SemaChecking.cpp revision 6edb029026d290f12393ed8389a3e1de596c77ec
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 Expr **Args, unsigned NumArgs) { 590 VariadicCallType CallType = 591 Method->isVariadic() ? VariadicMethod : VariadicDoesNotApply; 592 593 checkCall(Method, llvm::makeArrayRef<const Expr *>(Args, NumArgs), 594 Method->param_size(), 595 /*IsMemberFunction=*/false, 596 lbrac, Method->getSourceRange(), CallType); 597 598 return false; 599} 600 601bool Sema::CheckBlockCall(NamedDecl *NDecl, CallExpr *TheCall, 602 const FunctionProtoType *Proto) { 603 const VarDecl *V = dyn_cast<VarDecl>(NDecl); 604 if (!V) 605 return false; 606 607 QualType Ty = V->getType(); 608 if (!Ty->isBlockPointerType()) 609 return false; 610 611 VariadicCallType CallType = 612 Proto && Proto->isVariadic() ? VariadicBlock : VariadicDoesNotApply ; 613 unsigned NumProtoArgs = Proto ? Proto->getNumArgs() : 0; 614 615 checkCall(NDecl, 616 llvm::makeArrayRef<const Expr *>(TheCall->getArgs(), 617 TheCall->getNumArgs()), 618 NumProtoArgs, /*IsMemberFunction=*/false, 619 TheCall->getRParenLoc(), 620 TheCall->getCallee()->getSourceRange(), CallType); 621 622 return false; 623} 624 625ExprResult Sema::SemaAtomicOpsOverloaded(ExprResult TheCallResult, 626 AtomicExpr::AtomicOp Op) { 627 CallExpr *TheCall = cast<CallExpr>(TheCallResult.get()); 628 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 629 630 // All these operations take one of the following forms: 631 enum { 632 // C __c11_atomic_init(A *, C) 633 Init, 634 // C __c11_atomic_load(A *, int) 635 Load, 636 // void __atomic_load(A *, CP, int) 637 Copy, 638 // C __c11_atomic_add(A *, M, int) 639 Arithmetic, 640 // C __atomic_exchange_n(A *, CP, int) 641 Xchg, 642 // void __atomic_exchange(A *, C *, CP, int) 643 GNUXchg, 644 // bool __c11_atomic_compare_exchange_strong(A *, C *, CP, int, int) 645 C11CmpXchg, 646 // bool __atomic_compare_exchange(A *, C *, CP, bool, int, int) 647 GNUCmpXchg 648 } Form = Init; 649 const unsigned NumArgs[] = { 2, 2, 3, 3, 3, 4, 5, 6 }; 650 const unsigned NumVals[] = { 1, 0, 1, 1, 1, 2, 2, 3 }; 651 // where: 652 // C is an appropriate type, 653 // A is volatile _Atomic(C) for __c11 builtins and is C for GNU builtins, 654 // CP is C for __c11 builtins and GNU _n builtins and is C * otherwise, 655 // M is C if C is an integer, and ptrdiff_t if C is a pointer, and 656 // the int parameters are for orderings. 657 658 assert(AtomicExpr::AO__c11_atomic_init == 0 && 659 AtomicExpr::AO__c11_atomic_fetch_xor + 1 == AtomicExpr::AO__atomic_load 660 && "need to update code for modified C11 atomics"); 661 bool IsC11 = Op >= AtomicExpr::AO__c11_atomic_init && 662 Op <= AtomicExpr::AO__c11_atomic_fetch_xor; 663 bool IsN = Op == AtomicExpr::AO__atomic_load_n || 664 Op == AtomicExpr::AO__atomic_store_n || 665 Op == AtomicExpr::AO__atomic_exchange_n || 666 Op == AtomicExpr::AO__atomic_compare_exchange_n; 667 bool IsAddSub = false; 668 669 switch (Op) { 670 case AtomicExpr::AO__c11_atomic_init: 671 Form = Init; 672 break; 673 674 case AtomicExpr::AO__c11_atomic_load: 675 case AtomicExpr::AO__atomic_load_n: 676 Form = Load; 677 break; 678 679 case AtomicExpr::AO__c11_atomic_store: 680 case AtomicExpr::AO__atomic_load: 681 case AtomicExpr::AO__atomic_store: 682 case AtomicExpr::AO__atomic_store_n: 683 Form = Copy; 684 break; 685 686 case AtomicExpr::AO__c11_atomic_fetch_add: 687 case AtomicExpr::AO__c11_atomic_fetch_sub: 688 case AtomicExpr::AO__atomic_fetch_add: 689 case AtomicExpr::AO__atomic_fetch_sub: 690 case AtomicExpr::AO__atomic_add_fetch: 691 case AtomicExpr::AO__atomic_sub_fetch: 692 IsAddSub = true; 693 // Fall through. 694 case AtomicExpr::AO__c11_atomic_fetch_and: 695 case AtomicExpr::AO__c11_atomic_fetch_or: 696 case AtomicExpr::AO__c11_atomic_fetch_xor: 697 case AtomicExpr::AO__atomic_fetch_and: 698 case AtomicExpr::AO__atomic_fetch_or: 699 case AtomicExpr::AO__atomic_fetch_xor: 700 case AtomicExpr::AO__atomic_fetch_nand: 701 case AtomicExpr::AO__atomic_and_fetch: 702 case AtomicExpr::AO__atomic_or_fetch: 703 case AtomicExpr::AO__atomic_xor_fetch: 704 case AtomicExpr::AO__atomic_nand_fetch: 705 Form = Arithmetic; 706 break; 707 708 case AtomicExpr::AO__c11_atomic_exchange: 709 case AtomicExpr::AO__atomic_exchange_n: 710 Form = Xchg; 711 break; 712 713 case AtomicExpr::AO__atomic_exchange: 714 Form = GNUXchg; 715 break; 716 717 case AtomicExpr::AO__c11_atomic_compare_exchange_strong: 718 case AtomicExpr::AO__c11_atomic_compare_exchange_weak: 719 Form = C11CmpXchg; 720 break; 721 722 case AtomicExpr::AO__atomic_compare_exchange: 723 case AtomicExpr::AO__atomic_compare_exchange_n: 724 Form = GNUCmpXchg; 725 break; 726 } 727 728 // Check we have the right number of arguments. 729 if (TheCall->getNumArgs() < NumArgs[Form]) { 730 Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 731 << 0 << NumArgs[Form] << TheCall->getNumArgs() 732 << TheCall->getCallee()->getSourceRange(); 733 return ExprError(); 734 } else if (TheCall->getNumArgs() > NumArgs[Form]) { 735 Diag(TheCall->getArg(NumArgs[Form])->getLocStart(), 736 diag::err_typecheck_call_too_many_args) 737 << 0 << NumArgs[Form] << TheCall->getNumArgs() 738 << TheCall->getCallee()->getSourceRange(); 739 return ExprError(); 740 } 741 742 // Inspect the first argument of the atomic operation. 743 Expr *Ptr = TheCall->getArg(0); 744 Ptr = DefaultFunctionArrayLvalueConversion(Ptr).get(); 745 const PointerType *pointerType = Ptr->getType()->getAs<PointerType>(); 746 if (!pointerType) { 747 Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer) 748 << Ptr->getType() << Ptr->getSourceRange(); 749 return ExprError(); 750 } 751 752 // For a __c11 builtin, this should be a pointer to an _Atomic type. 753 QualType AtomTy = pointerType->getPointeeType(); // 'A' 754 QualType ValType = AtomTy; // 'C' 755 if (IsC11) { 756 if (!AtomTy->isAtomicType()) { 757 Diag(DRE->getLocStart(), diag::err_atomic_op_needs_atomic) 758 << Ptr->getType() << Ptr->getSourceRange(); 759 return ExprError(); 760 } 761 if (AtomTy.isConstQualified()) { 762 Diag(DRE->getLocStart(), diag::err_atomic_op_needs_non_const_atomic) 763 << Ptr->getType() << Ptr->getSourceRange(); 764 return ExprError(); 765 } 766 ValType = AtomTy->getAs<AtomicType>()->getValueType(); 767 } 768 769 // For an arithmetic operation, the implied arithmetic must be well-formed. 770 if (Form == Arithmetic) { 771 // gcc does not enforce these rules for GNU atomics, but we do so for sanity. 772 if (IsAddSub && !ValType->isIntegerType() && !ValType->isPointerType()) { 773 Diag(DRE->getLocStart(), diag::err_atomic_op_needs_atomic_int_or_ptr) 774 << IsC11 << Ptr->getType() << Ptr->getSourceRange(); 775 return ExprError(); 776 } 777 if (!IsAddSub && !ValType->isIntegerType()) { 778 Diag(DRE->getLocStart(), diag::err_atomic_op_bitwise_needs_atomic_int) 779 << IsC11 << Ptr->getType() << Ptr->getSourceRange(); 780 return ExprError(); 781 } 782 } else if (IsN && !ValType->isIntegerType() && !ValType->isPointerType()) { 783 // For __atomic_*_n operations, the value type must be a scalar integral or 784 // pointer type which is 1, 2, 4, 8 or 16 bytes in length. 785 Diag(DRE->getLocStart(), diag::err_atomic_op_needs_atomic_int_or_ptr) 786 << IsC11 << Ptr->getType() << Ptr->getSourceRange(); 787 return ExprError(); 788 } 789 790 if (!IsC11 && !AtomTy.isTriviallyCopyableType(Context)) { 791 // For GNU atomics, require a trivially-copyable type. This is not part of 792 // the GNU atomics specification, but we enforce it for sanity. 793 Diag(DRE->getLocStart(), diag::err_atomic_op_needs_trivial_copy) 794 << Ptr->getType() << Ptr->getSourceRange(); 795 return ExprError(); 796 } 797 798 // FIXME: For any builtin other than a load, the ValType must not be 799 // const-qualified. 800 801 switch (ValType.getObjCLifetime()) { 802 case Qualifiers::OCL_None: 803 case Qualifiers::OCL_ExplicitNone: 804 // okay 805 break; 806 807 case Qualifiers::OCL_Weak: 808 case Qualifiers::OCL_Strong: 809 case Qualifiers::OCL_Autoreleasing: 810 // FIXME: Can this happen? By this point, ValType should be known 811 // to be trivially copyable. 812 Diag(DRE->getLocStart(), diag::err_arc_atomic_ownership) 813 << ValType << Ptr->getSourceRange(); 814 return ExprError(); 815 } 816 817 QualType ResultType = ValType; 818 if (Form == Copy || Form == GNUXchg || Form == Init) 819 ResultType = Context.VoidTy; 820 else if (Form == C11CmpXchg || Form == GNUCmpXchg) 821 ResultType = Context.BoolTy; 822 823 // The type of a parameter passed 'by value'. In the GNU atomics, such 824 // arguments are actually passed as pointers. 825 QualType ByValType = ValType; // 'CP' 826 if (!IsC11 && !IsN) 827 ByValType = Ptr->getType(); 828 829 // The first argument --- the pointer --- has a fixed type; we 830 // deduce the types of the rest of the arguments accordingly. Walk 831 // the remaining arguments, converting them to the deduced value type. 832 for (unsigned i = 1; i != NumArgs[Form]; ++i) { 833 QualType Ty; 834 if (i < NumVals[Form] + 1) { 835 switch (i) { 836 case 1: 837 // The second argument is the non-atomic operand. For arithmetic, this 838 // is always passed by value, and for a compare_exchange it is always 839 // passed by address. For the rest, GNU uses by-address and C11 uses 840 // by-value. 841 assert(Form != Load); 842 if (Form == Init || (Form == Arithmetic && ValType->isIntegerType())) 843 Ty = ValType; 844 else if (Form == Copy || Form == Xchg) 845 Ty = ByValType; 846 else if (Form == Arithmetic) 847 Ty = Context.getPointerDiffType(); 848 else 849 Ty = Context.getPointerType(ValType.getUnqualifiedType()); 850 break; 851 case 2: 852 // The third argument to compare_exchange / GNU exchange is a 853 // (pointer to a) desired value. 854 Ty = ByValType; 855 break; 856 case 3: 857 // The fourth argument to GNU compare_exchange is a 'weak' flag. 858 Ty = Context.BoolTy; 859 break; 860 } 861 } else { 862 // The order(s) are always converted to int. 863 Ty = Context.IntTy; 864 } 865 866 InitializedEntity Entity = 867 InitializedEntity::InitializeParameter(Context, Ty, false); 868 ExprResult Arg = TheCall->getArg(i); 869 Arg = PerformCopyInitialization(Entity, SourceLocation(), Arg); 870 if (Arg.isInvalid()) 871 return true; 872 TheCall->setArg(i, Arg.get()); 873 } 874 875 // Permute the arguments into a 'consistent' order. 876 SmallVector<Expr*, 5> SubExprs; 877 SubExprs.push_back(Ptr); 878 switch (Form) { 879 case Init: 880 // Note, AtomicExpr::getVal1() has a special case for this atomic. 881 SubExprs.push_back(TheCall->getArg(1)); // Val1 882 break; 883 case Load: 884 SubExprs.push_back(TheCall->getArg(1)); // Order 885 break; 886 case Copy: 887 case Arithmetic: 888 case Xchg: 889 SubExprs.push_back(TheCall->getArg(2)); // Order 890 SubExprs.push_back(TheCall->getArg(1)); // Val1 891 break; 892 case GNUXchg: 893 // Note, AtomicExpr::getVal2() has a special case for this atomic. 894 SubExprs.push_back(TheCall->getArg(3)); // Order 895 SubExprs.push_back(TheCall->getArg(1)); // Val1 896 SubExprs.push_back(TheCall->getArg(2)); // Val2 897 break; 898 case C11CmpXchg: 899 SubExprs.push_back(TheCall->getArg(3)); // Order 900 SubExprs.push_back(TheCall->getArg(1)); // Val1 901 SubExprs.push_back(TheCall->getArg(4)); // OrderFail 902 SubExprs.push_back(TheCall->getArg(2)); // Val2 903 break; 904 case GNUCmpXchg: 905 SubExprs.push_back(TheCall->getArg(4)); // Order 906 SubExprs.push_back(TheCall->getArg(1)); // Val1 907 SubExprs.push_back(TheCall->getArg(5)); // OrderFail 908 SubExprs.push_back(TheCall->getArg(2)); // Val2 909 SubExprs.push_back(TheCall->getArg(3)); // Weak 910 break; 911 } 912 913 return Owned(new (Context) AtomicExpr(TheCall->getCallee()->getLocStart(), 914 SubExprs, ResultType, Op, 915 TheCall->getRParenLoc())); 916} 917 918 919/// checkBuiltinArgument - Given a call to a builtin function, perform 920/// normal type-checking on the given argument, updating the call in 921/// place. This is useful when a builtin function requires custom 922/// type-checking for some of its arguments but not necessarily all of 923/// them. 924/// 925/// Returns true on error. 926static bool checkBuiltinArgument(Sema &S, CallExpr *E, unsigned ArgIndex) { 927 FunctionDecl *Fn = E->getDirectCallee(); 928 assert(Fn && "builtin call without direct callee!"); 929 930 ParmVarDecl *Param = Fn->getParamDecl(ArgIndex); 931 InitializedEntity Entity = 932 InitializedEntity::InitializeParameter(S.Context, Param); 933 934 ExprResult Arg = E->getArg(0); 935 Arg = S.PerformCopyInitialization(Entity, SourceLocation(), Arg); 936 if (Arg.isInvalid()) 937 return true; 938 939 E->setArg(ArgIndex, Arg.take()); 940 return false; 941} 942 943/// SemaBuiltinAtomicOverloaded - We have a call to a function like 944/// __sync_fetch_and_add, which is an overloaded function based on the pointer 945/// type of its first argument. The main ActOnCallExpr routines have already 946/// promoted the types of arguments because all of these calls are prototyped as 947/// void(...). 948/// 949/// This function goes through and does final semantic checking for these 950/// builtins, 951ExprResult 952Sema::SemaBuiltinAtomicOverloaded(ExprResult TheCallResult) { 953 CallExpr *TheCall = (CallExpr *)TheCallResult.get(); 954 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 955 FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl()); 956 957 // Ensure that we have at least one argument to do type inference from. 958 if (TheCall->getNumArgs() < 1) { 959 Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args_at_least) 960 << 0 << 1 << TheCall->getNumArgs() 961 << TheCall->getCallee()->getSourceRange(); 962 return ExprError(); 963 } 964 965 // Inspect the first argument of the atomic builtin. This should always be 966 // a pointer type, whose element is an integral scalar or pointer type. 967 // Because it is a pointer type, we don't have to worry about any implicit 968 // casts here. 969 // FIXME: We don't allow floating point scalars as input. 970 Expr *FirstArg = TheCall->getArg(0); 971 ExprResult FirstArgResult = DefaultFunctionArrayLvalueConversion(FirstArg); 972 if (FirstArgResult.isInvalid()) 973 return ExprError(); 974 FirstArg = FirstArgResult.take(); 975 TheCall->setArg(0, FirstArg); 976 977 const PointerType *pointerType = FirstArg->getType()->getAs<PointerType>(); 978 if (!pointerType) { 979 Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer) 980 << FirstArg->getType() << FirstArg->getSourceRange(); 981 return ExprError(); 982 } 983 984 QualType ValType = pointerType->getPointeeType(); 985 if (!ValType->isIntegerType() && !ValType->isAnyPointerType() && 986 !ValType->isBlockPointerType()) { 987 Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer_intptr) 988 << FirstArg->getType() << FirstArg->getSourceRange(); 989 return ExprError(); 990 } 991 992 switch (ValType.getObjCLifetime()) { 993 case Qualifiers::OCL_None: 994 case Qualifiers::OCL_ExplicitNone: 995 // okay 996 break; 997 998 case Qualifiers::OCL_Weak: 999 case Qualifiers::OCL_Strong: 1000 case Qualifiers::OCL_Autoreleasing: 1001 Diag(DRE->getLocStart(), diag::err_arc_atomic_ownership) 1002 << ValType << FirstArg->getSourceRange(); 1003 return ExprError(); 1004 } 1005 1006 // Strip any qualifiers off ValType. 1007 ValType = ValType.getUnqualifiedType(); 1008 1009 // The majority of builtins return a value, but a few have special return 1010 // types, so allow them to override appropriately below. 1011 QualType ResultType = ValType; 1012 1013 // We need to figure out which concrete builtin this maps onto. For example, 1014 // __sync_fetch_and_add with a 2 byte object turns into 1015 // __sync_fetch_and_add_2. 1016#define BUILTIN_ROW(x) \ 1017 { Builtin::BI##x##_1, Builtin::BI##x##_2, Builtin::BI##x##_4, \ 1018 Builtin::BI##x##_8, Builtin::BI##x##_16 } 1019 1020 static const unsigned BuiltinIndices[][5] = { 1021 BUILTIN_ROW(__sync_fetch_and_add), 1022 BUILTIN_ROW(__sync_fetch_and_sub), 1023 BUILTIN_ROW(__sync_fetch_and_or), 1024 BUILTIN_ROW(__sync_fetch_and_and), 1025 BUILTIN_ROW(__sync_fetch_and_xor), 1026 1027 BUILTIN_ROW(__sync_add_and_fetch), 1028 BUILTIN_ROW(__sync_sub_and_fetch), 1029 BUILTIN_ROW(__sync_and_and_fetch), 1030 BUILTIN_ROW(__sync_or_and_fetch), 1031 BUILTIN_ROW(__sync_xor_and_fetch), 1032 1033 BUILTIN_ROW(__sync_val_compare_and_swap), 1034 BUILTIN_ROW(__sync_bool_compare_and_swap), 1035 BUILTIN_ROW(__sync_lock_test_and_set), 1036 BUILTIN_ROW(__sync_lock_release), 1037 BUILTIN_ROW(__sync_swap) 1038 }; 1039#undef BUILTIN_ROW 1040 1041 // Determine the index of the size. 1042 unsigned SizeIndex; 1043 switch (Context.getTypeSizeInChars(ValType).getQuantity()) { 1044 case 1: SizeIndex = 0; break; 1045 case 2: SizeIndex = 1; break; 1046 case 4: SizeIndex = 2; break; 1047 case 8: SizeIndex = 3; break; 1048 case 16: SizeIndex = 4; break; 1049 default: 1050 Diag(DRE->getLocStart(), diag::err_atomic_builtin_pointer_size) 1051 << FirstArg->getType() << FirstArg->getSourceRange(); 1052 return ExprError(); 1053 } 1054 1055 // Each of these builtins has one pointer argument, followed by some number of 1056 // values (0, 1 or 2) followed by a potentially empty varags list of stuff 1057 // that we ignore. Find out which row of BuiltinIndices to read from as well 1058 // as the number of fixed args. 1059 unsigned BuiltinID = FDecl->getBuiltinID(); 1060 unsigned BuiltinIndex, NumFixed = 1; 1061 switch (BuiltinID) { 1062 default: llvm_unreachable("Unknown overloaded atomic builtin!"); 1063 case Builtin::BI__sync_fetch_and_add: 1064 case Builtin::BI__sync_fetch_and_add_1: 1065 case Builtin::BI__sync_fetch_and_add_2: 1066 case Builtin::BI__sync_fetch_and_add_4: 1067 case Builtin::BI__sync_fetch_and_add_8: 1068 case Builtin::BI__sync_fetch_and_add_16: 1069 BuiltinIndex = 0; 1070 break; 1071 1072 case Builtin::BI__sync_fetch_and_sub: 1073 case Builtin::BI__sync_fetch_and_sub_1: 1074 case Builtin::BI__sync_fetch_and_sub_2: 1075 case Builtin::BI__sync_fetch_and_sub_4: 1076 case Builtin::BI__sync_fetch_and_sub_8: 1077 case Builtin::BI__sync_fetch_and_sub_16: 1078 BuiltinIndex = 1; 1079 break; 1080 1081 case Builtin::BI__sync_fetch_and_or: 1082 case Builtin::BI__sync_fetch_and_or_1: 1083 case Builtin::BI__sync_fetch_and_or_2: 1084 case Builtin::BI__sync_fetch_and_or_4: 1085 case Builtin::BI__sync_fetch_and_or_8: 1086 case Builtin::BI__sync_fetch_and_or_16: 1087 BuiltinIndex = 2; 1088 break; 1089 1090 case Builtin::BI__sync_fetch_and_and: 1091 case Builtin::BI__sync_fetch_and_and_1: 1092 case Builtin::BI__sync_fetch_and_and_2: 1093 case Builtin::BI__sync_fetch_and_and_4: 1094 case Builtin::BI__sync_fetch_and_and_8: 1095 case Builtin::BI__sync_fetch_and_and_16: 1096 BuiltinIndex = 3; 1097 break; 1098 1099 case Builtin::BI__sync_fetch_and_xor: 1100 case Builtin::BI__sync_fetch_and_xor_1: 1101 case Builtin::BI__sync_fetch_and_xor_2: 1102 case Builtin::BI__sync_fetch_and_xor_4: 1103 case Builtin::BI__sync_fetch_and_xor_8: 1104 case Builtin::BI__sync_fetch_and_xor_16: 1105 BuiltinIndex = 4; 1106 break; 1107 1108 case Builtin::BI__sync_add_and_fetch: 1109 case Builtin::BI__sync_add_and_fetch_1: 1110 case Builtin::BI__sync_add_and_fetch_2: 1111 case Builtin::BI__sync_add_and_fetch_4: 1112 case Builtin::BI__sync_add_and_fetch_8: 1113 case Builtin::BI__sync_add_and_fetch_16: 1114 BuiltinIndex = 5; 1115 break; 1116 1117 case Builtin::BI__sync_sub_and_fetch: 1118 case Builtin::BI__sync_sub_and_fetch_1: 1119 case Builtin::BI__sync_sub_and_fetch_2: 1120 case Builtin::BI__sync_sub_and_fetch_4: 1121 case Builtin::BI__sync_sub_and_fetch_8: 1122 case Builtin::BI__sync_sub_and_fetch_16: 1123 BuiltinIndex = 6; 1124 break; 1125 1126 case Builtin::BI__sync_and_and_fetch: 1127 case Builtin::BI__sync_and_and_fetch_1: 1128 case Builtin::BI__sync_and_and_fetch_2: 1129 case Builtin::BI__sync_and_and_fetch_4: 1130 case Builtin::BI__sync_and_and_fetch_8: 1131 case Builtin::BI__sync_and_and_fetch_16: 1132 BuiltinIndex = 7; 1133 break; 1134 1135 case Builtin::BI__sync_or_and_fetch: 1136 case Builtin::BI__sync_or_and_fetch_1: 1137 case Builtin::BI__sync_or_and_fetch_2: 1138 case Builtin::BI__sync_or_and_fetch_4: 1139 case Builtin::BI__sync_or_and_fetch_8: 1140 case Builtin::BI__sync_or_and_fetch_16: 1141 BuiltinIndex = 8; 1142 break; 1143 1144 case Builtin::BI__sync_xor_and_fetch: 1145 case Builtin::BI__sync_xor_and_fetch_1: 1146 case Builtin::BI__sync_xor_and_fetch_2: 1147 case Builtin::BI__sync_xor_and_fetch_4: 1148 case Builtin::BI__sync_xor_and_fetch_8: 1149 case Builtin::BI__sync_xor_and_fetch_16: 1150 BuiltinIndex = 9; 1151 break; 1152 1153 case Builtin::BI__sync_val_compare_and_swap: 1154 case Builtin::BI__sync_val_compare_and_swap_1: 1155 case Builtin::BI__sync_val_compare_and_swap_2: 1156 case Builtin::BI__sync_val_compare_and_swap_4: 1157 case Builtin::BI__sync_val_compare_and_swap_8: 1158 case Builtin::BI__sync_val_compare_and_swap_16: 1159 BuiltinIndex = 10; 1160 NumFixed = 2; 1161 break; 1162 1163 case Builtin::BI__sync_bool_compare_and_swap: 1164 case Builtin::BI__sync_bool_compare_and_swap_1: 1165 case Builtin::BI__sync_bool_compare_and_swap_2: 1166 case Builtin::BI__sync_bool_compare_and_swap_4: 1167 case Builtin::BI__sync_bool_compare_and_swap_8: 1168 case Builtin::BI__sync_bool_compare_and_swap_16: 1169 BuiltinIndex = 11; 1170 NumFixed = 2; 1171 ResultType = Context.BoolTy; 1172 break; 1173 1174 case Builtin::BI__sync_lock_test_and_set: 1175 case Builtin::BI__sync_lock_test_and_set_1: 1176 case Builtin::BI__sync_lock_test_and_set_2: 1177 case Builtin::BI__sync_lock_test_and_set_4: 1178 case Builtin::BI__sync_lock_test_and_set_8: 1179 case Builtin::BI__sync_lock_test_and_set_16: 1180 BuiltinIndex = 12; 1181 break; 1182 1183 case Builtin::BI__sync_lock_release: 1184 case Builtin::BI__sync_lock_release_1: 1185 case Builtin::BI__sync_lock_release_2: 1186 case Builtin::BI__sync_lock_release_4: 1187 case Builtin::BI__sync_lock_release_8: 1188 case Builtin::BI__sync_lock_release_16: 1189 BuiltinIndex = 13; 1190 NumFixed = 0; 1191 ResultType = Context.VoidTy; 1192 break; 1193 1194 case Builtin::BI__sync_swap: 1195 case Builtin::BI__sync_swap_1: 1196 case Builtin::BI__sync_swap_2: 1197 case Builtin::BI__sync_swap_4: 1198 case Builtin::BI__sync_swap_8: 1199 case Builtin::BI__sync_swap_16: 1200 BuiltinIndex = 14; 1201 break; 1202 } 1203 1204 // Now that we know how many fixed arguments we expect, first check that we 1205 // have at least that many. 1206 if (TheCall->getNumArgs() < 1+NumFixed) { 1207 Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args_at_least) 1208 << 0 << 1+NumFixed << TheCall->getNumArgs() 1209 << TheCall->getCallee()->getSourceRange(); 1210 return ExprError(); 1211 } 1212 1213 // Get the decl for the concrete builtin from this, we can tell what the 1214 // concrete integer type we should convert to is. 1215 unsigned NewBuiltinID = BuiltinIndices[BuiltinIndex][SizeIndex]; 1216 const char *NewBuiltinName = Context.BuiltinInfo.GetName(NewBuiltinID); 1217 FunctionDecl *NewBuiltinDecl; 1218 if (NewBuiltinID == BuiltinID) 1219 NewBuiltinDecl = FDecl; 1220 else { 1221 // Perform builtin lookup to avoid redeclaring it. 1222 DeclarationName DN(&Context.Idents.get(NewBuiltinName)); 1223 LookupResult Res(*this, DN, DRE->getLocStart(), LookupOrdinaryName); 1224 LookupName(Res, TUScope, /*AllowBuiltinCreation=*/true); 1225 assert(Res.getFoundDecl()); 1226 NewBuiltinDecl = dyn_cast<FunctionDecl>(Res.getFoundDecl()); 1227 if (NewBuiltinDecl == 0) 1228 return ExprError(); 1229 } 1230 1231 // The first argument --- the pointer --- has a fixed type; we 1232 // deduce the types of the rest of the arguments accordingly. Walk 1233 // the remaining arguments, converting them to the deduced value type. 1234 for (unsigned i = 0; i != NumFixed; ++i) { 1235 ExprResult Arg = TheCall->getArg(i+1); 1236 1237 // GCC does an implicit conversion to the pointer or integer ValType. This 1238 // can fail in some cases (1i -> int**), check for this error case now. 1239 // Initialize the argument. 1240 InitializedEntity Entity = InitializedEntity::InitializeParameter(Context, 1241 ValType, /*consume*/ false); 1242 Arg = PerformCopyInitialization(Entity, SourceLocation(), Arg); 1243 if (Arg.isInvalid()) 1244 return ExprError(); 1245 1246 // Okay, we have something that *can* be converted to the right type. Check 1247 // to see if there is a potentially weird extension going on here. This can 1248 // happen when you do an atomic operation on something like an char* and 1249 // pass in 42. The 42 gets converted to char. This is even more strange 1250 // for things like 45.123 -> char, etc. 1251 // FIXME: Do this check. 1252 TheCall->setArg(i+1, Arg.take()); 1253 } 1254 1255 ASTContext& Context = this->getASTContext(); 1256 1257 // Create a new DeclRefExpr to refer to the new decl. 1258 DeclRefExpr* NewDRE = DeclRefExpr::Create( 1259 Context, 1260 DRE->getQualifierLoc(), 1261 SourceLocation(), 1262 NewBuiltinDecl, 1263 /*enclosing*/ false, 1264 DRE->getLocation(), 1265 Context.BuiltinFnTy, 1266 DRE->getValueKind()); 1267 1268 // Set the callee in the CallExpr. 1269 // FIXME: This loses syntactic information. 1270 QualType CalleePtrTy = Context.getPointerType(NewBuiltinDecl->getType()); 1271 ExprResult PromotedCall = ImpCastExprToType(NewDRE, CalleePtrTy, 1272 CK_BuiltinFnToFnPtr); 1273 TheCall->setCallee(PromotedCall.take()); 1274 1275 // Change the result type of the call to match the original value type. This 1276 // is arbitrary, but the codegen for these builtins ins design to handle it 1277 // gracefully. 1278 TheCall->setType(ResultType); 1279 1280 return TheCallResult; 1281} 1282 1283/// CheckObjCString - Checks that the argument to the builtin 1284/// CFString constructor is correct 1285/// Note: It might also make sense to do the UTF-16 conversion here (would 1286/// simplify the backend). 1287bool Sema::CheckObjCString(Expr *Arg) { 1288 Arg = Arg->IgnoreParenCasts(); 1289 StringLiteral *Literal = dyn_cast<StringLiteral>(Arg); 1290 1291 if (!Literal || !Literal->isAscii()) { 1292 Diag(Arg->getLocStart(), diag::err_cfstring_literal_not_string_constant) 1293 << Arg->getSourceRange(); 1294 return true; 1295 } 1296 1297 if (Literal->containsNonAsciiOrNull()) { 1298 StringRef String = Literal->getString(); 1299 unsigned NumBytes = String.size(); 1300 SmallVector<UTF16, 128> ToBuf(NumBytes); 1301 const UTF8 *FromPtr = (const UTF8 *)String.data(); 1302 UTF16 *ToPtr = &ToBuf[0]; 1303 1304 ConversionResult Result = ConvertUTF8toUTF16(&FromPtr, FromPtr + NumBytes, 1305 &ToPtr, ToPtr + NumBytes, 1306 strictConversion); 1307 // Check for conversion failure. 1308 if (Result != conversionOK) 1309 Diag(Arg->getLocStart(), 1310 diag::warn_cfstring_truncated) << Arg->getSourceRange(); 1311 } 1312 return false; 1313} 1314 1315/// SemaBuiltinVAStart - Check the arguments to __builtin_va_start for validity. 1316/// Emit an error and return true on failure, return false on success. 1317bool Sema::SemaBuiltinVAStart(CallExpr *TheCall) { 1318 Expr *Fn = TheCall->getCallee(); 1319 if (TheCall->getNumArgs() > 2) { 1320 Diag(TheCall->getArg(2)->getLocStart(), 1321 diag::err_typecheck_call_too_many_args) 1322 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 1323 << Fn->getSourceRange() 1324 << SourceRange(TheCall->getArg(2)->getLocStart(), 1325 (*(TheCall->arg_end()-1))->getLocEnd()); 1326 return true; 1327 } 1328 1329 if (TheCall->getNumArgs() < 2) { 1330 return Diag(TheCall->getLocEnd(), 1331 diag::err_typecheck_call_too_few_args_at_least) 1332 << 0 /*function call*/ << 2 << TheCall->getNumArgs(); 1333 } 1334 1335 // Type-check the first argument normally. 1336 if (checkBuiltinArgument(*this, TheCall, 0)) 1337 return true; 1338 1339 // Determine whether the current function is variadic or not. 1340 BlockScopeInfo *CurBlock = getCurBlock(); 1341 bool isVariadic; 1342 if (CurBlock) 1343 isVariadic = CurBlock->TheDecl->isVariadic(); 1344 else if (FunctionDecl *FD = getCurFunctionDecl()) 1345 isVariadic = FD->isVariadic(); 1346 else 1347 isVariadic = getCurMethodDecl()->isVariadic(); 1348 1349 if (!isVariadic) { 1350 Diag(Fn->getLocStart(), diag::err_va_start_used_in_non_variadic_function); 1351 return true; 1352 } 1353 1354 // Verify that the second argument to the builtin is the last argument of the 1355 // current function or method. 1356 bool SecondArgIsLastNamedArgument = false; 1357 const Expr *Arg = TheCall->getArg(1)->IgnoreParenCasts(); 1358 1359 if (const DeclRefExpr *DR = dyn_cast<DeclRefExpr>(Arg)) { 1360 if (const ParmVarDecl *PV = dyn_cast<ParmVarDecl>(DR->getDecl())) { 1361 // FIXME: This isn't correct for methods (results in bogus warning). 1362 // Get the last formal in the current function. 1363 const ParmVarDecl *LastArg; 1364 if (CurBlock) 1365 LastArg = *(CurBlock->TheDecl->param_end()-1); 1366 else if (FunctionDecl *FD = getCurFunctionDecl()) 1367 LastArg = *(FD->param_end()-1); 1368 else 1369 LastArg = *(getCurMethodDecl()->param_end()-1); 1370 SecondArgIsLastNamedArgument = PV == LastArg; 1371 } 1372 } 1373 1374 if (!SecondArgIsLastNamedArgument) 1375 Diag(TheCall->getArg(1)->getLocStart(), 1376 diag::warn_second_parameter_of_va_start_not_last_named_argument); 1377 return false; 1378} 1379 1380/// SemaBuiltinUnorderedCompare - Handle functions like __builtin_isgreater and 1381/// friends. This is declared to take (...), so we have to check everything. 1382bool Sema::SemaBuiltinUnorderedCompare(CallExpr *TheCall) { 1383 if (TheCall->getNumArgs() < 2) 1384 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 1385 << 0 << 2 << TheCall->getNumArgs()/*function call*/; 1386 if (TheCall->getNumArgs() > 2) 1387 return Diag(TheCall->getArg(2)->getLocStart(), 1388 diag::err_typecheck_call_too_many_args) 1389 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 1390 << SourceRange(TheCall->getArg(2)->getLocStart(), 1391 (*(TheCall->arg_end()-1))->getLocEnd()); 1392 1393 ExprResult OrigArg0 = TheCall->getArg(0); 1394 ExprResult OrigArg1 = TheCall->getArg(1); 1395 1396 // Do standard promotions between the two arguments, returning their common 1397 // type. 1398 QualType Res = UsualArithmeticConversions(OrigArg0, OrigArg1, false); 1399 if (OrigArg0.isInvalid() || OrigArg1.isInvalid()) 1400 return true; 1401 1402 // Make sure any conversions are pushed back into the call; this is 1403 // type safe since unordered compare builtins are declared as "_Bool 1404 // foo(...)". 1405 TheCall->setArg(0, OrigArg0.get()); 1406 TheCall->setArg(1, OrigArg1.get()); 1407 1408 if (OrigArg0.get()->isTypeDependent() || OrigArg1.get()->isTypeDependent()) 1409 return false; 1410 1411 // If the common type isn't a real floating type, then the arguments were 1412 // invalid for this operation. 1413 if (Res.isNull() || !Res->isRealFloatingType()) 1414 return Diag(OrigArg0.get()->getLocStart(), 1415 diag::err_typecheck_call_invalid_ordered_compare) 1416 << OrigArg0.get()->getType() << OrigArg1.get()->getType() 1417 << SourceRange(OrigArg0.get()->getLocStart(), OrigArg1.get()->getLocEnd()); 1418 1419 return false; 1420} 1421 1422/// SemaBuiltinSemaBuiltinFPClassification - Handle functions like 1423/// __builtin_isnan and friends. This is declared to take (...), so we have 1424/// to check everything. We expect the last argument to be a floating point 1425/// value. 1426bool Sema::SemaBuiltinFPClassification(CallExpr *TheCall, unsigned NumArgs) { 1427 if (TheCall->getNumArgs() < NumArgs) 1428 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 1429 << 0 << NumArgs << TheCall->getNumArgs()/*function call*/; 1430 if (TheCall->getNumArgs() > NumArgs) 1431 return Diag(TheCall->getArg(NumArgs)->getLocStart(), 1432 diag::err_typecheck_call_too_many_args) 1433 << 0 /*function call*/ << NumArgs << TheCall->getNumArgs() 1434 << SourceRange(TheCall->getArg(NumArgs)->getLocStart(), 1435 (*(TheCall->arg_end()-1))->getLocEnd()); 1436 1437 Expr *OrigArg = TheCall->getArg(NumArgs-1); 1438 1439 if (OrigArg->isTypeDependent()) 1440 return false; 1441 1442 // This operation requires a non-_Complex floating-point number. 1443 if (!OrigArg->getType()->isRealFloatingType()) 1444 return Diag(OrigArg->getLocStart(), 1445 diag::err_typecheck_call_invalid_unary_fp) 1446 << OrigArg->getType() << OrigArg->getSourceRange(); 1447 1448 // If this is an implicit conversion from float -> double, remove it. 1449 if (ImplicitCastExpr *Cast = dyn_cast<ImplicitCastExpr>(OrigArg)) { 1450 Expr *CastArg = Cast->getSubExpr(); 1451 if (CastArg->getType()->isSpecificBuiltinType(BuiltinType::Float)) { 1452 assert(Cast->getType()->isSpecificBuiltinType(BuiltinType::Double) && 1453 "promotion from float to double is the only expected cast here"); 1454 Cast->setSubExpr(0); 1455 TheCall->setArg(NumArgs-1, CastArg); 1456 } 1457 } 1458 1459 return false; 1460} 1461 1462/// SemaBuiltinShuffleVector - Handle __builtin_shufflevector. 1463// This is declared to take (...), so we have to check everything. 1464ExprResult Sema::SemaBuiltinShuffleVector(CallExpr *TheCall) { 1465 if (TheCall->getNumArgs() < 2) 1466 return ExprError(Diag(TheCall->getLocEnd(), 1467 diag::err_typecheck_call_too_few_args_at_least) 1468 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 1469 << TheCall->getSourceRange()); 1470 1471 // Determine which of the following types of shufflevector we're checking: 1472 // 1) unary, vector mask: (lhs, mask) 1473 // 2) binary, vector mask: (lhs, rhs, mask) 1474 // 3) binary, scalar mask: (lhs, rhs, index, ..., index) 1475 QualType resType = TheCall->getArg(0)->getType(); 1476 unsigned numElements = 0; 1477 1478 if (!TheCall->getArg(0)->isTypeDependent() && 1479 !TheCall->getArg(1)->isTypeDependent()) { 1480 QualType LHSType = TheCall->getArg(0)->getType(); 1481 QualType RHSType = TheCall->getArg(1)->getType(); 1482 1483 if (!LHSType->isVectorType() || !RHSType->isVectorType()) { 1484 Diag(TheCall->getLocStart(), diag::err_shufflevector_non_vector) 1485 << SourceRange(TheCall->getArg(0)->getLocStart(), 1486 TheCall->getArg(1)->getLocEnd()); 1487 return ExprError(); 1488 } 1489 1490 numElements = LHSType->getAs<VectorType>()->getNumElements(); 1491 unsigned numResElements = TheCall->getNumArgs() - 2; 1492 1493 // Check to see if we have a call with 2 vector arguments, the unary shuffle 1494 // with mask. If so, verify that RHS is an integer vector type with the 1495 // same number of elts as lhs. 1496 if (TheCall->getNumArgs() == 2) { 1497 if (!RHSType->hasIntegerRepresentation() || 1498 RHSType->getAs<VectorType>()->getNumElements() != numElements) 1499 Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector) 1500 << SourceRange(TheCall->getArg(1)->getLocStart(), 1501 TheCall->getArg(1)->getLocEnd()); 1502 numResElements = numElements; 1503 } 1504 else if (!Context.hasSameUnqualifiedType(LHSType, RHSType)) { 1505 Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector) 1506 << SourceRange(TheCall->getArg(0)->getLocStart(), 1507 TheCall->getArg(1)->getLocEnd()); 1508 return ExprError(); 1509 } else if (numElements != numResElements) { 1510 QualType eltType = LHSType->getAs<VectorType>()->getElementType(); 1511 resType = Context.getVectorType(eltType, numResElements, 1512 VectorType::GenericVector); 1513 } 1514 } 1515 1516 for (unsigned i = 2; i < TheCall->getNumArgs(); i++) { 1517 if (TheCall->getArg(i)->isTypeDependent() || 1518 TheCall->getArg(i)->isValueDependent()) 1519 continue; 1520 1521 llvm::APSInt Result(32); 1522 if (!TheCall->getArg(i)->isIntegerConstantExpr(Result, Context)) 1523 return ExprError(Diag(TheCall->getLocStart(), 1524 diag::err_shufflevector_nonconstant_argument) 1525 << TheCall->getArg(i)->getSourceRange()); 1526 1527 if (Result.getActiveBits() > 64 || Result.getZExtValue() >= numElements*2) 1528 return ExprError(Diag(TheCall->getLocStart(), 1529 diag::err_shufflevector_argument_too_large) 1530 << TheCall->getArg(i)->getSourceRange()); 1531 } 1532 1533 SmallVector<Expr*, 32> exprs; 1534 1535 for (unsigned i = 0, e = TheCall->getNumArgs(); i != e; i++) { 1536 exprs.push_back(TheCall->getArg(i)); 1537 TheCall->setArg(i, 0); 1538 } 1539 1540 return Owned(new (Context) ShuffleVectorExpr(Context, exprs, resType, 1541 TheCall->getCallee()->getLocStart(), 1542 TheCall->getRParenLoc())); 1543} 1544 1545/// SemaBuiltinPrefetch - Handle __builtin_prefetch. 1546// This is declared to take (const void*, ...) and can take two 1547// optional constant int args. 1548bool Sema::SemaBuiltinPrefetch(CallExpr *TheCall) { 1549 unsigned NumArgs = TheCall->getNumArgs(); 1550 1551 if (NumArgs > 3) 1552 return Diag(TheCall->getLocEnd(), 1553 diag::err_typecheck_call_too_many_args_at_most) 1554 << 0 /*function call*/ << 3 << NumArgs 1555 << TheCall->getSourceRange(); 1556 1557 // Argument 0 is checked for us and the remaining arguments must be 1558 // constant integers. 1559 for (unsigned i = 1; i != NumArgs; ++i) { 1560 Expr *Arg = TheCall->getArg(i); 1561 1562 // We can't check the value of a dependent argument. 1563 if (Arg->isTypeDependent() || Arg->isValueDependent()) 1564 continue; 1565 1566 llvm::APSInt Result; 1567 if (SemaBuiltinConstantArg(TheCall, i, Result)) 1568 return true; 1569 1570 // FIXME: gcc issues a warning and rewrites these to 0. These 1571 // seems especially odd for the third argument since the default 1572 // is 3. 1573 if (i == 1) { 1574 if (Result.getLimitedValue() > 1) 1575 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 1576 << "0" << "1" << Arg->getSourceRange(); 1577 } else { 1578 if (Result.getLimitedValue() > 3) 1579 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 1580 << "0" << "3" << Arg->getSourceRange(); 1581 } 1582 } 1583 1584 return false; 1585} 1586 1587/// SemaBuiltinConstantArg - Handle a check if argument ArgNum of CallExpr 1588/// TheCall is a constant expression. 1589bool Sema::SemaBuiltinConstantArg(CallExpr *TheCall, int ArgNum, 1590 llvm::APSInt &Result) { 1591 Expr *Arg = TheCall->getArg(ArgNum); 1592 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 1593 FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl()); 1594 1595 if (Arg->isTypeDependent() || Arg->isValueDependent()) return false; 1596 1597 if (!Arg->isIntegerConstantExpr(Result, Context)) 1598 return Diag(TheCall->getLocStart(), diag::err_constant_integer_arg_type) 1599 << FDecl->getDeclName() << Arg->getSourceRange(); 1600 1601 return false; 1602} 1603 1604/// SemaBuiltinObjectSize - Handle __builtin_object_size(void *ptr, 1605/// int type). This simply type checks that type is one of the defined 1606/// constants (0-3). 1607// For compatibility check 0-3, llvm only handles 0 and 2. 1608bool Sema::SemaBuiltinObjectSize(CallExpr *TheCall) { 1609 llvm::APSInt Result; 1610 1611 // We can't check the value of a dependent argument. 1612 if (TheCall->getArg(1)->isTypeDependent() || 1613 TheCall->getArg(1)->isValueDependent()) 1614 return false; 1615 1616 // Check constant-ness first. 1617 if (SemaBuiltinConstantArg(TheCall, 1, Result)) 1618 return true; 1619 1620 Expr *Arg = TheCall->getArg(1); 1621 if (Result.getSExtValue() < 0 || Result.getSExtValue() > 3) { 1622 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 1623 << "0" << "3" << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 1624 } 1625 1626 return false; 1627} 1628 1629/// SemaBuiltinLongjmp - Handle __builtin_longjmp(void *env[5], int val). 1630/// This checks that val is a constant 1. 1631bool Sema::SemaBuiltinLongjmp(CallExpr *TheCall) { 1632 Expr *Arg = TheCall->getArg(1); 1633 llvm::APSInt Result; 1634 1635 // TODO: This is less than ideal. Overload this to take a value. 1636 if (SemaBuiltinConstantArg(TheCall, 1, Result)) 1637 return true; 1638 1639 if (Result != 1) 1640 return Diag(TheCall->getLocStart(), diag::err_builtin_longjmp_invalid_val) 1641 << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 1642 1643 return false; 1644} 1645 1646// Determine if an expression is a string literal or constant string. 1647// If this function returns false on the arguments to a function expecting a 1648// format string, we will usually need to emit a warning. 1649// True string literals are then checked by CheckFormatString. 1650Sema::StringLiteralCheckType 1651Sema::checkFormatStringExpr(const Expr *E, ArrayRef<const Expr *> Args, 1652 bool HasVAListArg, 1653 unsigned format_idx, unsigned firstDataArg, 1654 FormatStringType Type, VariadicCallType CallType, 1655 bool inFunctionCall) { 1656 tryAgain: 1657 if (E->isTypeDependent() || E->isValueDependent()) 1658 return SLCT_NotALiteral; 1659 1660 E = E->IgnoreParenCasts(); 1661 1662 if (E->isNullPointerConstant(Context, Expr::NPC_ValueDependentIsNotNull)) 1663 // Technically -Wformat-nonliteral does not warn about this case. 1664 // The behavior of printf and friends in this case is implementation 1665 // dependent. Ideally if the format string cannot be null then 1666 // it should have a 'nonnull' attribute in the function prototype. 1667 return SLCT_CheckedLiteral; 1668 1669 switch (E->getStmtClass()) { 1670 case Stmt::BinaryConditionalOperatorClass: 1671 case Stmt::ConditionalOperatorClass: { 1672 // The expression is a literal if both sub-expressions were, and it was 1673 // completely checked only if both sub-expressions were checked. 1674 const AbstractConditionalOperator *C = 1675 cast<AbstractConditionalOperator>(E); 1676 StringLiteralCheckType Left = 1677 checkFormatStringExpr(C->getTrueExpr(), Args, 1678 HasVAListArg, format_idx, firstDataArg, 1679 Type, CallType, inFunctionCall); 1680 if (Left == SLCT_NotALiteral) 1681 return SLCT_NotALiteral; 1682 StringLiteralCheckType Right = 1683 checkFormatStringExpr(C->getFalseExpr(), Args, 1684 HasVAListArg, format_idx, firstDataArg, 1685 Type, CallType, inFunctionCall); 1686 return Left < Right ? Left : Right; 1687 } 1688 1689 case Stmt::ImplicitCastExprClass: { 1690 E = cast<ImplicitCastExpr>(E)->getSubExpr(); 1691 goto tryAgain; 1692 } 1693 1694 case Stmt::OpaqueValueExprClass: 1695 if (const Expr *src = cast<OpaqueValueExpr>(E)->getSourceExpr()) { 1696 E = src; 1697 goto tryAgain; 1698 } 1699 return SLCT_NotALiteral; 1700 1701 case Stmt::PredefinedExprClass: 1702 // While __func__, etc., are technically not string literals, they 1703 // cannot contain format specifiers and thus are not a security 1704 // liability. 1705 return SLCT_UncheckedLiteral; 1706 1707 case Stmt::DeclRefExprClass: { 1708 const DeclRefExpr *DR = cast<DeclRefExpr>(E); 1709 1710 // As an exception, do not flag errors for variables binding to 1711 // const string literals. 1712 if (const VarDecl *VD = dyn_cast<VarDecl>(DR->getDecl())) { 1713 bool isConstant = false; 1714 QualType T = DR->getType(); 1715 1716 if (const ArrayType *AT = Context.getAsArrayType(T)) { 1717 isConstant = AT->getElementType().isConstant(Context); 1718 } else if (const PointerType *PT = T->getAs<PointerType>()) { 1719 isConstant = T.isConstant(Context) && 1720 PT->getPointeeType().isConstant(Context); 1721 } else if (T->isObjCObjectPointerType()) { 1722 // In ObjC, there is usually no "const ObjectPointer" type, 1723 // so don't check if the pointee type is constant. 1724 isConstant = T.isConstant(Context); 1725 } 1726 1727 if (isConstant) { 1728 if (const Expr *Init = VD->getAnyInitializer()) { 1729 // Look through initializers like const char c[] = { "foo" } 1730 if (const InitListExpr *InitList = dyn_cast<InitListExpr>(Init)) { 1731 if (InitList->isStringLiteralInit()) 1732 Init = InitList->getInit(0)->IgnoreParenImpCasts(); 1733 } 1734 return checkFormatStringExpr(Init, Args, 1735 HasVAListArg, format_idx, 1736 firstDataArg, Type, CallType, 1737 /*inFunctionCall*/false); 1738 } 1739 } 1740 1741 // For vprintf* functions (i.e., HasVAListArg==true), we add a 1742 // special check to see if the format string is a function parameter 1743 // of the function calling the printf function. If the function 1744 // has an attribute indicating it is a printf-like function, then we 1745 // should suppress warnings concerning non-literals being used in a call 1746 // to a vprintf function. For example: 1747 // 1748 // void 1749 // logmessage(char const *fmt __attribute__ (format (printf, 1, 2)), ...){ 1750 // va_list ap; 1751 // va_start(ap, fmt); 1752 // vprintf(fmt, ap); // Do NOT emit a warning about "fmt". 1753 // ... 1754 // 1755 if (HasVAListArg) { 1756 if (const ParmVarDecl *PV = dyn_cast<ParmVarDecl>(VD)) { 1757 if (const NamedDecl *ND = dyn_cast<NamedDecl>(PV->getDeclContext())) { 1758 int PVIndex = PV->getFunctionScopeIndex() + 1; 1759 for (specific_attr_iterator<FormatAttr> 1760 i = ND->specific_attr_begin<FormatAttr>(), 1761 e = ND->specific_attr_end<FormatAttr>(); i != e ; ++i) { 1762 FormatAttr *PVFormat = *i; 1763 // adjust for implicit parameter 1764 if (const CXXMethodDecl *MD = dyn_cast<CXXMethodDecl>(ND)) 1765 if (MD->isInstance()) 1766 ++PVIndex; 1767 // We also check if the formats are compatible. 1768 // We can't pass a 'scanf' string to a 'printf' function. 1769 if (PVIndex == PVFormat->getFormatIdx() && 1770 Type == GetFormatStringType(PVFormat)) 1771 return SLCT_UncheckedLiteral; 1772 } 1773 } 1774 } 1775 } 1776 } 1777 1778 return SLCT_NotALiteral; 1779 } 1780 1781 case Stmt::CallExprClass: 1782 case Stmt::CXXMemberCallExprClass: { 1783 const CallExpr *CE = cast<CallExpr>(E); 1784 if (const NamedDecl *ND = dyn_cast_or_null<NamedDecl>(CE->getCalleeDecl())) { 1785 if (const FormatArgAttr *FA = ND->getAttr<FormatArgAttr>()) { 1786 unsigned ArgIndex = FA->getFormatIdx(); 1787 if (const CXXMethodDecl *MD = dyn_cast<CXXMethodDecl>(ND)) 1788 if (MD->isInstance()) 1789 --ArgIndex; 1790 const Expr *Arg = CE->getArg(ArgIndex - 1); 1791 1792 return checkFormatStringExpr(Arg, Args, 1793 HasVAListArg, format_idx, firstDataArg, 1794 Type, CallType, inFunctionCall); 1795 } else if (const FunctionDecl *FD = dyn_cast<FunctionDecl>(ND)) { 1796 unsigned BuiltinID = FD->getBuiltinID(); 1797 if (BuiltinID == Builtin::BI__builtin___CFStringMakeConstantString || 1798 BuiltinID == Builtin::BI__builtin___NSStringMakeConstantString) { 1799 const Expr *Arg = CE->getArg(0); 1800 return checkFormatStringExpr(Arg, Args, 1801 HasVAListArg, format_idx, 1802 firstDataArg, Type, CallType, 1803 inFunctionCall); 1804 } 1805 } 1806 } 1807 1808 return SLCT_NotALiteral; 1809 } 1810 case Stmt::ObjCStringLiteralClass: 1811 case Stmt::StringLiteralClass: { 1812 const StringLiteral *StrE = NULL; 1813 1814 if (const ObjCStringLiteral *ObjCFExpr = dyn_cast<ObjCStringLiteral>(E)) 1815 StrE = ObjCFExpr->getString(); 1816 else 1817 StrE = cast<StringLiteral>(E); 1818 1819 if (StrE) { 1820 CheckFormatString(StrE, E, Args, HasVAListArg, format_idx, 1821 firstDataArg, Type, inFunctionCall, CallType); 1822 return SLCT_CheckedLiteral; 1823 } 1824 1825 return SLCT_NotALiteral; 1826 } 1827 1828 default: 1829 return SLCT_NotALiteral; 1830 } 1831} 1832 1833void 1834Sema::CheckNonNullArguments(const NonNullAttr *NonNull, 1835 const Expr * const *ExprArgs, 1836 SourceLocation CallSiteLoc) { 1837 for (NonNullAttr::args_iterator i = NonNull->args_begin(), 1838 e = NonNull->args_end(); 1839 i != e; ++i) { 1840 const Expr *ArgExpr = ExprArgs[*i]; 1841 1842 // As a special case, transparent unions initialized with zero are 1843 // considered null for the purposes of the nonnull attribute. 1844 if (const RecordType *UT = ArgExpr->getType()->getAsUnionType()) { 1845 if (UT->getDecl()->hasAttr<TransparentUnionAttr>()) 1846 if (const CompoundLiteralExpr *CLE = 1847 dyn_cast<CompoundLiteralExpr>(ArgExpr)) 1848 if (const InitListExpr *ILE = 1849 dyn_cast<InitListExpr>(CLE->getInitializer())) 1850 ArgExpr = ILE->getInit(0); 1851 } 1852 1853 bool Result; 1854 if (ArgExpr->EvaluateAsBooleanCondition(Result, Context) && !Result) 1855 Diag(CallSiteLoc, diag::warn_null_arg) << ArgExpr->getSourceRange(); 1856 } 1857} 1858 1859Sema::FormatStringType Sema::GetFormatStringType(const FormatAttr *Format) { 1860 return llvm::StringSwitch<FormatStringType>(Format->getType()) 1861 .Case("scanf", FST_Scanf) 1862 .Cases("printf", "printf0", FST_Printf) 1863 .Cases("NSString", "CFString", FST_NSString) 1864 .Case("strftime", FST_Strftime) 1865 .Case("strfmon", FST_Strfmon) 1866 .Cases("kprintf", "cmn_err", "vcmn_err", "zcmn_err", FST_Kprintf) 1867 .Default(FST_Unknown); 1868} 1869 1870/// CheckFormatArguments - Check calls to printf and scanf (and similar 1871/// functions) for correct use of format strings. 1872/// Returns true if a format string has been fully checked. 1873bool Sema::CheckFormatArguments(const FormatAttr *Format, 1874 ArrayRef<const Expr *> Args, 1875 bool IsCXXMember, 1876 VariadicCallType CallType, 1877 SourceLocation Loc, SourceRange Range) { 1878 FormatStringInfo FSI; 1879 if (getFormatStringInfo(Format, IsCXXMember, &FSI)) 1880 return CheckFormatArguments(Args, FSI.HasVAListArg, FSI.FormatIdx, 1881 FSI.FirstDataArg, GetFormatStringType(Format), 1882 CallType, Loc, Range); 1883 return false; 1884} 1885 1886bool Sema::CheckFormatArguments(ArrayRef<const Expr *> Args, 1887 bool HasVAListArg, unsigned format_idx, 1888 unsigned firstDataArg, FormatStringType Type, 1889 VariadicCallType CallType, 1890 SourceLocation Loc, SourceRange Range) { 1891 // CHECK: printf/scanf-like function is called with no format string. 1892 if (format_idx >= Args.size()) { 1893 Diag(Loc, diag::warn_missing_format_string) << Range; 1894 return false; 1895 } 1896 1897 const Expr *OrigFormatExpr = Args[format_idx]->IgnoreParenCasts(); 1898 1899 // CHECK: format string is not a string literal. 1900 // 1901 // Dynamically generated format strings are difficult to 1902 // automatically vet at compile time. Requiring that format strings 1903 // are string literals: (1) permits the checking of format strings by 1904 // the compiler and thereby (2) can practically remove the source of 1905 // many format string exploits. 1906 1907 // Format string can be either ObjC string (e.g. @"%d") or 1908 // C string (e.g. "%d") 1909 // ObjC string uses the same format specifiers as C string, so we can use 1910 // the same format string checking logic for both ObjC and C strings. 1911 StringLiteralCheckType CT = 1912 checkFormatStringExpr(OrigFormatExpr, Args, HasVAListArg, 1913 format_idx, firstDataArg, Type, CallType); 1914 if (CT != SLCT_NotALiteral) 1915 // Literal format string found, check done! 1916 return CT == SLCT_CheckedLiteral; 1917 1918 // Strftime is particular as it always uses a single 'time' argument, 1919 // so it is safe to pass a non-literal string. 1920 if (Type == FST_Strftime) 1921 return false; 1922 1923 // Do not emit diag when the string param is a macro expansion and the 1924 // format is either NSString or CFString. This is a hack to prevent 1925 // diag when using the NSLocalizedString and CFCopyLocalizedString macros 1926 // which are usually used in place of NS and CF string literals. 1927 if (Type == FST_NSString && 1928 SourceMgr.isInSystemMacro(Args[format_idx]->getLocStart())) 1929 return false; 1930 1931 // If there are no arguments specified, warn with -Wformat-security, otherwise 1932 // warn only with -Wformat-nonliteral. 1933 if (Args.size() == format_idx+1) 1934 Diag(Args[format_idx]->getLocStart(), 1935 diag::warn_format_nonliteral_noargs) 1936 << OrigFormatExpr->getSourceRange(); 1937 else 1938 Diag(Args[format_idx]->getLocStart(), 1939 diag::warn_format_nonliteral) 1940 << OrigFormatExpr->getSourceRange(); 1941 return false; 1942} 1943 1944namespace { 1945class CheckFormatHandler : public analyze_format_string::FormatStringHandler { 1946protected: 1947 Sema &S; 1948 const StringLiteral *FExpr; 1949 const Expr *OrigFormatExpr; 1950 const unsigned FirstDataArg; 1951 const unsigned NumDataArgs; 1952 const char *Beg; // Start of format string. 1953 const bool HasVAListArg; 1954 ArrayRef<const Expr *> Args; 1955 unsigned FormatIdx; 1956 llvm::BitVector CoveredArgs; 1957 bool usesPositionalArgs; 1958 bool atFirstArg; 1959 bool inFunctionCall; 1960 Sema::VariadicCallType CallType; 1961public: 1962 CheckFormatHandler(Sema &s, const StringLiteral *fexpr, 1963 const Expr *origFormatExpr, unsigned firstDataArg, 1964 unsigned numDataArgs, const char *beg, bool hasVAListArg, 1965 ArrayRef<const Expr *> Args, 1966 unsigned formatIdx, bool inFunctionCall, 1967 Sema::VariadicCallType callType) 1968 : S(s), FExpr(fexpr), OrigFormatExpr(origFormatExpr), 1969 FirstDataArg(firstDataArg), NumDataArgs(numDataArgs), 1970 Beg(beg), HasVAListArg(hasVAListArg), 1971 Args(Args), FormatIdx(formatIdx), 1972 usesPositionalArgs(false), atFirstArg(true), 1973 inFunctionCall(inFunctionCall), CallType(callType) { 1974 CoveredArgs.resize(numDataArgs); 1975 CoveredArgs.reset(); 1976 } 1977 1978 void DoneProcessing(); 1979 1980 void HandleIncompleteSpecifier(const char *startSpecifier, 1981 unsigned specifierLen); 1982 1983 void HandleInvalidLengthModifier( 1984 const analyze_format_string::FormatSpecifier &FS, 1985 const analyze_format_string::ConversionSpecifier &CS, 1986 const char *startSpecifier, unsigned specifierLen, unsigned DiagID); 1987 1988 void HandleNonStandardLengthModifier( 1989 const analyze_format_string::FormatSpecifier &FS, 1990 const char *startSpecifier, unsigned specifierLen); 1991 1992 void HandleNonStandardConversionSpecifier( 1993 const analyze_format_string::ConversionSpecifier &CS, 1994 const char *startSpecifier, unsigned specifierLen); 1995 1996 virtual void HandlePosition(const char *startPos, unsigned posLen); 1997 1998 virtual void HandleInvalidPosition(const char *startSpecifier, 1999 unsigned specifierLen, 2000 analyze_format_string::PositionContext p); 2001 2002 virtual void HandleZeroPosition(const char *startPos, unsigned posLen); 2003 2004 void HandleNullChar(const char *nullCharacter); 2005 2006 template <typename Range> 2007 static void EmitFormatDiagnostic(Sema &S, bool inFunctionCall, 2008 const Expr *ArgumentExpr, 2009 PartialDiagnostic PDiag, 2010 SourceLocation StringLoc, 2011 bool IsStringLocation, Range StringRange, 2012 ArrayRef<FixItHint> Fixit = ArrayRef<FixItHint>()); 2013 2014protected: 2015 bool HandleInvalidConversionSpecifier(unsigned argIndex, SourceLocation Loc, 2016 const char *startSpec, 2017 unsigned specifierLen, 2018 const char *csStart, unsigned csLen); 2019 2020 void HandlePositionalNonpositionalArgs(SourceLocation Loc, 2021 const char *startSpec, 2022 unsigned specifierLen); 2023 2024 SourceRange getFormatStringRange(); 2025 CharSourceRange getSpecifierRange(const char *startSpecifier, 2026 unsigned specifierLen); 2027 SourceLocation getLocationOfByte(const char *x); 2028 2029 const Expr *getDataArg(unsigned i) const; 2030 2031 bool CheckNumArgs(const analyze_format_string::FormatSpecifier &FS, 2032 const analyze_format_string::ConversionSpecifier &CS, 2033 const char *startSpecifier, unsigned specifierLen, 2034 unsigned argIndex); 2035 2036 template <typename Range> 2037 void EmitFormatDiagnostic(PartialDiagnostic PDiag, SourceLocation StringLoc, 2038 bool IsStringLocation, Range StringRange, 2039 ArrayRef<FixItHint> Fixit = ArrayRef<FixItHint>()); 2040 2041 void CheckPositionalAndNonpositionalArgs( 2042 const analyze_format_string::FormatSpecifier *FS); 2043}; 2044} 2045 2046SourceRange CheckFormatHandler::getFormatStringRange() { 2047 return OrigFormatExpr->getSourceRange(); 2048} 2049 2050CharSourceRange CheckFormatHandler:: 2051getSpecifierRange(const char *startSpecifier, unsigned specifierLen) { 2052 SourceLocation Start = getLocationOfByte(startSpecifier); 2053 SourceLocation End = getLocationOfByte(startSpecifier + specifierLen - 1); 2054 2055 // Advance the end SourceLocation by one due to half-open ranges. 2056 End = End.getLocWithOffset(1); 2057 2058 return CharSourceRange::getCharRange(Start, End); 2059} 2060 2061SourceLocation CheckFormatHandler::getLocationOfByte(const char *x) { 2062 return S.getLocationOfStringLiteralByte(FExpr, x - Beg); 2063} 2064 2065void CheckFormatHandler::HandleIncompleteSpecifier(const char *startSpecifier, 2066 unsigned specifierLen){ 2067 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_incomplete_specifier), 2068 getLocationOfByte(startSpecifier), 2069 /*IsStringLocation*/true, 2070 getSpecifierRange(startSpecifier, specifierLen)); 2071} 2072 2073void CheckFormatHandler::HandleInvalidLengthModifier( 2074 const analyze_format_string::FormatSpecifier &FS, 2075 const analyze_format_string::ConversionSpecifier &CS, 2076 const char *startSpecifier, unsigned specifierLen, unsigned DiagID) { 2077 using namespace analyze_format_string; 2078 2079 const LengthModifier &LM = FS.getLengthModifier(); 2080 CharSourceRange LMRange = getSpecifierRange(LM.getStart(), LM.getLength()); 2081 2082 // See if we know how to fix this length modifier. 2083 Optional<LengthModifier> FixedLM = FS.getCorrectedLengthModifier(); 2084 if (FixedLM) { 2085 EmitFormatDiagnostic(S.PDiag(DiagID) << LM.toString() << CS.toString(), 2086 getLocationOfByte(LM.getStart()), 2087 /*IsStringLocation*/true, 2088 getSpecifierRange(startSpecifier, specifierLen)); 2089 2090 S.Diag(getLocationOfByte(LM.getStart()), diag::note_format_fix_specifier) 2091 << FixedLM->toString() 2092 << FixItHint::CreateReplacement(LMRange, FixedLM->toString()); 2093 2094 } else { 2095 FixItHint Hint; 2096 if (DiagID == diag::warn_format_nonsensical_length) 2097 Hint = FixItHint::CreateRemoval(LMRange); 2098 2099 EmitFormatDiagnostic(S.PDiag(DiagID) << LM.toString() << CS.toString(), 2100 getLocationOfByte(LM.getStart()), 2101 /*IsStringLocation*/true, 2102 getSpecifierRange(startSpecifier, specifierLen), 2103 Hint); 2104 } 2105} 2106 2107void CheckFormatHandler::HandleNonStandardLengthModifier( 2108 const analyze_format_string::FormatSpecifier &FS, 2109 const char *startSpecifier, unsigned specifierLen) { 2110 using namespace analyze_format_string; 2111 2112 const LengthModifier &LM = FS.getLengthModifier(); 2113 CharSourceRange LMRange = getSpecifierRange(LM.getStart(), LM.getLength()); 2114 2115 // See if we know how to fix this length modifier. 2116 Optional<LengthModifier> FixedLM = FS.getCorrectedLengthModifier(); 2117 if (FixedLM) { 2118 EmitFormatDiagnostic(S.PDiag(diag::warn_format_non_standard) 2119 << LM.toString() << 0, 2120 getLocationOfByte(LM.getStart()), 2121 /*IsStringLocation*/true, 2122 getSpecifierRange(startSpecifier, specifierLen)); 2123 2124 S.Diag(getLocationOfByte(LM.getStart()), diag::note_format_fix_specifier) 2125 << FixedLM->toString() 2126 << FixItHint::CreateReplacement(LMRange, FixedLM->toString()); 2127 2128 } else { 2129 EmitFormatDiagnostic(S.PDiag(diag::warn_format_non_standard) 2130 << LM.toString() << 0, 2131 getLocationOfByte(LM.getStart()), 2132 /*IsStringLocation*/true, 2133 getSpecifierRange(startSpecifier, specifierLen)); 2134 } 2135} 2136 2137void CheckFormatHandler::HandleNonStandardConversionSpecifier( 2138 const analyze_format_string::ConversionSpecifier &CS, 2139 const char *startSpecifier, unsigned specifierLen) { 2140 using namespace analyze_format_string; 2141 2142 // See if we know how to fix this conversion specifier. 2143 Optional<ConversionSpecifier> FixedCS = CS.getStandardSpecifier(); 2144 if (FixedCS) { 2145 EmitFormatDiagnostic(S.PDiag(diag::warn_format_non_standard) 2146 << CS.toString() << /*conversion specifier*/1, 2147 getLocationOfByte(CS.getStart()), 2148 /*IsStringLocation*/true, 2149 getSpecifierRange(startSpecifier, specifierLen)); 2150 2151 CharSourceRange CSRange = getSpecifierRange(CS.getStart(), CS.getLength()); 2152 S.Diag(getLocationOfByte(CS.getStart()), diag::note_format_fix_specifier) 2153 << FixedCS->toString() 2154 << FixItHint::CreateReplacement(CSRange, FixedCS->toString()); 2155 } else { 2156 EmitFormatDiagnostic(S.PDiag(diag::warn_format_non_standard) 2157 << CS.toString() << /*conversion specifier*/1, 2158 getLocationOfByte(CS.getStart()), 2159 /*IsStringLocation*/true, 2160 getSpecifierRange(startSpecifier, specifierLen)); 2161 } 2162} 2163 2164void CheckFormatHandler::HandlePosition(const char *startPos, 2165 unsigned posLen) { 2166 EmitFormatDiagnostic(S.PDiag(diag::warn_format_non_standard_positional_arg), 2167 getLocationOfByte(startPos), 2168 /*IsStringLocation*/true, 2169 getSpecifierRange(startPos, posLen)); 2170} 2171 2172void 2173CheckFormatHandler::HandleInvalidPosition(const char *startPos, unsigned posLen, 2174 analyze_format_string::PositionContext p) { 2175 EmitFormatDiagnostic(S.PDiag(diag::warn_format_invalid_positional_specifier) 2176 << (unsigned) p, 2177 getLocationOfByte(startPos), /*IsStringLocation*/true, 2178 getSpecifierRange(startPos, posLen)); 2179} 2180 2181void CheckFormatHandler::HandleZeroPosition(const char *startPos, 2182 unsigned posLen) { 2183 EmitFormatDiagnostic(S.PDiag(diag::warn_format_zero_positional_specifier), 2184 getLocationOfByte(startPos), 2185 /*IsStringLocation*/true, 2186 getSpecifierRange(startPos, posLen)); 2187} 2188 2189void CheckFormatHandler::HandleNullChar(const char *nullCharacter) { 2190 if (!isa<ObjCStringLiteral>(OrigFormatExpr)) { 2191 // The presence of a null character is likely an error. 2192 EmitFormatDiagnostic( 2193 S.PDiag(diag::warn_printf_format_string_contains_null_char), 2194 getLocationOfByte(nullCharacter), /*IsStringLocation*/true, 2195 getFormatStringRange()); 2196 } 2197} 2198 2199// Note that this may return NULL if there was an error parsing or building 2200// one of the argument expressions. 2201const Expr *CheckFormatHandler::getDataArg(unsigned i) const { 2202 return Args[FirstDataArg + i]; 2203} 2204 2205void CheckFormatHandler::DoneProcessing() { 2206 // Does the number of data arguments exceed the number of 2207 // format conversions in the format string? 2208 if (!HasVAListArg) { 2209 // Find any arguments that weren't covered. 2210 CoveredArgs.flip(); 2211 signed notCoveredArg = CoveredArgs.find_first(); 2212 if (notCoveredArg >= 0) { 2213 assert((unsigned)notCoveredArg < NumDataArgs); 2214 if (const Expr *E = getDataArg((unsigned) notCoveredArg)) { 2215 SourceLocation Loc = E->getLocStart(); 2216 if (!S.getSourceManager().isInSystemMacro(Loc)) { 2217 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_data_arg_not_used), 2218 Loc, /*IsStringLocation*/false, 2219 getFormatStringRange()); 2220 } 2221 } 2222 } 2223 } 2224} 2225 2226bool 2227CheckFormatHandler::HandleInvalidConversionSpecifier(unsigned argIndex, 2228 SourceLocation Loc, 2229 const char *startSpec, 2230 unsigned specifierLen, 2231 const char *csStart, 2232 unsigned csLen) { 2233 2234 bool keepGoing = true; 2235 if (argIndex < NumDataArgs) { 2236 // Consider the argument coverered, even though the specifier doesn't 2237 // make sense. 2238 CoveredArgs.set(argIndex); 2239 } 2240 else { 2241 // If argIndex exceeds the number of data arguments we 2242 // don't issue a warning because that is just a cascade of warnings (and 2243 // they may have intended '%%' anyway). We don't want to continue processing 2244 // the format string after this point, however, as we will like just get 2245 // gibberish when trying to match arguments. 2246 keepGoing = false; 2247 } 2248 2249 EmitFormatDiagnostic(S.PDiag(diag::warn_format_invalid_conversion) 2250 << StringRef(csStart, csLen), 2251 Loc, /*IsStringLocation*/true, 2252 getSpecifierRange(startSpec, specifierLen)); 2253 2254 return keepGoing; 2255} 2256 2257void 2258CheckFormatHandler::HandlePositionalNonpositionalArgs(SourceLocation Loc, 2259 const char *startSpec, 2260 unsigned specifierLen) { 2261 EmitFormatDiagnostic( 2262 S.PDiag(diag::warn_format_mix_positional_nonpositional_args), 2263 Loc, /*isStringLoc*/true, getSpecifierRange(startSpec, specifierLen)); 2264} 2265 2266bool 2267CheckFormatHandler::CheckNumArgs( 2268 const analyze_format_string::FormatSpecifier &FS, 2269 const analyze_format_string::ConversionSpecifier &CS, 2270 const char *startSpecifier, unsigned specifierLen, unsigned argIndex) { 2271 2272 if (argIndex >= NumDataArgs) { 2273 PartialDiagnostic PDiag = FS.usesPositionalArg() 2274 ? (S.PDiag(diag::warn_printf_positional_arg_exceeds_data_args) 2275 << (argIndex+1) << NumDataArgs) 2276 : S.PDiag(diag::warn_printf_insufficient_data_args); 2277 EmitFormatDiagnostic( 2278 PDiag, getLocationOfByte(CS.getStart()), /*IsStringLocation*/true, 2279 getSpecifierRange(startSpecifier, specifierLen)); 2280 return false; 2281 } 2282 return true; 2283} 2284 2285template<typename Range> 2286void CheckFormatHandler::EmitFormatDiagnostic(PartialDiagnostic PDiag, 2287 SourceLocation Loc, 2288 bool IsStringLocation, 2289 Range StringRange, 2290 ArrayRef<FixItHint> FixIt) { 2291 EmitFormatDiagnostic(S, inFunctionCall, Args[FormatIdx], PDiag, 2292 Loc, IsStringLocation, StringRange, FixIt); 2293} 2294 2295/// \brief If the format string is not within the funcion call, emit a note 2296/// so that the function call and string are in diagnostic messages. 2297/// 2298/// \param InFunctionCall if true, the format string is within the function 2299/// call and only one diagnostic message will be produced. Otherwise, an 2300/// extra note will be emitted pointing to location of the format string. 2301/// 2302/// \param ArgumentExpr the expression that is passed as the format string 2303/// argument in the function call. Used for getting locations when two 2304/// diagnostics are emitted. 2305/// 2306/// \param PDiag the callee should already have provided any strings for the 2307/// diagnostic message. This function only adds locations and fixits 2308/// to diagnostics. 2309/// 2310/// \param Loc primary location for diagnostic. If two diagnostics are 2311/// required, one will be at Loc and a new SourceLocation will be created for 2312/// the other one. 2313/// 2314/// \param IsStringLocation if true, Loc points to the format string should be 2315/// used for the note. Otherwise, Loc points to the argument list and will 2316/// be used with PDiag. 2317/// 2318/// \param StringRange some or all of the string to highlight. This is 2319/// templated so it can accept either a CharSourceRange or a SourceRange. 2320/// 2321/// \param FixIt optional fix it hint for the format string. 2322template<typename Range> 2323void CheckFormatHandler::EmitFormatDiagnostic(Sema &S, bool InFunctionCall, 2324 const Expr *ArgumentExpr, 2325 PartialDiagnostic PDiag, 2326 SourceLocation Loc, 2327 bool IsStringLocation, 2328 Range StringRange, 2329 ArrayRef<FixItHint> FixIt) { 2330 if (InFunctionCall) { 2331 const Sema::SemaDiagnosticBuilder &D = S.Diag(Loc, PDiag); 2332 D << StringRange; 2333 for (ArrayRef<FixItHint>::iterator I = FixIt.begin(), E = FixIt.end(); 2334 I != E; ++I) { 2335 D << *I; 2336 } 2337 } else { 2338 S.Diag(IsStringLocation ? ArgumentExpr->getExprLoc() : Loc, PDiag) 2339 << ArgumentExpr->getSourceRange(); 2340 2341 const Sema::SemaDiagnosticBuilder &Note = 2342 S.Diag(IsStringLocation ? Loc : StringRange.getBegin(), 2343 diag::note_format_string_defined); 2344 2345 Note << StringRange; 2346 for (ArrayRef<FixItHint>::iterator I = FixIt.begin(), E = FixIt.end(); 2347 I != E; ++I) { 2348 Note << *I; 2349 } 2350 } 2351} 2352 2353//===--- CHECK: Printf format string checking ------------------------------===// 2354 2355namespace { 2356class CheckPrintfHandler : public CheckFormatHandler { 2357 bool ObjCContext; 2358public: 2359 CheckPrintfHandler(Sema &s, const StringLiteral *fexpr, 2360 const Expr *origFormatExpr, unsigned firstDataArg, 2361 unsigned numDataArgs, bool isObjC, 2362 const char *beg, bool hasVAListArg, 2363 ArrayRef<const Expr *> Args, 2364 unsigned formatIdx, bool inFunctionCall, 2365 Sema::VariadicCallType CallType) 2366 : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg, 2367 numDataArgs, beg, hasVAListArg, Args, 2368 formatIdx, inFunctionCall, CallType), ObjCContext(isObjC) 2369 {} 2370 2371 2372 bool HandleInvalidPrintfConversionSpecifier( 2373 const analyze_printf::PrintfSpecifier &FS, 2374 const char *startSpecifier, 2375 unsigned specifierLen); 2376 2377 bool HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier &FS, 2378 const char *startSpecifier, 2379 unsigned specifierLen); 2380 bool checkFormatExpr(const analyze_printf::PrintfSpecifier &FS, 2381 const char *StartSpecifier, 2382 unsigned SpecifierLen, 2383 const Expr *E); 2384 2385 bool HandleAmount(const analyze_format_string::OptionalAmount &Amt, unsigned k, 2386 const char *startSpecifier, unsigned specifierLen); 2387 void HandleInvalidAmount(const analyze_printf::PrintfSpecifier &FS, 2388 const analyze_printf::OptionalAmount &Amt, 2389 unsigned type, 2390 const char *startSpecifier, unsigned specifierLen); 2391 void HandleFlag(const analyze_printf::PrintfSpecifier &FS, 2392 const analyze_printf::OptionalFlag &flag, 2393 const char *startSpecifier, unsigned specifierLen); 2394 void HandleIgnoredFlag(const analyze_printf::PrintfSpecifier &FS, 2395 const analyze_printf::OptionalFlag &ignoredFlag, 2396 const analyze_printf::OptionalFlag &flag, 2397 const char *startSpecifier, unsigned specifierLen); 2398 bool checkForCStrMembers(const analyze_printf::ArgType &AT, 2399 const Expr *E, const CharSourceRange &CSR); 2400 2401}; 2402} 2403 2404bool CheckPrintfHandler::HandleInvalidPrintfConversionSpecifier( 2405 const analyze_printf::PrintfSpecifier &FS, 2406 const char *startSpecifier, 2407 unsigned specifierLen) { 2408 const analyze_printf::PrintfConversionSpecifier &CS = 2409 FS.getConversionSpecifier(); 2410 2411 return HandleInvalidConversionSpecifier(FS.getArgIndex(), 2412 getLocationOfByte(CS.getStart()), 2413 startSpecifier, specifierLen, 2414 CS.getStart(), CS.getLength()); 2415} 2416 2417bool CheckPrintfHandler::HandleAmount( 2418 const analyze_format_string::OptionalAmount &Amt, 2419 unsigned k, const char *startSpecifier, 2420 unsigned specifierLen) { 2421 2422 if (Amt.hasDataArgument()) { 2423 if (!HasVAListArg) { 2424 unsigned argIndex = Amt.getArgIndex(); 2425 if (argIndex >= NumDataArgs) { 2426 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_asterisk_missing_arg) 2427 << k, 2428 getLocationOfByte(Amt.getStart()), 2429 /*IsStringLocation*/true, 2430 getSpecifierRange(startSpecifier, specifierLen)); 2431 // Don't do any more checking. We will just emit 2432 // spurious errors. 2433 return false; 2434 } 2435 2436 // Type check the data argument. It should be an 'int'. 2437 // Although not in conformance with C99, we also allow the argument to be 2438 // an 'unsigned int' as that is a reasonably safe case. GCC also 2439 // doesn't emit a warning for that case. 2440 CoveredArgs.set(argIndex); 2441 const Expr *Arg = getDataArg(argIndex); 2442 if (!Arg) 2443 return false; 2444 2445 QualType T = Arg->getType(); 2446 2447 const analyze_printf::ArgType &AT = Amt.getArgType(S.Context); 2448 assert(AT.isValid()); 2449 2450 if (!AT.matchesType(S.Context, T)) { 2451 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_asterisk_wrong_type) 2452 << k << AT.getRepresentativeTypeName(S.Context) 2453 << T << Arg->getSourceRange(), 2454 getLocationOfByte(Amt.getStart()), 2455 /*IsStringLocation*/true, 2456 getSpecifierRange(startSpecifier, specifierLen)); 2457 // Don't do any more checking. We will just emit 2458 // spurious errors. 2459 return false; 2460 } 2461 } 2462 } 2463 return true; 2464} 2465 2466void CheckPrintfHandler::HandleInvalidAmount( 2467 const analyze_printf::PrintfSpecifier &FS, 2468 const analyze_printf::OptionalAmount &Amt, 2469 unsigned type, 2470 const char *startSpecifier, 2471 unsigned specifierLen) { 2472 const analyze_printf::PrintfConversionSpecifier &CS = 2473 FS.getConversionSpecifier(); 2474 2475 FixItHint fixit = 2476 Amt.getHowSpecified() == analyze_printf::OptionalAmount::Constant 2477 ? FixItHint::CreateRemoval(getSpecifierRange(Amt.getStart(), 2478 Amt.getConstantLength())) 2479 : FixItHint(); 2480 2481 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_nonsensical_optional_amount) 2482 << type << CS.toString(), 2483 getLocationOfByte(Amt.getStart()), 2484 /*IsStringLocation*/true, 2485 getSpecifierRange(startSpecifier, specifierLen), 2486 fixit); 2487} 2488 2489void CheckPrintfHandler::HandleFlag(const analyze_printf::PrintfSpecifier &FS, 2490 const analyze_printf::OptionalFlag &flag, 2491 const char *startSpecifier, 2492 unsigned specifierLen) { 2493 // Warn about pointless flag with a fixit removal. 2494 const analyze_printf::PrintfConversionSpecifier &CS = 2495 FS.getConversionSpecifier(); 2496 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_nonsensical_flag) 2497 << flag.toString() << CS.toString(), 2498 getLocationOfByte(flag.getPosition()), 2499 /*IsStringLocation*/true, 2500 getSpecifierRange(startSpecifier, specifierLen), 2501 FixItHint::CreateRemoval( 2502 getSpecifierRange(flag.getPosition(), 1))); 2503} 2504 2505void CheckPrintfHandler::HandleIgnoredFlag( 2506 const analyze_printf::PrintfSpecifier &FS, 2507 const analyze_printf::OptionalFlag &ignoredFlag, 2508 const analyze_printf::OptionalFlag &flag, 2509 const char *startSpecifier, 2510 unsigned specifierLen) { 2511 // Warn about ignored flag with a fixit removal. 2512 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_ignored_flag) 2513 << ignoredFlag.toString() << flag.toString(), 2514 getLocationOfByte(ignoredFlag.getPosition()), 2515 /*IsStringLocation*/true, 2516 getSpecifierRange(startSpecifier, specifierLen), 2517 FixItHint::CreateRemoval( 2518 getSpecifierRange(ignoredFlag.getPosition(), 1))); 2519} 2520 2521// Determines if the specified is a C++ class or struct containing 2522// a member with the specified name and kind (e.g. a CXXMethodDecl named 2523// "c_str()"). 2524template<typename MemberKind> 2525static llvm::SmallPtrSet<MemberKind*, 1> 2526CXXRecordMembersNamed(StringRef Name, Sema &S, QualType Ty) { 2527 const RecordType *RT = Ty->getAs<RecordType>(); 2528 llvm::SmallPtrSet<MemberKind*, 1> Results; 2529 2530 if (!RT) 2531 return Results; 2532 const CXXRecordDecl *RD = dyn_cast<CXXRecordDecl>(RT->getDecl()); 2533 if (!RD) 2534 return Results; 2535 2536 LookupResult R(S, &S.PP.getIdentifierTable().get(Name), SourceLocation(), 2537 Sema::LookupMemberName); 2538 2539 // We just need to include all members of the right kind turned up by the 2540 // filter, at this point. 2541 if (S.LookupQualifiedName(R, RT->getDecl())) 2542 for (LookupResult::iterator I = R.begin(), E = R.end(); I != E; ++I) { 2543 NamedDecl *decl = (*I)->getUnderlyingDecl(); 2544 if (MemberKind *FK = dyn_cast<MemberKind>(decl)) 2545 Results.insert(FK); 2546 } 2547 return Results; 2548} 2549 2550// Check if a (w)string was passed when a (w)char* was needed, and offer a 2551// better diagnostic if so. AT is assumed to be valid. 2552// Returns true when a c_str() conversion method is found. 2553bool CheckPrintfHandler::checkForCStrMembers( 2554 const analyze_printf::ArgType &AT, const Expr *E, 2555 const CharSourceRange &CSR) { 2556 typedef llvm::SmallPtrSet<CXXMethodDecl*, 1> MethodSet; 2557 2558 MethodSet Results = 2559 CXXRecordMembersNamed<CXXMethodDecl>("c_str", S, E->getType()); 2560 2561 for (MethodSet::iterator MI = Results.begin(), ME = Results.end(); 2562 MI != ME; ++MI) { 2563 const CXXMethodDecl *Method = *MI; 2564 if (Method->getNumParams() == 0 && 2565 AT.matchesType(S.Context, Method->getResultType())) { 2566 // FIXME: Suggest parens if the expression needs them. 2567 SourceLocation EndLoc = 2568 S.getPreprocessor().getLocForEndOfToken(E->getLocEnd()); 2569 S.Diag(E->getLocStart(), diag::note_printf_c_str) 2570 << "c_str()" 2571 << FixItHint::CreateInsertion(EndLoc, ".c_str()"); 2572 return true; 2573 } 2574 } 2575 2576 return false; 2577} 2578 2579bool 2580CheckPrintfHandler::HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier 2581 &FS, 2582 const char *startSpecifier, 2583 unsigned specifierLen) { 2584 2585 using namespace analyze_format_string; 2586 using namespace analyze_printf; 2587 const PrintfConversionSpecifier &CS = FS.getConversionSpecifier(); 2588 2589 if (FS.consumesDataArgument()) { 2590 if (atFirstArg) { 2591 atFirstArg = false; 2592 usesPositionalArgs = FS.usesPositionalArg(); 2593 } 2594 else if (usesPositionalArgs != FS.usesPositionalArg()) { 2595 HandlePositionalNonpositionalArgs(getLocationOfByte(CS.getStart()), 2596 startSpecifier, specifierLen); 2597 return false; 2598 } 2599 } 2600 2601 // First check if the field width, precision, and conversion specifier 2602 // have matching data arguments. 2603 if (!HandleAmount(FS.getFieldWidth(), /* field width */ 0, 2604 startSpecifier, specifierLen)) { 2605 return false; 2606 } 2607 2608 if (!HandleAmount(FS.getPrecision(), /* precision */ 1, 2609 startSpecifier, specifierLen)) { 2610 return false; 2611 } 2612 2613 if (!CS.consumesDataArgument()) { 2614 // FIXME: Technically specifying a precision or field width here 2615 // makes no sense. Worth issuing a warning at some point. 2616 return true; 2617 } 2618 2619 // Consume the argument. 2620 unsigned argIndex = FS.getArgIndex(); 2621 if (argIndex < NumDataArgs) { 2622 // The check to see if the argIndex is valid will come later. 2623 // We set the bit here because we may exit early from this 2624 // function if we encounter some other error. 2625 CoveredArgs.set(argIndex); 2626 } 2627 2628 // Check for using an Objective-C specific conversion specifier 2629 // in a non-ObjC literal. 2630 if (!ObjCContext && CS.isObjCArg()) { 2631 return HandleInvalidPrintfConversionSpecifier(FS, startSpecifier, 2632 specifierLen); 2633 } 2634 2635 // Check for invalid use of field width 2636 if (!FS.hasValidFieldWidth()) { 2637 HandleInvalidAmount(FS, FS.getFieldWidth(), /* field width */ 0, 2638 startSpecifier, specifierLen); 2639 } 2640 2641 // Check for invalid use of precision 2642 if (!FS.hasValidPrecision()) { 2643 HandleInvalidAmount(FS, FS.getPrecision(), /* precision */ 1, 2644 startSpecifier, specifierLen); 2645 } 2646 2647 // Check each flag does not conflict with any other component. 2648 if (!FS.hasValidThousandsGroupingPrefix()) 2649 HandleFlag(FS, FS.hasThousandsGrouping(), startSpecifier, specifierLen); 2650 if (!FS.hasValidLeadingZeros()) 2651 HandleFlag(FS, FS.hasLeadingZeros(), startSpecifier, specifierLen); 2652 if (!FS.hasValidPlusPrefix()) 2653 HandleFlag(FS, FS.hasPlusPrefix(), startSpecifier, specifierLen); 2654 if (!FS.hasValidSpacePrefix()) 2655 HandleFlag(FS, FS.hasSpacePrefix(), startSpecifier, specifierLen); 2656 if (!FS.hasValidAlternativeForm()) 2657 HandleFlag(FS, FS.hasAlternativeForm(), startSpecifier, specifierLen); 2658 if (!FS.hasValidLeftJustified()) 2659 HandleFlag(FS, FS.isLeftJustified(), startSpecifier, specifierLen); 2660 2661 // Check that flags are not ignored by another flag 2662 if (FS.hasSpacePrefix() && FS.hasPlusPrefix()) // ' ' ignored by '+' 2663 HandleIgnoredFlag(FS, FS.hasSpacePrefix(), FS.hasPlusPrefix(), 2664 startSpecifier, specifierLen); 2665 if (FS.hasLeadingZeros() && FS.isLeftJustified()) // '0' ignored by '-' 2666 HandleIgnoredFlag(FS, FS.hasLeadingZeros(), FS.isLeftJustified(), 2667 startSpecifier, specifierLen); 2668 2669 // Check the length modifier is valid with the given conversion specifier. 2670 if (!FS.hasValidLengthModifier(S.getASTContext().getTargetInfo())) 2671 HandleInvalidLengthModifier(FS, CS, startSpecifier, specifierLen, 2672 diag::warn_format_nonsensical_length); 2673 else if (!FS.hasStandardLengthModifier()) 2674 HandleNonStandardLengthModifier(FS, startSpecifier, specifierLen); 2675 else if (!FS.hasStandardLengthConversionCombination()) 2676 HandleInvalidLengthModifier(FS, CS, startSpecifier, specifierLen, 2677 diag::warn_format_non_standard_conversion_spec); 2678 2679 if (!FS.hasStandardConversionSpecifier(S.getLangOpts())) 2680 HandleNonStandardConversionSpecifier(CS, startSpecifier, specifierLen); 2681 2682 // The remaining checks depend on the data arguments. 2683 if (HasVAListArg) 2684 return true; 2685 2686 if (!CheckNumArgs(FS, CS, startSpecifier, specifierLen, argIndex)) 2687 return false; 2688 2689 const Expr *Arg = getDataArg(argIndex); 2690 if (!Arg) 2691 return true; 2692 2693 return checkFormatExpr(FS, startSpecifier, specifierLen, Arg); 2694} 2695 2696static bool requiresParensToAddCast(const Expr *E) { 2697 // FIXME: We should have a general way to reason about operator 2698 // precedence and whether parens are actually needed here. 2699 // Take care of a few common cases where they aren't. 2700 const Expr *Inside = E->IgnoreImpCasts(); 2701 if (const PseudoObjectExpr *POE = dyn_cast<PseudoObjectExpr>(Inside)) 2702 Inside = POE->getSyntacticForm()->IgnoreImpCasts(); 2703 2704 switch (Inside->getStmtClass()) { 2705 case Stmt::ArraySubscriptExprClass: 2706 case Stmt::CallExprClass: 2707 case Stmt::CharacterLiteralClass: 2708 case Stmt::CXXBoolLiteralExprClass: 2709 case Stmt::DeclRefExprClass: 2710 case Stmt::FloatingLiteralClass: 2711 case Stmt::IntegerLiteralClass: 2712 case Stmt::MemberExprClass: 2713 case Stmt::ObjCArrayLiteralClass: 2714 case Stmt::ObjCBoolLiteralExprClass: 2715 case Stmt::ObjCBoxedExprClass: 2716 case Stmt::ObjCDictionaryLiteralClass: 2717 case Stmt::ObjCEncodeExprClass: 2718 case Stmt::ObjCIvarRefExprClass: 2719 case Stmt::ObjCMessageExprClass: 2720 case Stmt::ObjCPropertyRefExprClass: 2721 case Stmt::ObjCStringLiteralClass: 2722 case Stmt::ObjCSubscriptRefExprClass: 2723 case Stmt::ParenExprClass: 2724 case Stmt::StringLiteralClass: 2725 case Stmt::UnaryOperatorClass: 2726 return false; 2727 default: 2728 return true; 2729 } 2730} 2731 2732bool 2733CheckPrintfHandler::checkFormatExpr(const analyze_printf::PrintfSpecifier &FS, 2734 const char *StartSpecifier, 2735 unsigned SpecifierLen, 2736 const Expr *E) { 2737 using namespace analyze_format_string; 2738 using namespace analyze_printf; 2739 // Now type check the data expression that matches the 2740 // format specifier. 2741 const analyze_printf::ArgType &AT = FS.getArgType(S.Context, 2742 ObjCContext); 2743 if (!AT.isValid()) 2744 return true; 2745 2746 QualType ExprTy = E->getType(); 2747 if (AT.matchesType(S.Context, ExprTy)) 2748 return true; 2749 2750 // Look through argument promotions for our error message's reported type. 2751 // This includes the integral and floating promotions, but excludes array 2752 // and function pointer decay; seeing that an argument intended to be a 2753 // string has type 'char [6]' is probably more confusing than 'char *'. 2754 if (const ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(E)) { 2755 if (ICE->getCastKind() == CK_IntegralCast || 2756 ICE->getCastKind() == CK_FloatingCast) { 2757 E = ICE->getSubExpr(); 2758 ExprTy = E->getType(); 2759 2760 // Check if we didn't match because of an implicit cast from a 'char' 2761 // or 'short' to an 'int'. This is done because printf is a varargs 2762 // function. 2763 if (ICE->getType() == S.Context.IntTy || 2764 ICE->getType() == S.Context.UnsignedIntTy) { 2765 // All further checking is done on the subexpression. 2766 if (AT.matchesType(S.Context, ExprTy)) 2767 return true; 2768 } 2769 } 2770 } else if (const CharacterLiteral *CL = dyn_cast<CharacterLiteral>(E)) { 2771 // Special case for 'a', which has type 'int' in C. 2772 // Note, however, that we do /not/ want to treat multibyte constants like 2773 // 'MooV' as characters! This form is deprecated but still exists. 2774 if (ExprTy == S.Context.IntTy) 2775 if (llvm::isUIntN(S.Context.getCharWidth(), CL->getValue())) 2776 ExprTy = S.Context.CharTy; 2777 } 2778 2779 // %C in an Objective-C context prints a unichar, not a wchar_t. 2780 // If the argument is an integer of some kind, believe the %C and suggest 2781 // a cast instead of changing the conversion specifier. 2782 QualType IntendedTy = ExprTy; 2783 if (ObjCContext && 2784 FS.getConversionSpecifier().getKind() == ConversionSpecifier::CArg) { 2785 if (ExprTy->isIntegralOrUnscopedEnumerationType() && 2786 !ExprTy->isCharType()) { 2787 // 'unichar' is defined as a typedef of unsigned short, but we should 2788 // prefer using the typedef if it is visible. 2789 IntendedTy = S.Context.UnsignedShortTy; 2790 2791 LookupResult Result(S, &S.Context.Idents.get("unichar"), E->getLocStart(), 2792 Sema::LookupOrdinaryName); 2793 if (S.LookupName(Result, S.getCurScope())) { 2794 NamedDecl *ND = Result.getFoundDecl(); 2795 if (TypedefNameDecl *TD = dyn_cast<TypedefNameDecl>(ND)) 2796 if (TD->getUnderlyingType() == IntendedTy) 2797 IntendedTy = S.Context.getTypedefType(TD); 2798 } 2799 } 2800 } 2801 2802 // Special-case some of Darwin's platform-independence types by suggesting 2803 // casts to primitive types that are known to be large enough. 2804 bool ShouldNotPrintDirectly = false; 2805 if (S.Context.getTargetInfo().getTriple().isOSDarwin()) { 2806 // Use a 'while' to peel off layers of typedefs. 2807 QualType TyTy = IntendedTy; 2808 while (const TypedefType *UserTy = TyTy->getAs<TypedefType>()) { 2809 StringRef Name = UserTy->getDecl()->getName(); 2810 QualType CastTy = llvm::StringSwitch<QualType>(Name) 2811 .Case("NSInteger", S.Context.LongTy) 2812 .Case("NSUInteger", S.Context.UnsignedLongTy) 2813 .Case("SInt32", S.Context.IntTy) 2814 .Case("UInt32", S.Context.UnsignedIntTy) 2815 .Default(QualType()); 2816 2817 if (!CastTy.isNull()) { 2818 ShouldNotPrintDirectly = true; 2819 IntendedTy = CastTy; 2820 break; 2821 } 2822 TyTy = UserTy->desugar(); 2823 } 2824 } 2825 2826 // We may be able to offer a FixItHint if it is a supported type. 2827 PrintfSpecifier fixedFS = FS; 2828 bool success = fixedFS.fixType(IntendedTy, S.getLangOpts(), 2829 S.Context, ObjCContext); 2830 2831 if (success) { 2832 // Get the fix string from the fixed format specifier 2833 SmallString<16> buf; 2834 llvm::raw_svector_ostream os(buf); 2835 fixedFS.toString(os); 2836 2837 CharSourceRange SpecRange = getSpecifierRange(StartSpecifier, SpecifierLen); 2838 2839 if (IntendedTy == ExprTy) { 2840 // In this case, the specifier is wrong and should be changed to match 2841 // the argument. 2842 EmitFormatDiagnostic( 2843 S.PDiag(diag::warn_printf_conversion_argument_type_mismatch) 2844 << AT.getRepresentativeTypeName(S.Context) << IntendedTy 2845 << E->getSourceRange(), 2846 E->getLocStart(), 2847 /*IsStringLocation*/false, 2848 SpecRange, 2849 FixItHint::CreateReplacement(SpecRange, os.str())); 2850 2851 } else { 2852 // The canonical type for formatting this value is different from the 2853 // actual type of the expression. (This occurs, for example, with Darwin's 2854 // NSInteger on 32-bit platforms, where it is typedef'd as 'int', but 2855 // should be printed as 'long' for 64-bit compatibility.) 2856 // Rather than emitting a normal format/argument mismatch, we want to 2857 // add a cast to the recommended type (and correct the format string 2858 // if necessary). 2859 SmallString<16> CastBuf; 2860 llvm::raw_svector_ostream CastFix(CastBuf); 2861 CastFix << "("; 2862 IntendedTy.print(CastFix, S.Context.getPrintingPolicy()); 2863 CastFix << ")"; 2864 2865 SmallVector<FixItHint,4> Hints; 2866 if (!AT.matchesType(S.Context, IntendedTy)) 2867 Hints.push_back(FixItHint::CreateReplacement(SpecRange, os.str())); 2868 2869 if (const CStyleCastExpr *CCast = dyn_cast<CStyleCastExpr>(E)) { 2870 // If there's already a cast present, just replace it. 2871 SourceRange CastRange(CCast->getLParenLoc(), CCast->getRParenLoc()); 2872 Hints.push_back(FixItHint::CreateReplacement(CastRange, CastFix.str())); 2873 2874 } else if (!requiresParensToAddCast(E)) { 2875 // If the expression has high enough precedence, 2876 // just write the C-style cast. 2877 Hints.push_back(FixItHint::CreateInsertion(E->getLocStart(), 2878 CastFix.str())); 2879 } else { 2880 // Otherwise, add parens around the expression as well as the cast. 2881 CastFix << "("; 2882 Hints.push_back(FixItHint::CreateInsertion(E->getLocStart(), 2883 CastFix.str())); 2884 2885 SourceLocation After = S.PP.getLocForEndOfToken(E->getLocEnd()); 2886 Hints.push_back(FixItHint::CreateInsertion(After, ")")); 2887 } 2888 2889 if (ShouldNotPrintDirectly) { 2890 // The expression has a type that should not be printed directly. 2891 // We extract the name from the typedef because we don't want to show 2892 // the underlying type in the diagnostic. 2893 StringRef Name = cast<TypedefType>(ExprTy)->getDecl()->getName(); 2894 2895 EmitFormatDiagnostic(S.PDiag(diag::warn_format_argument_needs_cast) 2896 << Name << IntendedTy 2897 << E->getSourceRange(), 2898 E->getLocStart(), /*IsStringLocation=*/false, 2899 SpecRange, Hints); 2900 } else { 2901 // In this case, the expression could be printed using a different 2902 // specifier, but we've decided that the specifier is probably correct 2903 // and we should cast instead. Just use the normal warning message. 2904 EmitFormatDiagnostic( 2905 S.PDiag(diag::warn_printf_conversion_argument_type_mismatch) 2906 << AT.getRepresentativeTypeName(S.Context) << ExprTy 2907 << E->getSourceRange(), 2908 E->getLocStart(), /*IsStringLocation*/false, 2909 SpecRange, Hints); 2910 } 2911 } 2912 } else { 2913 const CharSourceRange &CSR = getSpecifierRange(StartSpecifier, 2914 SpecifierLen); 2915 // Since the warning for passing non-POD types to variadic functions 2916 // was deferred until now, we emit a warning for non-POD 2917 // arguments here. 2918 if (S.isValidVarArgType(ExprTy) == Sema::VAK_Invalid) { 2919 unsigned DiagKind; 2920 if (ExprTy->isObjCObjectType()) 2921 DiagKind = diag::err_cannot_pass_objc_interface_to_vararg_format; 2922 else 2923 DiagKind = diag::warn_non_pod_vararg_with_format_string; 2924 2925 EmitFormatDiagnostic( 2926 S.PDiag(DiagKind) 2927 << S.getLangOpts().CPlusPlus11 2928 << ExprTy 2929 << CallType 2930 << AT.getRepresentativeTypeName(S.Context) 2931 << CSR 2932 << E->getSourceRange(), 2933 E->getLocStart(), /*IsStringLocation*/false, CSR); 2934 2935 checkForCStrMembers(AT, E, CSR); 2936 } else 2937 EmitFormatDiagnostic( 2938 S.PDiag(diag::warn_printf_conversion_argument_type_mismatch) 2939 << AT.getRepresentativeTypeName(S.Context) << ExprTy 2940 << CSR 2941 << E->getSourceRange(), 2942 E->getLocStart(), /*IsStringLocation*/false, CSR); 2943 } 2944 2945 return true; 2946} 2947 2948//===--- CHECK: Scanf format string checking ------------------------------===// 2949 2950namespace { 2951class CheckScanfHandler : public CheckFormatHandler { 2952public: 2953 CheckScanfHandler(Sema &s, const StringLiteral *fexpr, 2954 const Expr *origFormatExpr, unsigned firstDataArg, 2955 unsigned numDataArgs, const char *beg, bool hasVAListArg, 2956 ArrayRef<const Expr *> Args, 2957 unsigned formatIdx, bool inFunctionCall, 2958 Sema::VariadicCallType CallType) 2959 : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg, 2960 numDataArgs, beg, hasVAListArg, 2961 Args, formatIdx, inFunctionCall, CallType) 2962 {} 2963 2964 bool HandleScanfSpecifier(const analyze_scanf::ScanfSpecifier &FS, 2965 const char *startSpecifier, 2966 unsigned specifierLen); 2967 2968 bool HandleInvalidScanfConversionSpecifier( 2969 const analyze_scanf::ScanfSpecifier &FS, 2970 const char *startSpecifier, 2971 unsigned specifierLen); 2972 2973 void HandleIncompleteScanList(const char *start, const char *end); 2974}; 2975} 2976 2977void CheckScanfHandler::HandleIncompleteScanList(const char *start, 2978 const char *end) { 2979 EmitFormatDiagnostic(S.PDiag(diag::warn_scanf_scanlist_incomplete), 2980 getLocationOfByte(end), /*IsStringLocation*/true, 2981 getSpecifierRange(start, end - start)); 2982} 2983 2984bool CheckScanfHandler::HandleInvalidScanfConversionSpecifier( 2985 const analyze_scanf::ScanfSpecifier &FS, 2986 const char *startSpecifier, 2987 unsigned specifierLen) { 2988 2989 const analyze_scanf::ScanfConversionSpecifier &CS = 2990 FS.getConversionSpecifier(); 2991 2992 return HandleInvalidConversionSpecifier(FS.getArgIndex(), 2993 getLocationOfByte(CS.getStart()), 2994 startSpecifier, specifierLen, 2995 CS.getStart(), CS.getLength()); 2996} 2997 2998bool CheckScanfHandler::HandleScanfSpecifier( 2999 const analyze_scanf::ScanfSpecifier &FS, 3000 const char *startSpecifier, 3001 unsigned specifierLen) { 3002 3003 using namespace analyze_scanf; 3004 using namespace analyze_format_string; 3005 3006 const ScanfConversionSpecifier &CS = FS.getConversionSpecifier(); 3007 3008 // Handle case where '%' and '*' don't consume an argument. These shouldn't 3009 // be used to decide if we are using positional arguments consistently. 3010 if (FS.consumesDataArgument()) { 3011 if (atFirstArg) { 3012 atFirstArg = false; 3013 usesPositionalArgs = FS.usesPositionalArg(); 3014 } 3015 else if (usesPositionalArgs != FS.usesPositionalArg()) { 3016 HandlePositionalNonpositionalArgs(getLocationOfByte(CS.getStart()), 3017 startSpecifier, specifierLen); 3018 return false; 3019 } 3020 } 3021 3022 // Check if the field with is non-zero. 3023 const OptionalAmount &Amt = FS.getFieldWidth(); 3024 if (Amt.getHowSpecified() == OptionalAmount::Constant) { 3025 if (Amt.getConstantAmount() == 0) { 3026 const CharSourceRange &R = getSpecifierRange(Amt.getStart(), 3027 Amt.getConstantLength()); 3028 EmitFormatDiagnostic(S.PDiag(diag::warn_scanf_nonzero_width), 3029 getLocationOfByte(Amt.getStart()), 3030 /*IsStringLocation*/true, R, 3031 FixItHint::CreateRemoval(R)); 3032 } 3033 } 3034 3035 if (!FS.consumesDataArgument()) { 3036 // FIXME: Technically specifying a precision or field width here 3037 // makes no sense. Worth issuing a warning at some point. 3038 return true; 3039 } 3040 3041 // Consume the argument. 3042 unsigned argIndex = FS.getArgIndex(); 3043 if (argIndex < NumDataArgs) { 3044 // The check to see if the argIndex is valid will come later. 3045 // We set the bit here because we may exit early from this 3046 // function if we encounter some other error. 3047 CoveredArgs.set(argIndex); 3048 } 3049 3050 // Check the length modifier is valid with the given conversion specifier. 3051 if (!FS.hasValidLengthModifier(S.getASTContext().getTargetInfo())) 3052 HandleInvalidLengthModifier(FS, CS, startSpecifier, specifierLen, 3053 diag::warn_format_nonsensical_length); 3054 else if (!FS.hasStandardLengthModifier()) 3055 HandleNonStandardLengthModifier(FS, startSpecifier, specifierLen); 3056 else if (!FS.hasStandardLengthConversionCombination()) 3057 HandleInvalidLengthModifier(FS, CS, startSpecifier, specifierLen, 3058 diag::warn_format_non_standard_conversion_spec); 3059 3060 if (!FS.hasStandardConversionSpecifier(S.getLangOpts())) 3061 HandleNonStandardConversionSpecifier(CS, startSpecifier, specifierLen); 3062 3063 // The remaining checks depend on the data arguments. 3064 if (HasVAListArg) 3065 return true; 3066 3067 if (!CheckNumArgs(FS, CS, startSpecifier, specifierLen, argIndex)) 3068 return false; 3069 3070 // Check that the argument type matches the format specifier. 3071 const Expr *Ex = getDataArg(argIndex); 3072 if (!Ex) 3073 return true; 3074 3075 const analyze_format_string::ArgType &AT = FS.getArgType(S.Context); 3076 if (AT.isValid() && !AT.matchesType(S.Context, Ex->getType())) { 3077 ScanfSpecifier fixedFS = FS; 3078 bool success = fixedFS.fixType(Ex->getType(), S.getLangOpts(), 3079 S.Context); 3080 3081 if (success) { 3082 // Get the fix string from the fixed format specifier. 3083 SmallString<128> buf; 3084 llvm::raw_svector_ostream os(buf); 3085 fixedFS.toString(os); 3086 3087 EmitFormatDiagnostic( 3088 S.PDiag(diag::warn_printf_conversion_argument_type_mismatch) 3089 << AT.getRepresentativeTypeName(S.Context) << Ex->getType() 3090 << Ex->getSourceRange(), 3091 Ex->getLocStart(), 3092 /*IsStringLocation*/false, 3093 getSpecifierRange(startSpecifier, specifierLen), 3094 FixItHint::CreateReplacement( 3095 getSpecifierRange(startSpecifier, specifierLen), 3096 os.str())); 3097 } else { 3098 EmitFormatDiagnostic( 3099 S.PDiag(diag::warn_printf_conversion_argument_type_mismatch) 3100 << AT.getRepresentativeTypeName(S.Context) << Ex->getType() 3101 << Ex->getSourceRange(), 3102 Ex->getLocStart(), 3103 /*IsStringLocation*/false, 3104 getSpecifierRange(startSpecifier, specifierLen)); 3105 } 3106 } 3107 3108 return true; 3109} 3110 3111void Sema::CheckFormatString(const StringLiteral *FExpr, 3112 const Expr *OrigFormatExpr, 3113 ArrayRef<const Expr *> Args, 3114 bool HasVAListArg, unsigned format_idx, 3115 unsigned firstDataArg, FormatStringType Type, 3116 bool inFunctionCall, VariadicCallType CallType) { 3117 3118 // CHECK: is the format string a wide literal? 3119 if (!FExpr->isAscii() && !FExpr->isUTF8()) { 3120 CheckFormatHandler::EmitFormatDiagnostic( 3121 *this, inFunctionCall, Args[format_idx], 3122 PDiag(diag::warn_format_string_is_wide_literal), FExpr->getLocStart(), 3123 /*IsStringLocation*/true, OrigFormatExpr->getSourceRange()); 3124 return; 3125 } 3126 3127 // Str - The format string. NOTE: this is NOT null-terminated! 3128 StringRef StrRef = FExpr->getString(); 3129 const char *Str = StrRef.data(); 3130 unsigned StrLen = StrRef.size(); 3131 const unsigned numDataArgs = Args.size() - firstDataArg; 3132 3133 // CHECK: empty format string? 3134 if (StrLen == 0 && numDataArgs > 0) { 3135 CheckFormatHandler::EmitFormatDiagnostic( 3136 *this, inFunctionCall, Args[format_idx], 3137 PDiag(diag::warn_empty_format_string), FExpr->getLocStart(), 3138 /*IsStringLocation*/true, OrigFormatExpr->getSourceRange()); 3139 return; 3140 } 3141 3142 if (Type == FST_Printf || Type == FST_NSString) { 3143 CheckPrintfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg, 3144 numDataArgs, (Type == FST_NSString), 3145 Str, HasVAListArg, Args, format_idx, 3146 inFunctionCall, CallType); 3147 3148 if (!analyze_format_string::ParsePrintfString(H, Str, Str + StrLen, 3149 getLangOpts(), 3150 Context.getTargetInfo())) 3151 H.DoneProcessing(); 3152 } else if (Type == FST_Scanf) { 3153 CheckScanfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg, numDataArgs, 3154 Str, HasVAListArg, Args, format_idx, 3155 inFunctionCall, CallType); 3156 3157 if (!analyze_format_string::ParseScanfString(H, Str, Str + StrLen, 3158 getLangOpts(), 3159 Context.getTargetInfo())) 3160 H.DoneProcessing(); 3161 } // TODO: handle other formats 3162} 3163 3164//===--- CHECK: Standard memory functions ---------------------------------===// 3165 3166/// \brief Determine whether the given type is a dynamic class type (e.g., 3167/// whether it has a vtable). 3168static bool isDynamicClassType(QualType T) { 3169 if (CXXRecordDecl *Record = T->getAsCXXRecordDecl()) 3170 if (CXXRecordDecl *Definition = Record->getDefinition()) 3171 if (Definition->isDynamicClass()) 3172 return true; 3173 3174 return false; 3175} 3176 3177/// \brief If E is a sizeof expression, returns its argument expression, 3178/// otherwise returns NULL. 3179static const Expr *getSizeOfExprArg(const Expr* E) { 3180 if (const UnaryExprOrTypeTraitExpr *SizeOf = 3181 dyn_cast<UnaryExprOrTypeTraitExpr>(E)) 3182 if (SizeOf->getKind() == clang::UETT_SizeOf && !SizeOf->isArgumentType()) 3183 return SizeOf->getArgumentExpr()->IgnoreParenImpCasts(); 3184 3185 return 0; 3186} 3187 3188/// \brief If E is a sizeof expression, returns its argument type. 3189static QualType getSizeOfArgType(const Expr* E) { 3190 if (const UnaryExprOrTypeTraitExpr *SizeOf = 3191 dyn_cast<UnaryExprOrTypeTraitExpr>(E)) 3192 if (SizeOf->getKind() == clang::UETT_SizeOf) 3193 return SizeOf->getTypeOfArgument(); 3194 3195 return QualType(); 3196} 3197 3198/// \brief Check for dangerous or invalid arguments to memset(). 3199/// 3200/// This issues warnings on known problematic, dangerous or unspecified 3201/// arguments to the standard 'memset', 'memcpy', 'memmove', and 'memcmp' 3202/// function calls. 3203/// 3204/// \param Call The call expression to diagnose. 3205void Sema::CheckMemaccessArguments(const CallExpr *Call, 3206 unsigned BId, 3207 IdentifierInfo *FnName) { 3208 assert(BId != 0); 3209 3210 // It is possible to have a non-standard definition of memset. Validate 3211 // we have enough arguments, and if not, abort further checking. 3212 unsigned ExpectedNumArgs = (BId == Builtin::BIstrndup ? 2 : 3); 3213 if (Call->getNumArgs() < ExpectedNumArgs) 3214 return; 3215 3216 unsigned LastArg = (BId == Builtin::BImemset || 3217 BId == Builtin::BIstrndup ? 1 : 2); 3218 unsigned LenArg = (BId == Builtin::BIstrndup ? 1 : 2); 3219 const Expr *LenExpr = Call->getArg(LenArg)->IgnoreParenImpCasts(); 3220 3221 // We have special checking when the length is a sizeof expression. 3222 QualType SizeOfArgTy = getSizeOfArgType(LenExpr); 3223 const Expr *SizeOfArg = getSizeOfExprArg(LenExpr); 3224 llvm::FoldingSetNodeID SizeOfArgID; 3225 3226 for (unsigned ArgIdx = 0; ArgIdx != LastArg; ++ArgIdx) { 3227 const Expr *Dest = Call->getArg(ArgIdx)->IgnoreParenImpCasts(); 3228 SourceRange ArgRange = Call->getArg(ArgIdx)->getSourceRange(); 3229 3230 QualType DestTy = Dest->getType(); 3231 if (const PointerType *DestPtrTy = DestTy->getAs<PointerType>()) { 3232 QualType PointeeTy = DestPtrTy->getPointeeType(); 3233 3234 // Never warn about void type pointers. This can be used to suppress 3235 // false positives. 3236 if (PointeeTy->isVoidType()) 3237 continue; 3238 3239 // Catch "memset(p, 0, sizeof(p))" -- needs to be sizeof(*p). Do this by 3240 // actually comparing the expressions for equality. Because computing the 3241 // expression IDs can be expensive, we only do this if the diagnostic is 3242 // enabled. 3243 if (SizeOfArg && 3244 Diags.getDiagnosticLevel(diag::warn_sizeof_pointer_expr_memaccess, 3245 SizeOfArg->getExprLoc())) { 3246 // We only compute IDs for expressions if the warning is enabled, and 3247 // cache the sizeof arg's ID. 3248 if (SizeOfArgID == llvm::FoldingSetNodeID()) 3249 SizeOfArg->Profile(SizeOfArgID, Context, true); 3250 llvm::FoldingSetNodeID DestID; 3251 Dest->Profile(DestID, Context, true); 3252 if (DestID == SizeOfArgID) { 3253 // TODO: For strncpy() and friends, this could suggest sizeof(dst) 3254 // over sizeof(src) as well. 3255 unsigned ActionIdx = 0; // Default is to suggest dereferencing. 3256 StringRef ReadableName = FnName->getName(); 3257 3258 if (const UnaryOperator *UnaryOp = dyn_cast<UnaryOperator>(Dest)) 3259 if (UnaryOp->getOpcode() == UO_AddrOf) 3260 ActionIdx = 1; // If its an address-of operator, just remove it. 3261 if (!PointeeTy->isIncompleteType() && 3262 (Context.getTypeSize(PointeeTy) == Context.getCharWidth())) 3263 ActionIdx = 2; // If the pointee's size is sizeof(char), 3264 // suggest an explicit length. 3265 3266 // If the function is defined as a builtin macro, do not show macro 3267 // expansion. 3268 SourceLocation SL = SizeOfArg->getExprLoc(); 3269 SourceRange DSR = Dest->getSourceRange(); 3270 SourceRange SSR = SizeOfArg->getSourceRange(); 3271 SourceManager &SM = PP.getSourceManager(); 3272 3273 if (SM.isMacroArgExpansion(SL)) { 3274 ReadableName = Lexer::getImmediateMacroName(SL, SM, LangOpts); 3275 SL = SM.getSpellingLoc(SL); 3276 DSR = SourceRange(SM.getSpellingLoc(DSR.getBegin()), 3277 SM.getSpellingLoc(DSR.getEnd())); 3278 SSR = SourceRange(SM.getSpellingLoc(SSR.getBegin()), 3279 SM.getSpellingLoc(SSR.getEnd())); 3280 } 3281 3282 DiagRuntimeBehavior(SL, SizeOfArg, 3283 PDiag(diag::warn_sizeof_pointer_expr_memaccess) 3284 << ReadableName 3285 << PointeeTy 3286 << DestTy 3287 << DSR 3288 << SSR); 3289 DiagRuntimeBehavior(SL, SizeOfArg, 3290 PDiag(diag::warn_sizeof_pointer_expr_memaccess_note) 3291 << ActionIdx 3292 << SSR); 3293 3294 break; 3295 } 3296 } 3297 3298 // Also check for cases where the sizeof argument is the exact same 3299 // type as the memory argument, and where it points to a user-defined 3300 // record type. 3301 if (SizeOfArgTy != QualType()) { 3302 if (PointeeTy->isRecordType() && 3303 Context.typesAreCompatible(SizeOfArgTy, DestTy)) { 3304 DiagRuntimeBehavior(LenExpr->getExprLoc(), Dest, 3305 PDiag(diag::warn_sizeof_pointer_type_memaccess) 3306 << FnName << SizeOfArgTy << ArgIdx 3307 << PointeeTy << Dest->getSourceRange() 3308 << LenExpr->getSourceRange()); 3309 break; 3310 } 3311 } 3312 3313 // Always complain about dynamic classes. 3314 if (isDynamicClassType(PointeeTy)) { 3315 3316 unsigned OperationType = 0; 3317 // "overwritten" if we're warning about the destination for any call 3318 // but memcmp; otherwise a verb appropriate to the call. 3319 if (ArgIdx != 0 || BId == Builtin::BImemcmp) { 3320 if (BId == Builtin::BImemcpy) 3321 OperationType = 1; 3322 else if(BId == Builtin::BImemmove) 3323 OperationType = 2; 3324 else if (BId == Builtin::BImemcmp) 3325 OperationType = 3; 3326 } 3327 3328 DiagRuntimeBehavior( 3329 Dest->getExprLoc(), Dest, 3330 PDiag(diag::warn_dyn_class_memaccess) 3331 << (BId == Builtin::BImemcmp ? ArgIdx + 2 : ArgIdx) 3332 << FnName << PointeeTy 3333 << OperationType 3334 << Call->getCallee()->getSourceRange()); 3335 } else if (PointeeTy.hasNonTrivialObjCLifetime() && 3336 BId != Builtin::BImemset) 3337 DiagRuntimeBehavior( 3338 Dest->getExprLoc(), Dest, 3339 PDiag(diag::warn_arc_object_memaccess) 3340 << ArgIdx << FnName << PointeeTy 3341 << Call->getCallee()->getSourceRange()); 3342 else 3343 continue; 3344 3345 DiagRuntimeBehavior( 3346 Dest->getExprLoc(), Dest, 3347 PDiag(diag::note_bad_memaccess_silence) 3348 << FixItHint::CreateInsertion(ArgRange.getBegin(), "(void*)")); 3349 break; 3350 } 3351 } 3352} 3353 3354// A little helper routine: ignore addition and subtraction of integer literals. 3355// This intentionally does not ignore all integer constant expressions because 3356// we don't want to remove sizeof(). 3357static const Expr *ignoreLiteralAdditions(const Expr *Ex, ASTContext &Ctx) { 3358 Ex = Ex->IgnoreParenCasts(); 3359 3360 for (;;) { 3361 const BinaryOperator * BO = dyn_cast<BinaryOperator>(Ex); 3362 if (!BO || !BO->isAdditiveOp()) 3363 break; 3364 3365 const Expr *RHS = BO->getRHS()->IgnoreParenCasts(); 3366 const Expr *LHS = BO->getLHS()->IgnoreParenCasts(); 3367 3368 if (isa<IntegerLiteral>(RHS)) 3369 Ex = LHS; 3370 else if (isa<IntegerLiteral>(LHS)) 3371 Ex = RHS; 3372 else 3373 break; 3374 } 3375 3376 return Ex; 3377} 3378 3379static bool isConstantSizeArrayWithMoreThanOneElement(QualType Ty, 3380 ASTContext &Context) { 3381 // Only handle constant-sized or VLAs, but not flexible members. 3382 if (const ConstantArrayType *CAT = Context.getAsConstantArrayType(Ty)) { 3383 // Only issue the FIXIT for arrays of size > 1. 3384 if (CAT->getSize().getSExtValue() <= 1) 3385 return false; 3386 } else if (!Ty->isVariableArrayType()) { 3387 return false; 3388 } 3389 return true; 3390} 3391 3392// Warn if the user has made the 'size' argument to strlcpy or strlcat 3393// be the size of the source, instead of the destination. 3394void Sema::CheckStrlcpycatArguments(const CallExpr *Call, 3395 IdentifierInfo *FnName) { 3396 3397 // Don't crash if the user has the wrong number of arguments 3398 if (Call->getNumArgs() != 3) 3399 return; 3400 3401 const Expr *SrcArg = ignoreLiteralAdditions(Call->getArg(1), Context); 3402 const Expr *SizeArg = ignoreLiteralAdditions(Call->getArg(2), Context); 3403 const Expr *CompareWithSrc = NULL; 3404 3405 // Look for 'strlcpy(dst, x, sizeof(x))' 3406 if (const Expr *Ex = getSizeOfExprArg(SizeArg)) 3407 CompareWithSrc = Ex; 3408 else { 3409 // Look for 'strlcpy(dst, x, strlen(x))' 3410 if (const CallExpr *SizeCall = dyn_cast<CallExpr>(SizeArg)) { 3411 if (SizeCall->isBuiltinCall() == Builtin::BIstrlen 3412 && SizeCall->getNumArgs() == 1) 3413 CompareWithSrc = ignoreLiteralAdditions(SizeCall->getArg(0), Context); 3414 } 3415 } 3416 3417 if (!CompareWithSrc) 3418 return; 3419 3420 // Determine if the argument to sizeof/strlen is equal to the source 3421 // argument. In principle there's all kinds of things you could do 3422 // here, for instance creating an == expression and evaluating it with 3423 // EvaluateAsBooleanCondition, but this uses a more direct technique: 3424 const DeclRefExpr *SrcArgDRE = dyn_cast<DeclRefExpr>(SrcArg); 3425 if (!SrcArgDRE) 3426 return; 3427 3428 const DeclRefExpr *CompareWithSrcDRE = dyn_cast<DeclRefExpr>(CompareWithSrc); 3429 if (!CompareWithSrcDRE || 3430 SrcArgDRE->getDecl() != CompareWithSrcDRE->getDecl()) 3431 return; 3432 3433 const Expr *OriginalSizeArg = Call->getArg(2); 3434 Diag(CompareWithSrcDRE->getLocStart(), diag::warn_strlcpycat_wrong_size) 3435 << OriginalSizeArg->getSourceRange() << FnName; 3436 3437 // Output a FIXIT hint if the destination is an array (rather than a 3438 // pointer to an array). This could be enhanced to handle some 3439 // pointers if we know the actual size, like if DstArg is 'array+2' 3440 // we could say 'sizeof(array)-2'. 3441 const Expr *DstArg = Call->getArg(0)->IgnoreParenImpCasts(); 3442 if (!isConstantSizeArrayWithMoreThanOneElement(DstArg->getType(), Context)) 3443 return; 3444 3445 SmallString<128> sizeString; 3446 llvm::raw_svector_ostream OS(sizeString); 3447 OS << "sizeof("; 3448 DstArg->printPretty(OS, 0, getPrintingPolicy()); 3449 OS << ")"; 3450 3451 Diag(OriginalSizeArg->getLocStart(), diag::note_strlcpycat_wrong_size) 3452 << FixItHint::CreateReplacement(OriginalSizeArg->getSourceRange(), 3453 OS.str()); 3454} 3455 3456/// Check if two expressions refer to the same declaration. 3457static bool referToTheSameDecl(const Expr *E1, const Expr *E2) { 3458 if (const DeclRefExpr *D1 = dyn_cast_or_null<DeclRefExpr>(E1)) 3459 if (const DeclRefExpr *D2 = dyn_cast_or_null<DeclRefExpr>(E2)) 3460 return D1->getDecl() == D2->getDecl(); 3461 return false; 3462} 3463 3464static const Expr *getStrlenExprArg(const Expr *E) { 3465 if (const CallExpr *CE = dyn_cast<CallExpr>(E)) { 3466 const FunctionDecl *FD = CE->getDirectCallee(); 3467 if (!FD || FD->getMemoryFunctionKind() != Builtin::BIstrlen) 3468 return 0; 3469 return CE->getArg(0)->IgnoreParenCasts(); 3470 } 3471 return 0; 3472} 3473 3474// Warn on anti-patterns as the 'size' argument to strncat. 3475// The correct size argument should look like following: 3476// strncat(dst, src, sizeof(dst) - strlen(dest) - 1); 3477void Sema::CheckStrncatArguments(const CallExpr *CE, 3478 IdentifierInfo *FnName) { 3479 // Don't crash if the user has the wrong number of arguments. 3480 if (CE->getNumArgs() < 3) 3481 return; 3482 const Expr *DstArg = CE->getArg(0)->IgnoreParenCasts(); 3483 const Expr *SrcArg = CE->getArg(1)->IgnoreParenCasts(); 3484 const Expr *LenArg = CE->getArg(2)->IgnoreParenCasts(); 3485 3486 // Identify common expressions, which are wrongly used as the size argument 3487 // to strncat and may lead to buffer overflows. 3488 unsigned PatternType = 0; 3489 if (const Expr *SizeOfArg = getSizeOfExprArg(LenArg)) { 3490 // - sizeof(dst) 3491 if (referToTheSameDecl(SizeOfArg, DstArg)) 3492 PatternType = 1; 3493 // - sizeof(src) 3494 else if (referToTheSameDecl(SizeOfArg, SrcArg)) 3495 PatternType = 2; 3496 } else if (const BinaryOperator *BE = dyn_cast<BinaryOperator>(LenArg)) { 3497 if (BE->getOpcode() == BO_Sub) { 3498 const Expr *L = BE->getLHS()->IgnoreParenCasts(); 3499 const Expr *R = BE->getRHS()->IgnoreParenCasts(); 3500 // - sizeof(dst) - strlen(dst) 3501 if (referToTheSameDecl(DstArg, getSizeOfExprArg(L)) && 3502 referToTheSameDecl(DstArg, getStrlenExprArg(R))) 3503 PatternType = 1; 3504 // - sizeof(src) - (anything) 3505 else if (referToTheSameDecl(SrcArg, getSizeOfExprArg(L))) 3506 PatternType = 2; 3507 } 3508 } 3509 3510 if (PatternType == 0) 3511 return; 3512 3513 // Generate the diagnostic. 3514 SourceLocation SL = LenArg->getLocStart(); 3515 SourceRange SR = LenArg->getSourceRange(); 3516 SourceManager &SM = PP.getSourceManager(); 3517 3518 // If the function is defined as a builtin macro, do not show macro expansion. 3519 if (SM.isMacroArgExpansion(SL)) { 3520 SL = SM.getSpellingLoc(SL); 3521 SR = SourceRange(SM.getSpellingLoc(SR.getBegin()), 3522 SM.getSpellingLoc(SR.getEnd())); 3523 } 3524 3525 // Check if the destination is an array (rather than a pointer to an array). 3526 QualType DstTy = DstArg->getType(); 3527 bool isKnownSizeArray = isConstantSizeArrayWithMoreThanOneElement(DstTy, 3528 Context); 3529 if (!isKnownSizeArray) { 3530 if (PatternType == 1) 3531 Diag(SL, diag::warn_strncat_wrong_size) << SR; 3532 else 3533 Diag(SL, diag::warn_strncat_src_size) << SR; 3534 return; 3535 } 3536 3537 if (PatternType == 1) 3538 Diag(SL, diag::warn_strncat_large_size) << SR; 3539 else 3540 Diag(SL, diag::warn_strncat_src_size) << SR; 3541 3542 SmallString<128> sizeString; 3543 llvm::raw_svector_ostream OS(sizeString); 3544 OS << "sizeof("; 3545 DstArg->printPretty(OS, 0, getPrintingPolicy()); 3546 OS << ") - "; 3547 OS << "strlen("; 3548 DstArg->printPretty(OS, 0, getPrintingPolicy()); 3549 OS << ") - 1"; 3550 3551 Diag(SL, diag::note_strncat_wrong_size) 3552 << FixItHint::CreateReplacement(SR, OS.str()); 3553} 3554 3555//===--- CHECK: Return Address of Stack Variable --------------------------===// 3556 3557static Expr *EvalVal(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars, 3558 Decl *ParentDecl); 3559static Expr *EvalAddr(Expr* E, SmallVectorImpl<DeclRefExpr *> &refVars, 3560 Decl *ParentDecl); 3561 3562/// CheckReturnStackAddr - Check if a return statement returns the address 3563/// of a stack variable. 3564void 3565Sema::CheckReturnStackAddr(Expr *RetValExp, QualType lhsType, 3566 SourceLocation ReturnLoc) { 3567 3568 Expr *stackE = 0; 3569 SmallVector<DeclRefExpr *, 8> refVars; 3570 3571 // Perform checking for returned stack addresses, local blocks, 3572 // label addresses or references to temporaries. 3573 if (lhsType->isPointerType() || 3574 (!getLangOpts().ObjCAutoRefCount && lhsType->isBlockPointerType())) { 3575 stackE = EvalAddr(RetValExp, refVars, /*ParentDecl=*/0); 3576 } else if (lhsType->isReferenceType()) { 3577 stackE = EvalVal(RetValExp, refVars, /*ParentDecl=*/0); 3578 } 3579 3580 if (stackE == 0) 3581 return; // Nothing suspicious was found. 3582 3583 SourceLocation diagLoc; 3584 SourceRange diagRange; 3585 if (refVars.empty()) { 3586 diagLoc = stackE->getLocStart(); 3587 diagRange = stackE->getSourceRange(); 3588 } else { 3589 // We followed through a reference variable. 'stackE' contains the 3590 // problematic expression but we will warn at the return statement pointing 3591 // at the reference variable. We will later display the "trail" of 3592 // reference variables using notes. 3593 diagLoc = refVars[0]->getLocStart(); 3594 diagRange = refVars[0]->getSourceRange(); 3595 } 3596 3597 if (DeclRefExpr *DR = dyn_cast<DeclRefExpr>(stackE)) { //address of local var. 3598 Diag(diagLoc, lhsType->isReferenceType() ? diag::warn_ret_stack_ref 3599 : diag::warn_ret_stack_addr) 3600 << DR->getDecl()->getDeclName() << diagRange; 3601 } else if (isa<BlockExpr>(stackE)) { // local block. 3602 Diag(diagLoc, diag::err_ret_local_block) << diagRange; 3603 } else if (isa<AddrLabelExpr>(stackE)) { // address of label. 3604 Diag(diagLoc, diag::warn_ret_addr_label) << diagRange; 3605 } else { // local temporary. 3606 Diag(diagLoc, lhsType->isReferenceType() ? diag::warn_ret_local_temp_ref 3607 : diag::warn_ret_local_temp_addr) 3608 << diagRange; 3609 } 3610 3611 // Display the "trail" of reference variables that we followed until we 3612 // found the problematic expression using notes. 3613 for (unsigned i = 0, e = refVars.size(); i != e; ++i) { 3614 VarDecl *VD = cast<VarDecl>(refVars[i]->getDecl()); 3615 // If this var binds to another reference var, show the range of the next 3616 // var, otherwise the var binds to the problematic expression, in which case 3617 // show the range of the expression. 3618 SourceRange range = (i < e-1) ? refVars[i+1]->getSourceRange() 3619 : stackE->getSourceRange(); 3620 Diag(VD->getLocation(), diag::note_ref_var_local_bind) 3621 << VD->getDeclName() << range; 3622 } 3623} 3624 3625/// EvalAddr - EvalAddr and EvalVal are mutually recursive functions that 3626/// check if the expression in a return statement evaluates to an address 3627/// to a location on the stack, a local block, an address of a label, or a 3628/// reference to local temporary. The recursion is used to traverse the 3629/// AST of the return expression, with recursion backtracking when we 3630/// encounter a subexpression that (1) clearly does not lead to one of the 3631/// above problematic expressions (2) is something we cannot determine leads to 3632/// a problematic expression based on such local checking. 3633/// 3634/// Both EvalAddr and EvalVal follow through reference variables to evaluate 3635/// the expression that they point to. Such variables are added to the 3636/// 'refVars' vector so that we know what the reference variable "trail" was. 3637/// 3638/// EvalAddr processes expressions that are pointers that are used as 3639/// references (and not L-values). EvalVal handles all other values. 3640/// At the base case of the recursion is a check for the above problematic 3641/// expressions. 3642/// 3643/// This implementation handles: 3644/// 3645/// * pointer-to-pointer casts 3646/// * implicit conversions from array references to pointers 3647/// * taking the address of fields 3648/// * arbitrary interplay between "&" and "*" operators 3649/// * pointer arithmetic from an address of a stack variable 3650/// * taking the address of an array element where the array is on the stack 3651static Expr *EvalAddr(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars, 3652 Decl *ParentDecl) { 3653 if (E->isTypeDependent()) 3654 return NULL; 3655 3656 // We should only be called for evaluating pointer expressions. 3657 assert((E->getType()->isAnyPointerType() || 3658 E->getType()->isBlockPointerType() || 3659 E->getType()->isObjCQualifiedIdType()) && 3660 "EvalAddr only works on pointers"); 3661 3662 E = E->IgnoreParens(); 3663 3664 // Our "symbolic interpreter" is just a dispatch off the currently 3665 // viewed AST node. We then recursively traverse the AST by calling 3666 // EvalAddr and EvalVal appropriately. 3667 switch (E->getStmtClass()) { 3668 case Stmt::DeclRefExprClass: { 3669 DeclRefExpr *DR = cast<DeclRefExpr>(E); 3670 3671 if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl())) 3672 // If this is a reference variable, follow through to the expression that 3673 // it points to. 3674 if (V->hasLocalStorage() && 3675 V->getType()->isReferenceType() && V->hasInit()) { 3676 // Add the reference variable to the "trail". 3677 refVars.push_back(DR); 3678 return EvalAddr(V->getInit(), refVars, ParentDecl); 3679 } 3680 3681 return NULL; 3682 } 3683 3684 case Stmt::UnaryOperatorClass: { 3685 // The only unary operator that make sense to handle here 3686 // is AddrOf. All others don't make sense as pointers. 3687 UnaryOperator *U = cast<UnaryOperator>(E); 3688 3689 if (U->getOpcode() == UO_AddrOf) 3690 return EvalVal(U->getSubExpr(), refVars, ParentDecl); 3691 else 3692 return NULL; 3693 } 3694 3695 case Stmt::BinaryOperatorClass: { 3696 // Handle pointer arithmetic. All other binary operators are not valid 3697 // in this context. 3698 BinaryOperator *B = cast<BinaryOperator>(E); 3699 BinaryOperatorKind op = B->getOpcode(); 3700 3701 if (op != BO_Add && op != BO_Sub) 3702 return NULL; 3703 3704 Expr *Base = B->getLHS(); 3705 3706 // Determine which argument is the real pointer base. It could be 3707 // the RHS argument instead of the LHS. 3708 if (!Base->getType()->isPointerType()) Base = B->getRHS(); 3709 3710 assert (Base->getType()->isPointerType()); 3711 return EvalAddr(Base, refVars, ParentDecl); 3712 } 3713 3714 // For conditional operators we need to see if either the LHS or RHS are 3715 // valid DeclRefExpr*s. If one of them is valid, we return it. 3716 case Stmt::ConditionalOperatorClass: { 3717 ConditionalOperator *C = cast<ConditionalOperator>(E); 3718 3719 // Handle the GNU extension for missing LHS. 3720 if (Expr *lhsExpr = C->getLHS()) { 3721 // In C++, we can have a throw-expression, which has 'void' type. 3722 if (!lhsExpr->getType()->isVoidType()) 3723 if (Expr* LHS = EvalAddr(lhsExpr, refVars, ParentDecl)) 3724 return LHS; 3725 } 3726 3727 // In C++, we can have a throw-expression, which has 'void' type. 3728 if (C->getRHS()->getType()->isVoidType()) 3729 return NULL; 3730 3731 return EvalAddr(C->getRHS(), refVars, ParentDecl); 3732 } 3733 3734 case Stmt::BlockExprClass: 3735 if (cast<BlockExpr>(E)->getBlockDecl()->hasCaptures()) 3736 return E; // local block. 3737 return NULL; 3738 3739 case Stmt::AddrLabelExprClass: 3740 return E; // address of label. 3741 3742 case Stmt::ExprWithCleanupsClass: 3743 return EvalAddr(cast<ExprWithCleanups>(E)->getSubExpr(), refVars, 3744 ParentDecl); 3745 3746 // For casts, we need to handle conversions from arrays to 3747 // pointer values, and pointer-to-pointer conversions. 3748 case Stmt::ImplicitCastExprClass: 3749 case Stmt::CStyleCastExprClass: 3750 case Stmt::CXXFunctionalCastExprClass: 3751 case Stmt::ObjCBridgedCastExprClass: 3752 case Stmt::CXXStaticCastExprClass: 3753 case Stmt::CXXDynamicCastExprClass: 3754 case Stmt::CXXConstCastExprClass: 3755 case Stmt::CXXReinterpretCastExprClass: { 3756 Expr* SubExpr = cast<CastExpr>(E)->getSubExpr(); 3757 switch (cast<CastExpr>(E)->getCastKind()) { 3758 case CK_BitCast: 3759 case CK_LValueToRValue: 3760 case CK_NoOp: 3761 case CK_BaseToDerived: 3762 case CK_DerivedToBase: 3763 case CK_UncheckedDerivedToBase: 3764 case CK_Dynamic: 3765 case CK_CPointerToObjCPointerCast: 3766 case CK_BlockPointerToObjCPointerCast: 3767 case CK_AnyPointerToBlockPointerCast: 3768 return EvalAddr(SubExpr, refVars, ParentDecl); 3769 3770 case CK_ArrayToPointerDecay: 3771 return EvalVal(SubExpr, refVars, ParentDecl); 3772 3773 default: 3774 return 0; 3775 } 3776 } 3777 3778 case Stmt::MaterializeTemporaryExprClass: 3779 if (Expr *Result = EvalAddr( 3780 cast<MaterializeTemporaryExpr>(E)->GetTemporaryExpr(), 3781 refVars, ParentDecl)) 3782 return Result; 3783 3784 return E; 3785 3786 // Everything else: we simply don't reason about them. 3787 default: 3788 return NULL; 3789 } 3790} 3791 3792 3793/// EvalVal - This function is complements EvalAddr in the mutual recursion. 3794/// See the comments for EvalAddr for more details. 3795static Expr *EvalVal(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars, 3796 Decl *ParentDecl) { 3797do { 3798 // We should only be called for evaluating non-pointer expressions, or 3799 // expressions with a pointer type that are not used as references but instead 3800 // are l-values (e.g., DeclRefExpr with a pointer type). 3801 3802 // Our "symbolic interpreter" is just a dispatch off the currently 3803 // viewed AST node. We then recursively traverse the AST by calling 3804 // EvalAddr and EvalVal appropriately. 3805 3806 E = E->IgnoreParens(); 3807 switch (E->getStmtClass()) { 3808 case Stmt::ImplicitCastExprClass: { 3809 ImplicitCastExpr *IE = cast<ImplicitCastExpr>(E); 3810 if (IE->getValueKind() == VK_LValue) { 3811 E = IE->getSubExpr(); 3812 continue; 3813 } 3814 return NULL; 3815 } 3816 3817 case Stmt::ExprWithCleanupsClass: 3818 return EvalVal(cast<ExprWithCleanups>(E)->getSubExpr(), refVars,ParentDecl); 3819 3820 case Stmt::DeclRefExprClass: { 3821 // When we hit a DeclRefExpr we are looking at code that refers to a 3822 // variable's name. If it's not a reference variable we check if it has 3823 // local storage within the function, and if so, return the expression. 3824 DeclRefExpr *DR = cast<DeclRefExpr>(E); 3825 3826 if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl())) { 3827 // Check if it refers to itself, e.g. "int& i = i;". 3828 if (V == ParentDecl) 3829 return DR; 3830 3831 if (V->hasLocalStorage()) { 3832 if (!V->getType()->isReferenceType()) 3833 return DR; 3834 3835 // Reference variable, follow through to the expression that 3836 // it points to. 3837 if (V->hasInit()) { 3838 // Add the reference variable to the "trail". 3839 refVars.push_back(DR); 3840 return EvalVal(V->getInit(), refVars, V); 3841 } 3842 } 3843 } 3844 3845 return NULL; 3846 } 3847 3848 case Stmt::UnaryOperatorClass: { 3849 // The only unary operator that make sense to handle here 3850 // is Deref. All others don't resolve to a "name." This includes 3851 // handling all sorts of rvalues passed to a unary operator. 3852 UnaryOperator *U = cast<UnaryOperator>(E); 3853 3854 if (U->getOpcode() == UO_Deref) 3855 return EvalAddr(U->getSubExpr(), refVars, ParentDecl); 3856 3857 return NULL; 3858 } 3859 3860 case Stmt::ArraySubscriptExprClass: { 3861 // Array subscripts are potential references to data on the stack. We 3862 // retrieve the DeclRefExpr* for the array variable if it indeed 3863 // has local storage. 3864 return EvalAddr(cast<ArraySubscriptExpr>(E)->getBase(), refVars,ParentDecl); 3865 } 3866 3867 case Stmt::ConditionalOperatorClass: { 3868 // For conditional operators we need to see if either the LHS or RHS are 3869 // non-NULL Expr's. If one is non-NULL, we return it. 3870 ConditionalOperator *C = cast<ConditionalOperator>(E); 3871 3872 // Handle the GNU extension for missing LHS. 3873 if (Expr *lhsExpr = C->getLHS()) 3874 if (Expr *LHS = EvalVal(lhsExpr, refVars, ParentDecl)) 3875 return LHS; 3876 3877 return EvalVal(C->getRHS(), refVars, ParentDecl); 3878 } 3879 3880 // Accesses to members are potential references to data on the stack. 3881 case Stmt::MemberExprClass: { 3882 MemberExpr *M = cast<MemberExpr>(E); 3883 3884 // Check for indirect access. We only want direct field accesses. 3885 if (M->isArrow()) 3886 return NULL; 3887 3888 // Check whether the member type is itself a reference, in which case 3889 // we're not going to refer to the member, but to what the member refers to. 3890 if (M->getMemberDecl()->getType()->isReferenceType()) 3891 return NULL; 3892 3893 return EvalVal(M->getBase(), refVars, ParentDecl); 3894 } 3895 3896 case Stmt::MaterializeTemporaryExprClass: 3897 if (Expr *Result = EvalVal( 3898 cast<MaterializeTemporaryExpr>(E)->GetTemporaryExpr(), 3899 refVars, ParentDecl)) 3900 return Result; 3901 3902 return E; 3903 3904 default: 3905 // Check that we don't return or take the address of a reference to a 3906 // temporary. This is only useful in C++. 3907 if (!E->isTypeDependent() && E->isRValue()) 3908 return E; 3909 3910 // Everything else: we simply don't reason about them. 3911 return NULL; 3912 } 3913} while (true); 3914} 3915 3916//===--- CHECK: Floating-Point comparisons (-Wfloat-equal) ---------------===// 3917 3918/// Check for comparisons of floating point operands using != and ==. 3919/// Issue a warning if these are no self-comparisons, as they are not likely 3920/// to do what the programmer intended. 3921void Sema::CheckFloatComparison(SourceLocation Loc, Expr* LHS, Expr *RHS) { 3922 Expr* LeftExprSansParen = LHS->IgnoreParenImpCasts(); 3923 Expr* RightExprSansParen = RHS->IgnoreParenImpCasts(); 3924 3925 // Special case: check for x == x (which is OK). 3926 // Do not emit warnings for such cases. 3927 if (DeclRefExpr* DRL = dyn_cast<DeclRefExpr>(LeftExprSansParen)) 3928 if (DeclRefExpr* DRR = dyn_cast<DeclRefExpr>(RightExprSansParen)) 3929 if (DRL->getDecl() == DRR->getDecl()) 3930 return; 3931 3932 3933 // Special case: check for comparisons against literals that can be exactly 3934 // represented by APFloat. In such cases, do not emit a warning. This 3935 // is a heuristic: often comparison against such literals are used to 3936 // detect if a value in a variable has not changed. This clearly can 3937 // lead to false negatives. 3938 if (FloatingLiteral* FLL = dyn_cast<FloatingLiteral>(LeftExprSansParen)) { 3939 if (FLL->isExact()) 3940 return; 3941 } else 3942 if (FloatingLiteral* FLR = dyn_cast<FloatingLiteral>(RightExprSansParen)) 3943 if (FLR->isExact()) 3944 return; 3945 3946 // Check for comparisons with builtin types. 3947 if (CallExpr* CL = dyn_cast<CallExpr>(LeftExprSansParen)) 3948 if (CL->isBuiltinCall()) 3949 return; 3950 3951 if (CallExpr* CR = dyn_cast<CallExpr>(RightExprSansParen)) 3952 if (CR->isBuiltinCall()) 3953 return; 3954 3955 // Emit the diagnostic. 3956 Diag(Loc, diag::warn_floatingpoint_eq) 3957 << LHS->getSourceRange() << RHS->getSourceRange(); 3958} 3959 3960//===--- CHECK: Integer mixed-sign comparisons (-Wsign-compare) --------===// 3961//===--- CHECK: Lossy implicit conversions (-Wconversion) --------------===// 3962 3963namespace { 3964 3965/// Structure recording the 'active' range of an integer-valued 3966/// expression. 3967struct IntRange { 3968 /// The number of bits active in the int. 3969 unsigned Width; 3970 3971 /// True if the int is known not to have negative values. 3972 bool NonNegative; 3973 3974 IntRange(unsigned Width, bool NonNegative) 3975 : Width(Width), NonNegative(NonNegative) 3976 {} 3977 3978 /// Returns the range of the bool type. 3979 static IntRange forBoolType() { 3980 return IntRange(1, true); 3981 } 3982 3983 /// Returns the range of an opaque value of the given integral type. 3984 static IntRange forValueOfType(ASTContext &C, QualType T) { 3985 return forValueOfCanonicalType(C, 3986 T->getCanonicalTypeInternal().getTypePtr()); 3987 } 3988 3989 /// Returns the range of an opaque value of a canonical integral type. 3990 static IntRange forValueOfCanonicalType(ASTContext &C, const Type *T) { 3991 assert(T->isCanonicalUnqualified()); 3992 3993 if (const VectorType *VT = dyn_cast<VectorType>(T)) 3994 T = VT->getElementType().getTypePtr(); 3995 if (const ComplexType *CT = dyn_cast<ComplexType>(T)) 3996 T = CT->getElementType().getTypePtr(); 3997 3998 // For enum types, use the known bit width of the enumerators. 3999 if (const EnumType *ET = dyn_cast<EnumType>(T)) { 4000 EnumDecl *Enum = ET->getDecl(); 4001 if (!Enum->isCompleteDefinition()) 4002 return IntRange(C.getIntWidth(QualType(T, 0)), false); 4003 4004 unsigned NumPositive = Enum->getNumPositiveBits(); 4005 unsigned NumNegative = Enum->getNumNegativeBits(); 4006 4007 if (NumNegative == 0) 4008 return IntRange(NumPositive, true/*NonNegative*/); 4009 else 4010 return IntRange(std::max(NumPositive + 1, NumNegative), 4011 false/*NonNegative*/); 4012 } 4013 4014 const BuiltinType *BT = cast<BuiltinType>(T); 4015 assert(BT->isInteger()); 4016 4017 return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger()); 4018 } 4019 4020 /// Returns the "target" range of a canonical integral type, i.e. 4021 /// the range of values expressible in the type. 4022 /// 4023 /// This matches forValueOfCanonicalType except that enums have the 4024 /// full range of their type, not the range of their enumerators. 4025 static IntRange forTargetOfCanonicalType(ASTContext &C, const Type *T) { 4026 assert(T->isCanonicalUnqualified()); 4027 4028 if (const VectorType *VT = dyn_cast<VectorType>(T)) 4029 T = VT->getElementType().getTypePtr(); 4030 if (const ComplexType *CT = dyn_cast<ComplexType>(T)) 4031 T = CT->getElementType().getTypePtr(); 4032 if (const EnumType *ET = dyn_cast<EnumType>(T)) 4033 T = C.getCanonicalType(ET->getDecl()->getIntegerType()).getTypePtr(); 4034 4035 const BuiltinType *BT = cast<BuiltinType>(T); 4036 assert(BT->isInteger()); 4037 4038 return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger()); 4039 } 4040 4041 /// Returns the supremum of two ranges: i.e. their conservative merge. 4042 static IntRange join(IntRange L, IntRange R) { 4043 return IntRange(std::max(L.Width, R.Width), 4044 L.NonNegative && R.NonNegative); 4045 } 4046 4047 /// Returns the infinum of two ranges: i.e. their aggressive merge. 4048 static IntRange meet(IntRange L, IntRange R) { 4049 return IntRange(std::min(L.Width, R.Width), 4050 L.NonNegative || R.NonNegative); 4051 } 4052}; 4053 4054static IntRange GetValueRange(ASTContext &C, llvm::APSInt &value, 4055 unsigned MaxWidth) { 4056 if (value.isSigned() && value.isNegative()) 4057 return IntRange(value.getMinSignedBits(), false); 4058 4059 if (value.getBitWidth() > MaxWidth) 4060 value = value.trunc(MaxWidth); 4061 4062 // isNonNegative() just checks the sign bit without considering 4063 // signedness. 4064 return IntRange(value.getActiveBits(), true); 4065} 4066 4067static IntRange GetValueRange(ASTContext &C, APValue &result, QualType Ty, 4068 unsigned MaxWidth) { 4069 if (result.isInt()) 4070 return GetValueRange(C, result.getInt(), MaxWidth); 4071 4072 if (result.isVector()) { 4073 IntRange R = GetValueRange(C, result.getVectorElt(0), Ty, MaxWidth); 4074 for (unsigned i = 1, e = result.getVectorLength(); i != e; ++i) { 4075 IntRange El = GetValueRange(C, result.getVectorElt(i), Ty, MaxWidth); 4076 R = IntRange::join(R, El); 4077 } 4078 return R; 4079 } 4080 4081 if (result.isComplexInt()) { 4082 IntRange R = GetValueRange(C, result.getComplexIntReal(), MaxWidth); 4083 IntRange I = GetValueRange(C, result.getComplexIntImag(), MaxWidth); 4084 return IntRange::join(R, I); 4085 } 4086 4087 // This can happen with lossless casts to intptr_t of "based" lvalues. 4088 // Assume it might use arbitrary bits. 4089 // FIXME: The only reason we need to pass the type in here is to get 4090 // the sign right on this one case. It would be nice if APValue 4091 // preserved this. 4092 assert(result.isLValue() || result.isAddrLabelDiff()); 4093 return IntRange(MaxWidth, Ty->isUnsignedIntegerOrEnumerationType()); 4094} 4095 4096/// Pseudo-evaluate the given integer expression, estimating the 4097/// range of values it might take. 4098/// 4099/// \param MaxWidth - the width to which the value will be truncated 4100static IntRange GetExprRange(ASTContext &C, Expr *E, unsigned MaxWidth) { 4101 E = E->IgnoreParens(); 4102 4103 // Try a full evaluation first. 4104 Expr::EvalResult result; 4105 if (E->EvaluateAsRValue(result, C)) 4106 return GetValueRange(C, result.Val, E->getType(), MaxWidth); 4107 4108 // I think we only want to look through implicit casts here; if the 4109 // user has an explicit widening cast, we should treat the value as 4110 // being of the new, wider type. 4111 if (ImplicitCastExpr *CE = dyn_cast<ImplicitCastExpr>(E)) { 4112 if (CE->getCastKind() == CK_NoOp || CE->getCastKind() == CK_LValueToRValue) 4113 return GetExprRange(C, CE->getSubExpr(), MaxWidth); 4114 4115 IntRange OutputTypeRange = IntRange::forValueOfType(C, CE->getType()); 4116 4117 bool isIntegerCast = (CE->getCastKind() == CK_IntegralCast); 4118 4119 // Assume that non-integer casts can span the full range of the type. 4120 if (!isIntegerCast) 4121 return OutputTypeRange; 4122 4123 IntRange SubRange 4124 = GetExprRange(C, CE->getSubExpr(), 4125 std::min(MaxWidth, OutputTypeRange.Width)); 4126 4127 // Bail out if the subexpr's range is as wide as the cast type. 4128 if (SubRange.Width >= OutputTypeRange.Width) 4129 return OutputTypeRange; 4130 4131 // Otherwise, we take the smaller width, and we're non-negative if 4132 // either the output type or the subexpr is. 4133 return IntRange(SubRange.Width, 4134 SubRange.NonNegative || OutputTypeRange.NonNegative); 4135 } 4136 4137 if (ConditionalOperator *CO = dyn_cast<ConditionalOperator>(E)) { 4138 // If we can fold the condition, just take that operand. 4139 bool CondResult; 4140 if (CO->getCond()->EvaluateAsBooleanCondition(CondResult, C)) 4141 return GetExprRange(C, CondResult ? CO->getTrueExpr() 4142 : CO->getFalseExpr(), 4143 MaxWidth); 4144 4145 // Otherwise, conservatively merge. 4146 IntRange L = GetExprRange(C, CO->getTrueExpr(), MaxWidth); 4147 IntRange R = GetExprRange(C, CO->getFalseExpr(), MaxWidth); 4148 return IntRange::join(L, R); 4149 } 4150 4151 if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) { 4152 switch (BO->getOpcode()) { 4153 4154 // Boolean-valued operations are single-bit and positive. 4155 case BO_LAnd: 4156 case BO_LOr: 4157 case BO_LT: 4158 case BO_GT: 4159 case BO_LE: 4160 case BO_GE: 4161 case BO_EQ: 4162 case BO_NE: 4163 return IntRange::forBoolType(); 4164 4165 // The type of the assignments is the type of the LHS, so the RHS 4166 // is not necessarily the same type. 4167 case BO_MulAssign: 4168 case BO_DivAssign: 4169 case BO_RemAssign: 4170 case BO_AddAssign: 4171 case BO_SubAssign: 4172 case BO_XorAssign: 4173 case BO_OrAssign: 4174 // TODO: bitfields? 4175 return IntRange::forValueOfType(C, E->getType()); 4176 4177 // Simple assignments just pass through the RHS, which will have 4178 // been coerced to the LHS type. 4179 case BO_Assign: 4180 // TODO: bitfields? 4181 return GetExprRange(C, BO->getRHS(), MaxWidth); 4182 4183 // Operations with opaque sources are black-listed. 4184 case BO_PtrMemD: 4185 case BO_PtrMemI: 4186 return IntRange::forValueOfType(C, E->getType()); 4187 4188 // Bitwise-and uses the *infinum* of the two source ranges. 4189 case BO_And: 4190 case BO_AndAssign: 4191 return IntRange::meet(GetExprRange(C, BO->getLHS(), MaxWidth), 4192 GetExprRange(C, BO->getRHS(), MaxWidth)); 4193 4194 // Left shift gets black-listed based on a judgement call. 4195 case BO_Shl: 4196 // ...except that we want to treat '1 << (blah)' as logically 4197 // positive. It's an important idiom. 4198 if (IntegerLiteral *I 4199 = dyn_cast<IntegerLiteral>(BO->getLHS()->IgnoreParenCasts())) { 4200 if (I->getValue() == 1) { 4201 IntRange R = IntRange::forValueOfType(C, E->getType()); 4202 return IntRange(R.Width, /*NonNegative*/ true); 4203 } 4204 } 4205 // fallthrough 4206 4207 case BO_ShlAssign: 4208 return IntRange::forValueOfType(C, E->getType()); 4209 4210 // Right shift by a constant can narrow its left argument. 4211 case BO_Shr: 4212 case BO_ShrAssign: { 4213 IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth); 4214 4215 // If the shift amount is a positive constant, drop the width by 4216 // that much. 4217 llvm::APSInt shift; 4218 if (BO->getRHS()->isIntegerConstantExpr(shift, C) && 4219 shift.isNonNegative()) { 4220 unsigned zext = shift.getZExtValue(); 4221 if (zext >= L.Width) 4222 L.Width = (L.NonNegative ? 0 : 1); 4223 else 4224 L.Width -= zext; 4225 } 4226 4227 return L; 4228 } 4229 4230 // Comma acts as its right operand. 4231 case BO_Comma: 4232 return GetExprRange(C, BO->getRHS(), MaxWidth); 4233 4234 // Black-list pointer subtractions. 4235 case BO_Sub: 4236 if (BO->getLHS()->getType()->isPointerType()) 4237 return IntRange::forValueOfType(C, E->getType()); 4238 break; 4239 4240 // The width of a division result is mostly determined by the size 4241 // of the LHS. 4242 case BO_Div: { 4243 // Don't 'pre-truncate' the operands. 4244 unsigned opWidth = C.getIntWidth(E->getType()); 4245 IntRange L = GetExprRange(C, BO->getLHS(), opWidth); 4246 4247 // If the divisor is constant, use that. 4248 llvm::APSInt divisor; 4249 if (BO->getRHS()->isIntegerConstantExpr(divisor, C)) { 4250 unsigned log2 = divisor.logBase2(); // floor(log_2(divisor)) 4251 if (log2 >= L.Width) 4252 L.Width = (L.NonNegative ? 0 : 1); 4253 else 4254 L.Width = std::min(L.Width - log2, MaxWidth); 4255 return L; 4256 } 4257 4258 // Otherwise, just use the LHS's width. 4259 IntRange R = GetExprRange(C, BO->getRHS(), opWidth); 4260 return IntRange(L.Width, L.NonNegative && R.NonNegative); 4261 } 4262 4263 // The result of a remainder can't be larger than the result of 4264 // either side. 4265 case BO_Rem: { 4266 // Don't 'pre-truncate' the operands. 4267 unsigned opWidth = C.getIntWidth(E->getType()); 4268 IntRange L = GetExprRange(C, BO->getLHS(), opWidth); 4269 IntRange R = GetExprRange(C, BO->getRHS(), opWidth); 4270 4271 IntRange meet = IntRange::meet(L, R); 4272 meet.Width = std::min(meet.Width, MaxWidth); 4273 return meet; 4274 } 4275 4276 // The default behavior is okay for these. 4277 case BO_Mul: 4278 case BO_Add: 4279 case BO_Xor: 4280 case BO_Or: 4281 break; 4282 } 4283 4284 // The default case is to treat the operation as if it were closed 4285 // on the narrowest type that encompasses both operands. 4286 IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth); 4287 IntRange R = GetExprRange(C, BO->getRHS(), MaxWidth); 4288 return IntRange::join(L, R); 4289 } 4290 4291 if (UnaryOperator *UO = dyn_cast<UnaryOperator>(E)) { 4292 switch (UO->getOpcode()) { 4293 // Boolean-valued operations are white-listed. 4294 case UO_LNot: 4295 return IntRange::forBoolType(); 4296 4297 // Operations with opaque sources are black-listed. 4298 case UO_Deref: 4299 case UO_AddrOf: // should be impossible 4300 return IntRange::forValueOfType(C, E->getType()); 4301 4302 default: 4303 return GetExprRange(C, UO->getSubExpr(), MaxWidth); 4304 } 4305 } 4306 4307 if (dyn_cast<OffsetOfExpr>(E)) { 4308 IntRange::forValueOfType(C, E->getType()); 4309 } 4310 4311 if (FieldDecl *BitField = E->getBitField()) 4312 return IntRange(BitField->getBitWidthValue(C), 4313 BitField->getType()->isUnsignedIntegerOrEnumerationType()); 4314 4315 return IntRange::forValueOfType(C, E->getType()); 4316} 4317 4318static IntRange GetExprRange(ASTContext &C, Expr *E) { 4319 return GetExprRange(C, E, C.getIntWidth(E->getType())); 4320} 4321 4322/// Checks whether the given value, which currently has the given 4323/// source semantics, has the same value when coerced through the 4324/// target semantics. 4325static bool IsSameFloatAfterCast(const llvm::APFloat &value, 4326 const llvm::fltSemantics &Src, 4327 const llvm::fltSemantics &Tgt) { 4328 llvm::APFloat truncated = value; 4329 4330 bool ignored; 4331 truncated.convert(Src, llvm::APFloat::rmNearestTiesToEven, &ignored); 4332 truncated.convert(Tgt, llvm::APFloat::rmNearestTiesToEven, &ignored); 4333 4334 return truncated.bitwiseIsEqual(value); 4335} 4336 4337/// Checks whether the given value, which currently has the given 4338/// source semantics, has the same value when coerced through the 4339/// target semantics. 4340/// 4341/// The value might be a vector of floats (or a complex number). 4342static bool IsSameFloatAfterCast(const APValue &value, 4343 const llvm::fltSemantics &Src, 4344 const llvm::fltSemantics &Tgt) { 4345 if (value.isFloat()) 4346 return IsSameFloatAfterCast(value.getFloat(), Src, Tgt); 4347 4348 if (value.isVector()) { 4349 for (unsigned i = 0, e = value.getVectorLength(); i != e; ++i) 4350 if (!IsSameFloatAfterCast(value.getVectorElt(i), Src, Tgt)) 4351 return false; 4352 return true; 4353 } 4354 4355 assert(value.isComplexFloat()); 4356 return (IsSameFloatAfterCast(value.getComplexFloatReal(), Src, Tgt) && 4357 IsSameFloatAfterCast(value.getComplexFloatImag(), Src, Tgt)); 4358} 4359 4360static void AnalyzeImplicitConversions(Sema &S, Expr *E, SourceLocation CC); 4361 4362static bool IsZero(Sema &S, Expr *E) { 4363 // Suppress cases where we are comparing against an enum constant. 4364 if (const DeclRefExpr *DR = 4365 dyn_cast<DeclRefExpr>(E->IgnoreParenImpCasts())) 4366 if (isa<EnumConstantDecl>(DR->getDecl())) 4367 return false; 4368 4369 // Suppress cases where the '0' value is expanded from a macro. 4370 if (E->getLocStart().isMacroID()) 4371 return false; 4372 4373 llvm::APSInt Value; 4374 return E->isIntegerConstantExpr(Value, S.Context) && Value == 0; 4375} 4376 4377static bool HasEnumType(Expr *E) { 4378 // Strip off implicit integral promotions. 4379 while (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(E)) { 4380 if (ICE->getCastKind() != CK_IntegralCast && 4381 ICE->getCastKind() != CK_NoOp) 4382 break; 4383 E = ICE->getSubExpr(); 4384 } 4385 4386 return E->getType()->isEnumeralType(); 4387} 4388 4389static void CheckTrivialUnsignedComparison(Sema &S, BinaryOperator *E) { 4390 BinaryOperatorKind op = E->getOpcode(); 4391 if (E->isValueDependent()) 4392 return; 4393 4394 if (op == BO_LT && IsZero(S, E->getRHS())) { 4395 S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison) 4396 << "< 0" << "false" << HasEnumType(E->getLHS()) 4397 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 4398 } else if (op == BO_GE && IsZero(S, E->getRHS())) { 4399 S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison) 4400 << ">= 0" << "true" << HasEnumType(E->getLHS()) 4401 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 4402 } else if (op == BO_GT && IsZero(S, E->getLHS())) { 4403 S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison) 4404 << "0 >" << "false" << HasEnumType(E->getRHS()) 4405 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 4406 } else if (op == BO_LE && IsZero(S, E->getLHS())) { 4407 S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison) 4408 << "0 <=" << "true" << HasEnumType(E->getRHS()) 4409 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 4410 } 4411} 4412 4413static void DiagnoseOutOfRangeComparison(Sema &S, BinaryOperator *E, 4414 Expr *Constant, Expr *Other, 4415 llvm::APSInt Value, 4416 bool RhsConstant) { 4417 // 0 values are handled later by CheckTrivialUnsignedComparison(). 4418 if (Value == 0) 4419 return; 4420 4421 BinaryOperatorKind op = E->getOpcode(); 4422 QualType OtherT = Other->getType(); 4423 QualType ConstantT = Constant->getType(); 4424 QualType CommonT = E->getLHS()->getType(); 4425 if (S.Context.hasSameUnqualifiedType(OtherT, ConstantT)) 4426 return; 4427 assert((OtherT->isIntegerType() && ConstantT->isIntegerType()) 4428 && "comparison with non-integer type"); 4429 4430 bool ConstantSigned = ConstantT->isSignedIntegerType(); 4431 bool CommonSigned = CommonT->isSignedIntegerType(); 4432 4433 bool EqualityOnly = false; 4434 4435 // TODO: Investigate using GetExprRange() to get tighter bounds on 4436 // on the bit ranges. 4437 IntRange OtherRange = IntRange::forValueOfType(S.Context, OtherT); 4438 unsigned OtherWidth = OtherRange.Width; 4439 4440 if (CommonSigned) { 4441 // The common type is signed, therefore no signed to unsigned conversion. 4442 if (!OtherRange.NonNegative) { 4443 // Check that the constant is representable in type OtherT. 4444 if (ConstantSigned) { 4445 if (OtherWidth >= Value.getMinSignedBits()) 4446 return; 4447 } else { // !ConstantSigned 4448 if (OtherWidth >= Value.getActiveBits() + 1) 4449 return; 4450 } 4451 } else { // !OtherSigned 4452 // Check that the constant is representable in type OtherT. 4453 // Negative values are out of range. 4454 if (ConstantSigned) { 4455 if (Value.isNonNegative() && OtherWidth >= Value.getActiveBits()) 4456 return; 4457 } else { // !ConstantSigned 4458 if (OtherWidth >= Value.getActiveBits()) 4459 return; 4460 } 4461 } 4462 } else { // !CommonSigned 4463 if (OtherRange.NonNegative) { 4464 if (OtherWidth >= Value.getActiveBits()) 4465 return; 4466 } else if (!OtherRange.NonNegative && !ConstantSigned) { 4467 // Check to see if the constant is representable in OtherT. 4468 if (OtherWidth > Value.getActiveBits()) 4469 return; 4470 // Check to see if the constant is equivalent to a negative value 4471 // cast to CommonT. 4472 if (S.Context.getIntWidth(ConstantT) == S.Context.getIntWidth(CommonT) && 4473 Value.isNegative() && Value.getMinSignedBits() <= OtherWidth) 4474 return; 4475 // The constant value rests between values that OtherT can represent after 4476 // conversion. Relational comparison still works, but equality 4477 // comparisons will be tautological. 4478 EqualityOnly = true; 4479 } else { // OtherSigned && ConstantSigned 4480 assert(0 && "Two signed types converted to unsigned types."); 4481 } 4482 } 4483 4484 bool PositiveConstant = !ConstantSigned || Value.isNonNegative(); 4485 4486 bool IsTrue = true; 4487 if (op == BO_EQ || op == BO_NE) { 4488 IsTrue = op == BO_NE; 4489 } else if (EqualityOnly) { 4490 return; 4491 } else if (RhsConstant) { 4492 if (op == BO_GT || op == BO_GE) 4493 IsTrue = !PositiveConstant; 4494 else // op == BO_LT || op == BO_LE 4495 IsTrue = PositiveConstant; 4496 } else { 4497 if (op == BO_LT || op == BO_LE) 4498 IsTrue = !PositiveConstant; 4499 else // op == BO_GT || op == BO_GE 4500 IsTrue = PositiveConstant; 4501 } 4502 4503 // If this is a comparison to an enum constant, include that 4504 // constant in the diagnostic. 4505 const EnumConstantDecl *ED = 0; 4506 if (const DeclRefExpr *DR = dyn_cast<DeclRefExpr>(Constant)) 4507 ED = dyn_cast<EnumConstantDecl>(DR->getDecl()); 4508 4509 SmallString<64> PrettySourceValue; 4510 llvm::raw_svector_ostream OS(PrettySourceValue); 4511 if (ED) 4512 OS << '\'' << *ED << "' (" << Value << ")"; 4513 else 4514 OS << Value; 4515 4516 S.Diag(E->getOperatorLoc(), diag::warn_out_of_range_compare) 4517 << OS.str() << OtherT << IsTrue 4518 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 4519} 4520 4521/// Analyze the operands of the given comparison. Implements the 4522/// fallback case from AnalyzeComparison. 4523static void AnalyzeImpConvsInComparison(Sema &S, BinaryOperator *E) { 4524 AnalyzeImplicitConversions(S, E->getLHS(), E->getOperatorLoc()); 4525 AnalyzeImplicitConversions(S, E->getRHS(), E->getOperatorLoc()); 4526} 4527 4528/// \brief Implements -Wsign-compare. 4529/// 4530/// \param E the binary operator to check for warnings 4531static void AnalyzeComparison(Sema &S, BinaryOperator *E) { 4532 // The type the comparison is being performed in. 4533 QualType T = E->getLHS()->getType(); 4534 assert(S.Context.hasSameUnqualifiedType(T, E->getRHS()->getType()) 4535 && "comparison with mismatched types"); 4536 if (E->isValueDependent()) 4537 return AnalyzeImpConvsInComparison(S, E); 4538 4539 Expr *LHS = E->getLHS()->IgnoreParenImpCasts(); 4540 Expr *RHS = E->getRHS()->IgnoreParenImpCasts(); 4541 4542 bool IsComparisonConstant = false; 4543 4544 // Check whether an integer constant comparison results in a value 4545 // of 'true' or 'false'. 4546 if (T->isIntegralType(S.Context)) { 4547 llvm::APSInt RHSValue; 4548 bool IsRHSIntegralLiteral = 4549 RHS->isIntegerConstantExpr(RHSValue, S.Context); 4550 llvm::APSInt LHSValue; 4551 bool IsLHSIntegralLiteral = 4552 LHS->isIntegerConstantExpr(LHSValue, S.Context); 4553 if (IsRHSIntegralLiteral && !IsLHSIntegralLiteral) 4554 DiagnoseOutOfRangeComparison(S, E, RHS, LHS, RHSValue, true); 4555 else if (!IsRHSIntegralLiteral && IsLHSIntegralLiteral) 4556 DiagnoseOutOfRangeComparison(S, E, LHS, RHS, LHSValue, false); 4557 else 4558 IsComparisonConstant = 4559 (IsRHSIntegralLiteral && IsLHSIntegralLiteral); 4560 } else if (!T->hasUnsignedIntegerRepresentation()) 4561 IsComparisonConstant = E->isIntegerConstantExpr(S.Context); 4562 4563 // We don't do anything special if this isn't an unsigned integral 4564 // comparison: we're only interested in integral comparisons, and 4565 // signed comparisons only happen in cases we don't care to warn about. 4566 // 4567 // We also don't care about value-dependent expressions or expressions 4568 // whose result is a constant. 4569 if (!T->hasUnsignedIntegerRepresentation() || IsComparisonConstant) 4570 return AnalyzeImpConvsInComparison(S, E); 4571 4572 // Check to see if one of the (unmodified) operands is of different 4573 // signedness. 4574 Expr *signedOperand, *unsignedOperand; 4575 if (LHS->getType()->hasSignedIntegerRepresentation()) { 4576 assert(!RHS->getType()->hasSignedIntegerRepresentation() && 4577 "unsigned comparison between two signed integer expressions?"); 4578 signedOperand = LHS; 4579 unsignedOperand = RHS; 4580 } else if (RHS->getType()->hasSignedIntegerRepresentation()) { 4581 signedOperand = RHS; 4582 unsignedOperand = LHS; 4583 } else { 4584 CheckTrivialUnsignedComparison(S, E); 4585 return AnalyzeImpConvsInComparison(S, E); 4586 } 4587 4588 // Otherwise, calculate the effective range of the signed operand. 4589 IntRange signedRange = GetExprRange(S.Context, signedOperand); 4590 4591 // Go ahead and analyze implicit conversions in the operands. Note 4592 // that we skip the implicit conversions on both sides. 4593 AnalyzeImplicitConversions(S, LHS, E->getOperatorLoc()); 4594 AnalyzeImplicitConversions(S, RHS, E->getOperatorLoc()); 4595 4596 // If the signed range is non-negative, -Wsign-compare won't fire, 4597 // but we should still check for comparisons which are always true 4598 // or false. 4599 if (signedRange.NonNegative) 4600 return CheckTrivialUnsignedComparison(S, E); 4601 4602 // For (in)equality comparisons, if the unsigned operand is a 4603 // constant which cannot collide with a overflowed signed operand, 4604 // then reinterpreting the signed operand as unsigned will not 4605 // change the result of the comparison. 4606 if (E->isEqualityOp()) { 4607 unsigned comparisonWidth = S.Context.getIntWidth(T); 4608 IntRange unsignedRange = GetExprRange(S.Context, unsignedOperand); 4609 4610 // We should never be unable to prove that the unsigned operand is 4611 // non-negative. 4612 assert(unsignedRange.NonNegative && "unsigned range includes negative?"); 4613 4614 if (unsignedRange.Width < comparisonWidth) 4615 return; 4616 } 4617 4618 S.DiagRuntimeBehavior(E->getOperatorLoc(), E, 4619 S.PDiag(diag::warn_mixed_sign_comparison) 4620 << LHS->getType() << RHS->getType() 4621 << LHS->getSourceRange() << RHS->getSourceRange()); 4622} 4623 4624/// Analyzes an attempt to assign the given value to a bitfield. 4625/// 4626/// Returns true if there was something fishy about the attempt. 4627static bool AnalyzeBitFieldAssignment(Sema &S, FieldDecl *Bitfield, Expr *Init, 4628 SourceLocation InitLoc) { 4629 assert(Bitfield->isBitField()); 4630 if (Bitfield->isInvalidDecl()) 4631 return false; 4632 4633 // White-list bool bitfields. 4634 if (Bitfield->getType()->isBooleanType()) 4635 return false; 4636 4637 // Ignore value- or type-dependent expressions. 4638 if (Bitfield->getBitWidth()->isValueDependent() || 4639 Bitfield->getBitWidth()->isTypeDependent() || 4640 Init->isValueDependent() || 4641 Init->isTypeDependent()) 4642 return false; 4643 4644 Expr *OriginalInit = Init->IgnoreParenImpCasts(); 4645 4646 llvm::APSInt Value; 4647 if (!OriginalInit->EvaluateAsInt(Value, S.Context, Expr::SE_AllowSideEffects)) 4648 return false; 4649 4650 unsigned OriginalWidth = Value.getBitWidth(); 4651 unsigned FieldWidth = Bitfield->getBitWidthValue(S.Context); 4652 4653 if (OriginalWidth <= FieldWidth) 4654 return false; 4655 4656 // Compute the value which the bitfield will contain. 4657 llvm::APSInt TruncatedValue = Value.trunc(FieldWidth); 4658 TruncatedValue.setIsSigned(Bitfield->getType()->isSignedIntegerType()); 4659 4660 // Check whether the stored value is equal to the original value. 4661 TruncatedValue = TruncatedValue.extend(OriginalWidth); 4662 if (llvm::APSInt::isSameValue(Value, TruncatedValue)) 4663 return false; 4664 4665 // Special-case bitfields of width 1: booleans are naturally 0/1, and 4666 // therefore don't strictly fit into a signed bitfield of width 1. 4667 if (FieldWidth == 1 && Value == 1) 4668 return false; 4669 4670 std::string PrettyValue = Value.toString(10); 4671 std::string PrettyTrunc = TruncatedValue.toString(10); 4672 4673 S.Diag(InitLoc, diag::warn_impcast_bitfield_precision_constant) 4674 << PrettyValue << PrettyTrunc << OriginalInit->getType() 4675 << Init->getSourceRange(); 4676 4677 return true; 4678} 4679 4680/// Analyze the given simple or compound assignment for warning-worthy 4681/// operations. 4682static void AnalyzeAssignment(Sema &S, BinaryOperator *E) { 4683 // Just recurse on the LHS. 4684 AnalyzeImplicitConversions(S, E->getLHS(), E->getOperatorLoc()); 4685 4686 // We want to recurse on the RHS as normal unless we're assigning to 4687 // a bitfield. 4688 if (FieldDecl *Bitfield = E->getLHS()->getBitField()) { 4689 if (AnalyzeBitFieldAssignment(S, Bitfield, E->getRHS(), 4690 E->getOperatorLoc())) { 4691 // Recurse, ignoring any implicit conversions on the RHS. 4692 return AnalyzeImplicitConversions(S, E->getRHS()->IgnoreParenImpCasts(), 4693 E->getOperatorLoc()); 4694 } 4695 } 4696 4697 AnalyzeImplicitConversions(S, E->getRHS(), E->getOperatorLoc()); 4698} 4699 4700/// Diagnose an implicit cast; purely a helper for CheckImplicitConversion. 4701static void DiagnoseImpCast(Sema &S, Expr *E, QualType SourceType, QualType T, 4702 SourceLocation CContext, unsigned diag, 4703 bool pruneControlFlow = false) { 4704 if (pruneControlFlow) { 4705 S.DiagRuntimeBehavior(E->getExprLoc(), E, 4706 S.PDiag(diag) 4707 << SourceType << T << E->getSourceRange() 4708 << SourceRange(CContext)); 4709 return; 4710 } 4711 S.Diag(E->getExprLoc(), diag) 4712 << SourceType << T << E->getSourceRange() << SourceRange(CContext); 4713} 4714 4715/// Diagnose an implicit cast; purely a helper for CheckImplicitConversion. 4716static void DiagnoseImpCast(Sema &S, Expr *E, QualType T, 4717 SourceLocation CContext, unsigned diag, 4718 bool pruneControlFlow = false) { 4719 DiagnoseImpCast(S, E, E->getType(), T, CContext, diag, pruneControlFlow); 4720} 4721 4722/// Diagnose an implicit cast from a literal expression. Does not warn when the 4723/// cast wouldn't lose information. 4724void DiagnoseFloatingLiteralImpCast(Sema &S, FloatingLiteral *FL, QualType T, 4725 SourceLocation CContext) { 4726 // Try to convert the literal exactly to an integer. If we can, don't warn. 4727 bool isExact = false; 4728 const llvm::APFloat &Value = FL->getValue(); 4729 llvm::APSInt IntegerValue(S.Context.getIntWidth(T), 4730 T->hasUnsignedIntegerRepresentation()); 4731 if (Value.convertToInteger(IntegerValue, 4732 llvm::APFloat::rmTowardZero, &isExact) 4733 == llvm::APFloat::opOK && isExact) 4734 return; 4735 4736 SmallString<16> PrettySourceValue; 4737 Value.toString(PrettySourceValue); 4738 SmallString<16> PrettyTargetValue; 4739 if (T->isSpecificBuiltinType(BuiltinType::Bool)) 4740 PrettyTargetValue = IntegerValue == 0 ? "false" : "true"; 4741 else 4742 IntegerValue.toString(PrettyTargetValue); 4743 4744 S.Diag(FL->getExprLoc(), diag::warn_impcast_literal_float_to_integer) 4745 << FL->getType() << T.getUnqualifiedType() << PrettySourceValue 4746 << PrettyTargetValue << FL->getSourceRange() << SourceRange(CContext); 4747} 4748 4749std::string PrettyPrintInRange(const llvm::APSInt &Value, IntRange Range) { 4750 if (!Range.Width) return "0"; 4751 4752 llvm::APSInt ValueInRange = Value; 4753 ValueInRange.setIsSigned(!Range.NonNegative); 4754 ValueInRange = ValueInRange.trunc(Range.Width); 4755 return ValueInRange.toString(10); 4756} 4757 4758static bool IsImplicitBoolFloatConversion(Sema &S, Expr *Ex, bool ToBool) { 4759 if (!isa<ImplicitCastExpr>(Ex)) 4760 return false; 4761 4762 Expr *InnerE = Ex->IgnoreParenImpCasts(); 4763 const Type *Target = S.Context.getCanonicalType(Ex->getType()).getTypePtr(); 4764 const Type *Source = 4765 S.Context.getCanonicalType(InnerE->getType()).getTypePtr(); 4766 if (Target->isDependentType()) 4767 return false; 4768 4769 const BuiltinType *FloatCandidateBT = 4770 dyn_cast<BuiltinType>(ToBool ? Source : Target); 4771 const Type *BoolCandidateType = ToBool ? Target : Source; 4772 4773 return (BoolCandidateType->isSpecificBuiltinType(BuiltinType::Bool) && 4774 FloatCandidateBT && (FloatCandidateBT->isFloatingPoint())); 4775} 4776 4777void CheckImplicitArgumentConversions(Sema &S, CallExpr *TheCall, 4778 SourceLocation CC) { 4779 unsigned NumArgs = TheCall->getNumArgs(); 4780 for (unsigned i = 0; i < NumArgs; ++i) { 4781 Expr *CurrA = TheCall->getArg(i); 4782 if (!IsImplicitBoolFloatConversion(S, CurrA, true)) 4783 continue; 4784 4785 bool IsSwapped = ((i > 0) && 4786 IsImplicitBoolFloatConversion(S, TheCall->getArg(i - 1), false)); 4787 IsSwapped |= ((i < (NumArgs - 1)) && 4788 IsImplicitBoolFloatConversion(S, TheCall->getArg(i + 1), false)); 4789 if (IsSwapped) { 4790 // Warn on this floating-point to bool conversion. 4791 DiagnoseImpCast(S, CurrA->IgnoreParenImpCasts(), 4792 CurrA->getType(), CC, 4793 diag::warn_impcast_floating_point_to_bool); 4794 } 4795 } 4796} 4797 4798void CheckImplicitConversion(Sema &S, Expr *E, QualType T, 4799 SourceLocation CC, bool *ICContext = 0) { 4800 if (E->isTypeDependent() || E->isValueDependent()) return; 4801 4802 const Type *Source = S.Context.getCanonicalType(E->getType()).getTypePtr(); 4803 const Type *Target = S.Context.getCanonicalType(T).getTypePtr(); 4804 if (Source == Target) return; 4805 if (Target->isDependentType()) return; 4806 4807 // If the conversion context location is invalid don't complain. We also 4808 // don't want to emit a warning if the issue occurs from the expansion of 4809 // a system macro. The problem is that 'getSpellingLoc()' is slow, so we 4810 // delay this check as long as possible. Once we detect we are in that 4811 // scenario, we just return. 4812 if (CC.isInvalid()) 4813 return; 4814 4815 // Diagnose implicit casts to bool. 4816 if (Target->isSpecificBuiltinType(BuiltinType::Bool)) { 4817 if (isa<StringLiteral>(E)) 4818 // Warn on string literal to bool. Checks for string literals in logical 4819 // expressions, for instances, assert(0 && "error here"), is prevented 4820 // by a check in AnalyzeImplicitConversions(). 4821 return DiagnoseImpCast(S, E, T, CC, 4822 diag::warn_impcast_string_literal_to_bool); 4823 if (Source->isFunctionType()) { 4824 // Warn on function to bool. Checks free functions and static member 4825 // functions. Weakly imported functions are excluded from the check, 4826 // since it's common to test their value to check whether the linker 4827 // found a definition for them. 4828 ValueDecl *D = 0; 4829 if (DeclRefExpr* R = dyn_cast<DeclRefExpr>(E)) { 4830 D = R->getDecl(); 4831 } else if (MemberExpr *M = dyn_cast<MemberExpr>(E)) { 4832 D = M->getMemberDecl(); 4833 } 4834 4835 if (D && !D->isWeak()) { 4836 if (FunctionDecl* F = dyn_cast<FunctionDecl>(D)) { 4837 S.Diag(E->getExprLoc(), diag::warn_impcast_function_to_bool) 4838 << F << E->getSourceRange() << SourceRange(CC); 4839 S.Diag(E->getExprLoc(), diag::note_function_to_bool_silence) 4840 << FixItHint::CreateInsertion(E->getExprLoc(), "&"); 4841 QualType ReturnType; 4842 UnresolvedSet<4> NonTemplateOverloads; 4843 S.isExprCallable(*E, ReturnType, NonTemplateOverloads); 4844 if (!ReturnType.isNull() 4845 && ReturnType->isSpecificBuiltinType(BuiltinType::Bool)) 4846 S.Diag(E->getExprLoc(), diag::note_function_to_bool_call) 4847 << FixItHint::CreateInsertion( 4848 S.getPreprocessor().getLocForEndOfToken(E->getLocEnd()), "()"); 4849 return; 4850 } 4851 } 4852 } 4853 } 4854 4855 // Strip vector types. 4856 if (isa<VectorType>(Source)) { 4857 if (!isa<VectorType>(Target)) { 4858 if (S.SourceMgr.isInSystemMacro(CC)) 4859 return; 4860 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_vector_scalar); 4861 } 4862 4863 // If the vector cast is cast between two vectors of the same size, it is 4864 // a bitcast, not a conversion. 4865 if (S.Context.getTypeSize(Source) == S.Context.getTypeSize(Target)) 4866 return; 4867 4868 Source = cast<VectorType>(Source)->getElementType().getTypePtr(); 4869 Target = cast<VectorType>(Target)->getElementType().getTypePtr(); 4870 } 4871 4872 // Strip complex types. 4873 if (isa<ComplexType>(Source)) { 4874 if (!isa<ComplexType>(Target)) { 4875 if (S.SourceMgr.isInSystemMacro(CC)) 4876 return; 4877 4878 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_complex_scalar); 4879 } 4880 4881 Source = cast<ComplexType>(Source)->getElementType().getTypePtr(); 4882 Target = cast<ComplexType>(Target)->getElementType().getTypePtr(); 4883 } 4884 4885 const BuiltinType *SourceBT = dyn_cast<BuiltinType>(Source); 4886 const BuiltinType *TargetBT = dyn_cast<BuiltinType>(Target); 4887 4888 // If the source is floating point... 4889 if (SourceBT && SourceBT->isFloatingPoint()) { 4890 // ...and the target is floating point... 4891 if (TargetBT && TargetBT->isFloatingPoint()) { 4892 // ...then warn if we're dropping FP rank. 4893 4894 // Builtin FP kinds are ordered by increasing FP rank. 4895 if (SourceBT->getKind() > TargetBT->getKind()) { 4896 // Don't warn about float constants that are precisely 4897 // representable in the target type. 4898 Expr::EvalResult result; 4899 if (E->EvaluateAsRValue(result, S.Context)) { 4900 // Value might be a float, a float vector, or a float complex. 4901 if (IsSameFloatAfterCast(result.Val, 4902 S.Context.getFloatTypeSemantics(QualType(TargetBT, 0)), 4903 S.Context.getFloatTypeSemantics(QualType(SourceBT, 0)))) 4904 return; 4905 } 4906 4907 if (S.SourceMgr.isInSystemMacro(CC)) 4908 return; 4909 4910 DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_float_precision); 4911 } 4912 return; 4913 } 4914 4915 // If the target is integral, always warn. 4916 if (TargetBT && TargetBT->isInteger()) { 4917 if (S.SourceMgr.isInSystemMacro(CC)) 4918 return; 4919 4920 Expr *InnerE = E->IgnoreParenImpCasts(); 4921 // We also want to warn on, e.g., "int i = -1.234" 4922 if (UnaryOperator *UOp = dyn_cast<UnaryOperator>(InnerE)) 4923 if (UOp->getOpcode() == UO_Minus || UOp->getOpcode() == UO_Plus) 4924 InnerE = UOp->getSubExpr()->IgnoreParenImpCasts(); 4925 4926 if (FloatingLiteral *FL = dyn_cast<FloatingLiteral>(InnerE)) { 4927 DiagnoseFloatingLiteralImpCast(S, FL, T, CC); 4928 } else { 4929 DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_float_integer); 4930 } 4931 } 4932 4933 // If the target is bool, warn if expr is a function or method call. 4934 if (Target->isSpecificBuiltinType(BuiltinType::Bool) && 4935 isa<CallExpr>(E)) { 4936 // Check last argument of function call to see if it is an 4937 // implicit cast from a type matching the type the result 4938 // is being cast to. 4939 CallExpr *CEx = cast<CallExpr>(E); 4940 unsigned NumArgs = CEx->getNumArgs(); 4941 if (NumArgs > 0) { 4942 Expr *LastA = CEx->getArg(NumArgs - 1); 4943 Expr *InnerE = LastA->IgnoreParenImpCasts(); 4944 const Type *InnerType = 4945 S.Context.getCanonicalType(InnerE->getType()).getTypePtr(); 4946 if (isa<ImplicitCastExpr>(LastA) && (InnerType == Target)) { 4947 // Warn on this floating-point to bool conversion 4948 DiagnoseImpCast(S, E, T, CC, 4949 diag::warn_impcast_floating_point_to_bool); 4950 } 4951 } 4952 } 4953 return; 4954 } 4955 4956 if ((E->isNullPointerConstant(S.Context, Expr::NPC_ValueDependentIsNotNull) 4957 == Expr::NPCK_GNUNull) && !Target->isAnyPointerType() 4958 && !Target->isBlockPointerType() && !Target->isMemberPointerType() 4959 && Target->isScalarType() && !Target->isNullPtrType()) { 4960 SourceLocation Loc = E->getSourceRange().getBegin(); 4961 if (Loc.isMacroID()) 4962 Loc = S.SourceMgr.getImmediateExpansionRange(Loc).first; 4963 if (!Loc.isMacroID() || CC.isMacroID()) 4964 S.Diag(Loc, diag::warn_impcast_null_pointer_to_integer) 4965 << T << clang::SourceRange(CC) 4966 << FixItHint::CreateReplacement(Loc, S.getFixItZeroLiteralForType(T)); 4967 } 4968 4969 if (!Source->isIntegerType() || !Target->isIntegerType()) 4970 return; 4971 4972 // TODO: remove this early return once the false positives for constant->bool 4973 // in templates, macros, etc, are reduced or removed. 4974 if (Target->isSpecificBuiltinType(BuiltinType::Bool)) 4975 return; 4976 4977 IntRange SourceRange = GetExprRange(S.Context, E); 4978 IntRange TargetRange = IntRange::forTargetOfCanonicalType(S.Context, Target); 4979 4980 if (SourceRange.Width > TargetRange.Width) { 4981 // If the source is a constant, use a default-on diagnostic. 4982 // TODO: this should happen for bitfield stores, too. 4983 llvm::APSInt Value(32); 4984 if (E->isIntegerConstantExpr(Value, S.Context)) { 4985 if (S.SourceMgr.isInSystemMacro(CC)) 4986 return; 4987 4988 std::string PrettySourceValue = Value.toString(10); 4989 std::string PrettyTargetValue = PrettyPrintInRange(Value, TargetRange); 4990 4991 S.DiagRuntimeBehavior(E->getExprLoc(), E, 4992 S.PDiag(diag::warn_impcast_integer_precision_constant) 4993 << PrettySourceValue << PrettyTargetValue 4994 << E->getType() << T << E->getSourceRange() 4995 << clang::SourceRange(CC)); 4996 return; 4997 } 4998 4999 // People want to build with -Wshorten-64-to-32 and not -Wconversion. 5000 if (S.SourceMgr.isInSystemMacro(CC)) 5001 return; 5002 5003 if (TargetRange.Width == 32 && S.Context.getIntWidth(E->getType()) == 64) 5004 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_integer_64_32, 5005 /* pruneControlFlow */ true); 5006 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_integer_precision); 5007 } 5008 5009 if ((TargetRange.NonNegative && !SourceRange.NonNegative) || 5010 (!TargetRange.NonNegative && SourceRange.NonNegative && 5011 SourceRange.Width == TargetRange.Width)) { 5012 5013 if (S.SourceMgr.isInSystemMacro(CC)) 5014 return; 5015 5016 unsigned DiagID = diag::warn_impcast_integer_sign; 5017 5018 // Traditionally, gcc has warned about this under -Wsign-compare. 5019 // We also want to warn about it in -Wconversion. 5020 // So if -Wconversion is off, use a completely identical diagnostic 5021 // in the sign-compare group. 5022 // The conditional-checking code will 5023 if (ICContext) { 5024 DiagID = diag::warn_impcast_integer_sign_conditional; 5025 *ICContext = true; 5026 } 5027 5028 return DiagnoseImpCast(S, E, T, CC, DiagID); 5029 } 5030 5031 // Diagnose conversions between different enumeration types. 5032 // In C, we pretend that the type of an EnumConstantDecl is its enumeration 5033 // type, to give us better diagnostics. 5034 QualType SourceType = E->getType(); 5035 if (!S.getLangOpts().CPlusPlus) { 5036 if (DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(E)) 5037 if (EnumConstantDecl *ECD = dyn_cast<EnumConstantDecl>(DRE->getDecl())) { 5038 EnumDecl *Enum = cast<EnumDecl>(ECD->getDeclContext()); 5039 SourceType = S.Context.getTypeDeclType(Enum); 5040 Source = S.Context.getCanonicalType(SourceType).getTypePtr(); 5041 } 5042 } 5043 5044 if (const EnumType *SourceEnum = Source->getAs<EnumType>()) 5045 if (const EnumType *TargetEnum = Target->getAs<EnumType>()) 5046 if (SourceEnum->getDecl()->hasNameForLinkage() && 5047 TargetEnum->getDecl()->hasNameForLinkage() && 5048 SourceEnum != TargetEnum) { 5049 if (S.SourceMgr.isInSystemMacro(CC)) 5050 return; 5051 5052 return DiagnoseImpCast(S, E, SourceType, T, CC, 5053 diag::warn_impcast_different_enum_types); 5054 } 5055 5056 return; 5057} 5058 5059void CheckConditionalOperator(Sema &S, ConditionalOperator *E, 5060 SourceLocation CC, QualType T); 5061 5062void CheckConditionalOperand(Sema &S, Expr *E, QualType T, 5063 SourceLocation CC, bool &ICContext) { 5064 E = E->IgnoreParenImpCasts(); 5065 5066 if (isa<ConditionalOperator>(E)) 5067 return CheckConditionalOperator(S, cast<ConditionalOperator>(E), CC, T); 5068 5069 AnalyzeImplicitConversions(S, E, CC); 5070 if (E->getType() != T) 5071 return CheckImplicitConversion(S, E, T, CC, &ICContext); 5072 return; 5073} 5074 5075void CheckConditionalOperator(Sema &S, ConditionalOperator *E, 5076 SourceLocation CC, QualType T) { 5077 AnalyzeImplicitConversions(S, E->getCond(), CC); 5078 5079 bool Suspicious = false; 5080 CheckConditionalOperand(S, E->getTrueExpr(), T, CC, Suspicious); 5081 CheckConditionalOperand(S, E->getFalseExpr(), T, CC, Suspicious); 5082 5083 // If -Wconversion would have warned about either of the candidates 5084 // for a signedness conversion to the context type... 5085 if (!Suspicious) return; 5086 5087 // ...but it's currently ignored... 5088 if (S.Diags.getDiagnosticLevel(diag::warn_impcast_integer_sign_conditional, 5089 CC)) 5090 return; 5091 5092 // ...then check whether it would have warned about either of the 5093 // candidates for a signedness conversion to the condition type. 5094 if (E->getType() == T) return; 5095 5096 Suspicious = false; 5097 CheckImplicitConversion(S, E->getTrueExpr()->IgnoreParenImpCasts(), 5098 E->getType(), CC, &Suspicious); 5099 if (!Suspicious) 5100 CheckImplicitConversion(S, E->getFalseExpr()->IgnoreParenImpCasts(), 5101 E->getType(), CC, &Suspicious); 5102} 5103 5104/// AnalyzeImplicitConversions - Find and report any interesting 5105/// implicit conversions in the given expression. There are a couple 5106/// of competing diagnostics here, -Wconversion and -Wsign-compare. 5107void AnalyzeImplicitConversions(Sema &S, Expr *OrigE, SourceLocation CC) { 5108 QualType T = OrigE->getType(); 5109 Expr *E = OrigE->IgnoreParenImpCasts(); 5110 5111 if (E->isTypeDependent() || E->isValueDependent()) 5112 return; 5113 5114 // For conditional operators, we analyze the arguments as if they 5115 // were being fed directly into the output. 5116 if (isa<ConditionalOperator>(E)) { 5117 ConditionalOperator *CO = cast<ConditionalOperator>(E); 5118 CheckConditionalOperator(S, CO, CC, T); 5119 return; 5120 } 5121 5122 // Check implicit argument conversions for function calls. 5123 if (CallExpr *Call = dyn_cast<CallExpr>(E)) 5124 CheckImplicitArgumentConversions(S, Call, CC); 5125 5126 // Go ahead and check any implicit conversions we might have skipped. 5127 // The non-canonical typecheck is just an optimization; 5128 // CheckImplicitConversion will filter out dead implicit conversions. 5129 if (E->getType() != T) 5130 CheckImplicitConversion(S, E, T, CC); 5131 5132 // Now continue drilling into this expression. 5133 5134 // Skip past explicit casts. 5135 if (isa<ExplicitCastExpr>(E)) { 5136 E = cast<ExplicitCastExpr>(E)->getSubExpr()->IgnoreParenImpCasts(); 5137 return AnalyzeImplicitConversions(S, E, CC); 5138 } 5139 5140 if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) { 5141 // Do a somewhat different check with comparison operators. 5142 if (BO->isComparisonOp()) 5143 return AnalyzeComparison(S, BO); 5144 5145 // And with simple assignments. 5146 if (BO->getOpcode() == BO_Assign) 5147 return AnalyzeAssignment(S, BO); 5148 } 5149 5150 // These break the otherwise-useful invariant below. Fortunately, 5151 // we don't really need to recurse into them, because any internal 5152 // expressions should have been analyzed already when they were 5153 // built into statements. 5154 if (isa<StmtExpr>(E)) return; 5155 5156 // Don't descend into unevaluated contexts. 5157 if (isa<UnaryExprOrTypeTraitExpr>(E)) return; 5158 5159 // Now just recurse over the expression's children. 5160 CC = E->getExprLoc(); 5161 BinaryOperator *BO = dyn_cast<BinaryOperator>(E); 5162 bool IsLogicalOperator = BO && BO->isLogicalOp(); 5163 for (Stmt::child_range I = E->children(); I; ++I) { 5164 Expr *ChildExpr = dyn_cast_or_null<Expr>(*I); 5165 if (!ChildExpr) 5166 continue; 5167 5168 if (IsLogicalOperator && 5169 isa<StringLiteral>(ChildExpr->IgnoreParenImpCasts())) 5170 // Ignore checking string literals that are in logical operators. 5171 continue; 5172 AnalyzeImplicitConversions(S, ChildExpr, CC); 5173 } 5174} 5175 5176} // end anonymous namespace 5177 5178/// Diagnoses "dangerous" implicit conversions within the given 5179/// expression (which is a full expression). Implements -Wconversion 5180/// and -Wsign-compare. 5181/// 5182/// \param CC the "context" location of the implicit conversion, i.e. 5183/// the most location of the syntactic entity requiring the implicit 5184/// conversion 5185void Sema::CheckImplicitConversions(Expr *E, SourceLocation CC) { 5186 // Don't diagnose in unevaluated contexts. 5187 if (isUnevaluatedContext()) 5188 return; 5189 5190 // Don't diagnose for value- or type-dependent expressions. 5191 if (E->isTypeDependent() || E->isValueDependent()) 5192 return; 5193 5194 // Check for array bounds violations in cases where the check isn't triggered 5195 // elsewhere for other Expr types (like BinaryOperators), e.g. when an 5196 // ArraySubscriptExpr is on the RHS of a variable initialization. 5197 CheckArrayAccess(E); 5198 5199 // This is not the right CC for (e.g.) a variable initialization. 5200 AnalyzeImplicitConversions(*this, E, CC); 5201} 5202 5203/// Diagnose when expression is an integer constant expression and its evaluation 5204/// results in integer overflow 5205void Sema::CheckForIntOverflow (Expr *E) { 5206 if (isa<BinaryOperator>(E->IgnoreParens())) { 5207 llvm::SmallVector<PartialDiagnosticAt, 4> Diags; 5208 E->EvaluateForOverflow(Context, &Diags); 5209 } 5210} 5211 5212namespace { 5213/// \brief Visitor for expressions which looks for unsequenced operations on the 5214/// same object. 5215class SequenceChecker : public EvaluatedExprVisitor<SequenceChecker> { 5216 /// \brief A tree of sequenced regions within an expression. Two regions are 5217 /// unsequenced if one is an ancestor or a descendent of the other. When we 5218 /// finish processing an expression with sequencing, such as a comma 5219 /// expression, we fold its tree nodes into its parent, since they are 5220 /// unsequenced with respect to nodes we will visit later. 5221 class SequenceTree { 5222 struct Value { 5223 explicit Value(unsigned Parent) : Parent(Parent), Merged(false) {} 5224 unsigned Parent : 31; 5225 bool Merged : 1; 5226 }; 5227 llvm::SmallVector<Value, 8> Values; 5228 5229 public: 5230 /// \brief A region within an expression which may be sequenced with respect 5231 /// to some other region. 5232 class Seq { 5233 explicit Seq(unsigned N) : Index(N) {} 5234 unsigned Index; 5235 friend class SequenceTree; 5236 public: 5237 Seq() : Index(0) {} 5238 }; 5239 5240 SequenceTree() { Values.push_back(Value(0)); } 5241 Seq root() const { return Seq(0); } 5242 5243 /// \brief Create a new sequence of operations, which is an unsequenced 5244 /// subset of \p Parent. This sequence of operations is sequenced with 5245 /// respect to other children of \p Parent. 5246 Seq allocate(Seq Parent) { 5247 Values.push_back(Value(Parent.Index)); 5248 return Seq(Values.size() - 1); 5249 } 5250 5251 /// \brief Merge a sequence of operations into its parent. 5252 void merge(Seq S) { 5253 Values[S.Index].Merged = true; 5254 } 5255 5256 /// \brief Determine whether two operations are unsequenced. This operation 5257 /// is asymmetric: \p Cur should be the more recent sequence, and \p Old 5258 /// should have been merged into its parent as appropriate. 5259 bool isUnsequenced(Seq Cur, Seq Old) { 5260 unsigned C = representative(Cur.Index); 5261 unsigned Target = representative(Old.Index); 5262 while (C >= Target) { 5263 if (C == Target) 5264 return true; 5265 C = Values[C].Parent; 5266 } 5267 return false; 5268 } 5269 5270 private: 5271 /// \brief Pick a representative for a sequence. 5272 unsigned representative(unsigned K) { 5273 if (Values[K].Merged) 5274 // Perform path compression as we go. 5275 return Values[K].Parent = representative(Values[K].Parent); 5276 return K; 5277 } 5278 }; 5279 5280 /// An object for which we can track unsequenced uses. 5281 typedef NamedDecl *Object; 5282 5283 /// Different flavors of object usage which we track. We only track the 5284 /// least-sequenced usage of each kind. 5285 enum UsageKind { 5286 /// A read of an object. Multiple unsequenced reads are OK. 5287 UK_Use, 5288 /// A modification of an object which is sequenced before the value 5289 /// computation of the expression, such as ++n. 5290 UK_ModAsValue, 5291 /// A modification of an object which is not sequenced before the value 5292 /// computation of the expression, such as n++. 5293 UK_ModAsSideEffect, 5294 5295 UK_Count = UK_ModAsSideEffect + 1 5296 }; 5297 5298 struct Usage { 5299 Usage() : Use(0), Seq() {} 5300 Expr *Use; 5301 SequenceTree::Seq Seq; 5302 }; 5303 5304 struct UsageInfo { 5305 UsageInfo() : Diagnosed(false) {} 5306 Usage Uses[UK_Count]; 5307 /// Have we issued a diagnostic for this variable already? 5308 bool Diagnosed; 5309 }; 5310 typedef llvm::SmallDenseMap<Object, UsageInfo, 16> UsageInfoMap; 5311 5312 Sema &SemaRef; 5313 /// Sequenced regions within the expression. 5314 SequenceTree Tree; 5315 /// Declaration modifications and references which we have seen. 5316 UsageInfoMap UsageMap; 5317 /// The region we are currently within. 5318 SequenceTree::Seq Region; 5319 /// Filled in with declarations which were modified as a side-effect 5320 /// (that is, post-increment operations). 5321 llvm::SmallVectorImpl<std::pair<Object, Usage> > *ModAsSideEffect; 5322 /// Expressions to check later. We defer checking these to reduce 5323 /// stack usage. 5324 llvm::SmallVectorImpl<Expr*> &WorkList; 5325 5326 /// RAII object wrapping the visitation of a sequenced subexpression of an 5327 /// expression. At the end of this process, the side-effects of the evaluation 5328 /// become sequenced with respect to the value computation of the result, so 5329 /// we downgrade any UK_ModAsSideEffect within the evaluation to 5330 /// UK_ModAsValue. 5331 struct SequencedSubexpression { 5332 SequencedSubexpression(SequenceChecker &Self) 5333 : Self(Self), OldModAsSideEffect(Self.ModAsSideEffect) { 5334 Self.ModAsSideEffect = &ModAsSideEffect; 5335 } 5336 ~SequencedSubexpression() { 5337 for (unsigned I = 0, E = ModAsSideEffect.size(); I != E; ++I) { 5338 UsageInfo &U = Self.UsageMap[ModAsSideEffect[I].first]; 5339 U.Uses[UK_ModAsSideEffect] = ModAsSideEffect[I].second; 5340 Self.addUsage(U, ModAsSideEffect[I].first, 5341 ModAsSideEffect[I].second.Use, UK_ModAsValue); 5342 } 5343 Self.ModAsSideEffect = OldModAsSideEffect; 5344 } 5345 5346 SequenceChecker &Self; 5347 llvm::SmallVector<std::pair<Object, Usage>, 4> ModAsSideEffect; 5348 llvm::SmallVectorImpl<std::pair<Object, Usage> > *OldModAsSideEffect; 5349 }; 5350 5351 /// \brief Find the object which is produced by the specified expression, 5352 /// if any. 5353 Object getObject(Expr *E, bool Mod) const { 5354 E = E->IgnoreParenCasts(); 5355 if (UnaryOperator *UO = dyn_cast<UnaryOperator>(E)) { 5356 if (Mod && (UO->getOpcode() == UO_PreInc || UO->getOpcode() == UO_PreDec)) 5357 return getObject(UO->getSubExpr(), Mod); 5358 } else if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) { 5359 if (BO->getOpcode() == BO_Comma) 5360 return getObject(BO->getRHS(), Mod); 5361 if (Mod && BO->isAssignmentOp()) 5362 return getObject(BO->getLHS(), Mod); 5363 } else if (MemberExpr *ME = dyn_cast<MemberExpr>(E)) { 5364 // FIXME: Check for more interesting cases, like "x.n = ++x.n". 5365 if (isa<CXXThisExpr>(ME->getBase()->IgnoreParenCasts())) 5366 return ME->getMemberDecl(); 5367 } else if (DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(E)) 5368 // FIXME: If this is a reference, map through to its value. 5369 return DRE->getDecl(); 5370 return 0; 5371 } 5372 5373 /// \brief Note that an object was modified or used by an expression. 5374 void addUsage(UsageInfo &UI, Object O, Expr *Ref, UsageKind UK) { 5375 Usage &U = UI.Uses[UK]; 5376 if (!U.Use || !Tree.isUnsequenced(Region, U.Seq)) { 5377 if (UK == UK_ModAsSideEffect && ModAsSideEffect) 5378 ModAsSideEffect->push_back(std::make_pair(O, U)); 5379 U.Use = Ref; 5380 U.Seq = Region; 5381 } 5382 } 5383 /// \brief Check whether a modification or use conflicts with a prior usage. 5384 void checkUsage(Object O, UsageInfo &UI, Expr *Ref, UsageKind OtherKind, 5385 bool IsModMod) { 5386 if (UI.Diagnosed) 5387 return; 5388 5389 const Usage &U = UI.Uses[OtherKind]; 5390 if (!U.Use || !Tree.isUnsequenced(Region, U.Seq)) 5391 return; 5392 5393 Expr *Mod = U.Use; 5394 Expr *ModOrUse = Ref; 5395 if (OtherKind == UK_Use) 5396 std::swap(Mod, ModOrUse); 5397 5398 SemaRef.Diag(Mod->getExprLoc(), 5399 IsModMod ? diag::warn_unsequenced_mod_mod 5400 : diag::warn_unsequenced_mod_use) 5401 << O << SourceRange(ModOrUse->getExprLoc()); 5402 UI.Diagnosed = true; 5403 } 5404 5405 void notePreUse(Object O, Expr *Use) { 5406 UsageInfo &U = UsageMap[O]; 5407 // Uses conflict with other modifications. 5408 checkUsage(O, U, Use, UK_ModAsValue, false); 5409 } 5410 void notePostUse(Object O, Expr *Use) { 5411 UsageInfo &U = UsageMap[O]; 5412 checkUsage(O, U, Use, UK_ModAsSideEffect, false); 5413 addUsage(U, O, Use, UK_Use); 5414 } 5415 5416 void notePreMod(Object O, Expr *Mod) { 5417 UsageInfo &U = UsageMap[O]; 5418 // Modifications conflict with other modifications and with uses. 5419 checkUsage(O, U, Mod, UK_ModAsValue, true); 5420 checkUsage(O, U, Mod, UK_Use, false); 5421 } 5422 void notePostMod(Object O, Expr *Use, UsageKind UK) { 5423 UsageInfo &U = UsageMap[O]; 5424 checkUsage(O, U, Use, UK_ModAsSideEffect, true); 5425 addUsage(U, O, Use, UK); 5426 } 5427 5428public: 5429 SequenceChecker(Sema &S, Expr *E, 5430 llvm::SmallVectorImpl<Expr*> &WorkList) 5431 : EvaluatedExprVisitor<SequenceChecker>(S.Context), SemaRef(S), 5432 Region(Tree.root()), ModAsSideEffect(0), WorkList(WorkList) { 5433 Visit(E); 5434 } 5435 5436 void VisitStmt(Stmt *S) { 5437 // Skip all statements which aren't expressions for now. 5438 } 5439 5440 void VisitExpr(Expr *E) { 5441 // By default, just recurse to evaluated subexpressions. 5442 EvaluatedExprVisitor<SequenceChecker>::VisitStmt(E); 5443 } 5444 5445 void VisitCastExpr(CastExpr *E) { 5446 Object O = Object(); 5447 if (E->getCastKind() == CK_LValueToRValue) 5448 O = getObject(E->getSubExpr(), false); 5449 5450 if (O) 5451 notePreUse(O, E); 5452 VisitExpr(E); 5453 if (O) 5454 notePostUse(O, E); 5455 } 5456 5457 void VisitBinComma(BinaryOperator *BO) { 5458 // C++11 [expr.comma]p1: 5459 // Every value computation and side effect associated with the left 5460 // expression is sequenced before every value computation and side 5461 // effect associated with the right expression. 5462 SequenceTree::Seq LHS = Tree.allocate(Region); 5463 SequenceTree::Seq RHS = Tree.allocate(Region); 5464 SequenceTree::Seq OldRegion = Region; 5465 5466 { 5467 SequencedSubexpression SeqLHS(*this); 5468 Region = LHS; 5469 Visit(BO->getLHS()); 5470 } 5471 5472 Region = RHS; 5473 Visit(BO->getRHS()); 5474 5475 Region = OldRegion; 5476 5477 // Forget that LHS and RHS are sequenced. They are both unsequenced 5478 // with respect to other stuff. 5479 Tree.merge(LHS); 5480 Tree.merge(RHS); 5481 } 5482 5483 void VisitBinAssign(BinaryOperator *BO) { 5484 // The modification is sequenced after the value computation of the LHS 5485 // and RHS, so check it before inspecting the operands and update the 5486 // map afterwards. 5487 Object O = getObject(BO->getLHS(), true); 5488 if (!O) 5489 return VisitExpr(BO); 5490 5491 notePreMod(O, BO); 5492 5493 // C++11 [expr.ass]p7: 5494 // E1 op= E2 is equivalent to E1 = E1 op E2, except that E1 is evaluated 5495 // only once. 5496 // 5497 // Therefore, for a compound assignment operator, O is considered used 5498 // everywhere except within the evaluation of E1 itself. 5499 if (isa<CompoundAssignOperator>(BO)) 5500 notePreUse(O, BO); 5501 5502 Visit(BO->getLHS()); 5503 5504 if (isa<CompoundAssignOperator>(BO)) 5505 notePostUse(O, BO); 5506 5507 Visit(BO->getRHS()); 5508 5509 notePostMod(O, BO, UK_ModAsValue); 5510 } 5511 void VisitCompoundAssignOperator(CompoundAssignOperator *CAO) { 5512 VisitBinAssign(CAO); 5513 } 5514 5515 void VisitUnaryPreInc(UnaryOperator *UO) { VisitUnaryPreIncDec(UO); } 5516 void VisitUnaryPreDec(UnaryOperator *UO) { VisitUnaryPreIncDec(UO); } 5517 void VisitUnaryPreIncDec(UnaryOperator *UO) { 5518 Object O = getObject(UO->getSubExpr(), true); 5519 if (!O) 5520 return VisitExpr(UO); 5521 5522 notePreMod(O, UO); 5523 Visit(UO->getSubExpr()); 5524 notePostMod(O, UO, UK_ModAsValue); 5525 } 5526 5527 void VisitUnaryPostInc(UnaryOperator *UO) { VisitUnaryPostIncDec(UO); } 5528 void VisitUnaryPostDec(UnaryOperator *UO) { VisitUnaryPostIncDec(UO); } 5529 void VisitUnaryPostIncDec(UnaryOperator *UO) { 5530 Object O = getObject(UO->getSubExpr(), true); 5531 if (!O) 5532 return VisitExpr(UO); 5533 5534 notePreMod(O, UO); 5535 Visit(UO->getSubExpr()); 5536 notePostMod(O, UO, UK_ModAsSideEffect); 5537 } 5538 5539 /// Don't visit the RHS of '&&' or '||' if it might not be evaluated. 5540 void VisitBinLOr(BinaryOperator *BO) { 5541 // The side-effects of the LHS of an '&&' are sequenced before the 5542 // value computation of the RHS, and hence before the value computation 5543 // of the '&&' itself, unless the LHS evaluates to zero. We treat them 5544 // as if they were unconditionally sequenced. 5545 { 5546 SequencedSubexpression Sequenced(*this); 5547 Visit(BO->getLHS()); 5548 } 5549 5550 bool Result; 5551 if (!BO->getLHS()->isValueDependent() && 5552 BO->getLHS()->EvaluateAsBooleanCondition(Result, SemaRef.Context)) { 5553 if (!Result) 5554 Visit(BO->getRHS()); 5555 } else { 5556 // Check for unsequenced operations in the RHS, treating it as an 5557 // entirely separate evaluation. 5558 // 5559 // FIXME: If there are operations in the RHS which are unsequenced 5560 // with respect to operations outside the RHS, and those operations 5561 // are unconditionally evaluated, diagnose them. 5562 WorkList.push_back(BO->getRHS()); 5563 } 5564 } 5565 void VisitBinLAnd(BinaryOperator *BO) { 5566 { 5567 SequencedSubexpression Sequenced(*this); 5568 Visit(BO->getLHS()); 5569 } 5570 5571 bool Result; 5572 if (!BO->getLHS()->isValueDependent() && 5573 BO->getLHS()->EvaluateAsBooleanCondition(Result, SemaRef.Context)) { 5574 if (Result) 5575 Visit(BO->getRHS()); 5576 } else { 5577 WorkList.push_back(BO->getRHS()); 5578 } 5579 } 5580 5581 // Only visit the condition, unless we can be sure which subexpression will 5582 // be chosen. 5583 void VisitAbstractConditionalOperator(AbstractConditionalOperator *CO) { 5584 SequencedSubexpression Sequenced(*this); 5585 Visit(CO->getCond()); 5586 5587 bool Result; 5588 if (!CO->getCond()->isValueDependent() && 5589 CO->getCond()->EvaluateAsBooleanCondition(Result, SemaRef.Context)) 5590 Visit(Result ? CO->getTrueExpr() : CO->getFalseExpr()); 5591 else { 5592 WorkList.push_back(CO->getTrueExpr()); 5593 WorkList.push_back(CO->getFalseExpr()); 5594 } 5595 } 5596 5597 void VisitCXXConstructExpr(CXXConstructExpr *CCE) { 5598 if (!CCE->isListInitialization()) 5599 return VisitExpr(CCE); 5600 5601 // In C++11, list initializations are sequenced. 5602 llvm::SmallVector<SequenceTree::Seq, 32> Elts; 5603 SequenceTree::Seq Parent = Region; 5604 for (CXXConstructExpr::arg_iterator I = CCE->arg_begin(), 5605 E = CCE->arg_end(); 5606 I != E; ++I) { 5607 Region = Tree.allocate(Parent); 5608 Elts.push_back(Region); 5609 Visit(*I); 5610 } 5611 5612 // Forget that the initializers are sequenced. 5613 Region = Parent; 5614 for (unsigned I = 0; I < Elts.size(); ++I) 5615 Tree.merge(Elts[I]); 5616 } 5617 5618 void VisitInitListExpr(InitListExpr *ILE) { 5619 if (!SemaRef.getLangOpts().CPlusPlus11) 5620 return VisitExpr(ILE); 5621 5622 // In C++11, list initializations are sequenced. 5623 llvm::SmallVector<SequenceTree::Seq, 32> Elts; 5624 SequenceTree::Seq Parent = Region; 5625 for (unsigned I = 0; I < ILE->getNumInits(); ++I) { 5626 Expr *E = ILE->getInit(I); 5627 if (!E) continue; 5628 Region = Tree.allocate(Parent); 5629 Elts.push_back(Region); 5630 Visit(E); 5631 } 5632 5633 // Forget that the initializers are sequenced. 5634 Region = Parent; 5635 for (unsigned I = 0; I < Elts.size(); ++I) 5636 Tree.merge(Elts[I]); 5637 } 5638}; 5639} 5640 5641void Sema::CheckUnsequencedOperations(Expr *E) { 5642 llvm::SmallVector<Expr*, 8> WorkList; 5643 WorkList.push_back(E); 5644 while (!WorkList.empty()) { 5645 Expr *Item = WorkList.back(); 5646 WorkList.pop_back(); 5647 SequenceChecker(*this, Item, WorkList); 5648 } 5649} 5650 5651void Sema::CheckCompletedExpr(Expr *E, SourceLocation CheckLoc, 5652 bool IsConstexpr) { 5653 CheckImplicitConversions(E, CheckLoc); 5654 CheckUnsequencedOperations(E); 5655 if (!IsConstexpr && !E->isValueDependent()) 5656 CheckForIntOverflow(E); 5657} 5658 5659void Sema::CheckBitFieldInitialization(SourceLocation InitLoc, 5660 FieldDecl *BitField, 5661 Expr *Init) { 5662 (void) AnalyzeBitFieldAssignment(*this, BitField, Init, InitLoc); 5663} 5664 5665/// CheckParmsForFunctionDef - Check that the parameters of the given 5666/// function are appropriate for the definition of a function. This 5667/// takes care of any checks that cannot be performed on the 5668/// declaration itself, e.g., that the types of each of the function 5669/// parameters are complete. 5670bool Sema::CheckParmsForFunctionDef(ParmVarDecl **P, ParmVarDecl **PEnd, 5671 bool CheckParameterNames) { 5672 bool HasInvalidParm = false; 5673 for (; P != PEnd; ++P) { 5674 ParmVarDecl *Param = *P; 5675 5676 // C99 6.7.5.3p4: the parameters in a parameter type list in a 5677 // function declarator that is part of a function definition of 5678 // that function shall not have incomplete type. 5679 // 5680 // This is also C++ [dcl.fct]p6. 5681 if (!Param->isInvalidDecl() && 5682 RequireCompleteType(Param->getLocation(), Param->getType(), 5683 diag::err_typecheck_decl_incomplete_type)) { 5684 Param->setInvalidDecl(); 5685 HasInvalidParm = true; 5686 } 5687 5688 // C99 6.9.1p5: If the declarator includes a parameter type list, the 5689 // declaration of each parameter shall include an identifier. 5690 if (CheckParameterNames && 5691 Param->getIdentifier() == 0 && 5692 !Param->isImplicit() && 5693 !getLangOpts().CPlusPlus) 5694 Diag(Param->getLocation(), diag::err_parameter_name_omitted); 5695 5696 // C99 6.7.5.3p12: 5697 // If the function declarator is not part of a definition of that 5698 // function, parameters may have incomplete type and may use the [*] 5699 // notation in their sequences of declarator specifiers to specify 5700 // variable length array types. 5701 QualType PType = Param->getOriginalType(); 5702 if (const ArrayType *AT = Context.getAsArrayType(PType)) { 5703 if (AT->getSizeModifier() == ArrayType::Star) { 5704 // FIXME: This diagnostic should point the '[*]' if source-location 5705 // information is added for it. 5706 Diag(Param->getLocation(), diag::err_array_star_in_function_definition); 5707 } 5708 } 5709 } 5710 5711 return HasInvalidParm; 5712} 5713 5714/// CheckCastAlign - Implements -Wcast-align, which warns when a 5715/// pointer cast increases the alignment requirements. 5716void Sema::CheckCastAlign(Expr *Op, QualType T, SourceRange TRange) { 5717 // This is actually a lot of work to potentially be doing on every 5718 // cast; don't do it if we're ignoring -Wcast_align (as is the default). 5719 if (getDiagnostics().getDiagnosticLevel(diag::warn_cast_align, 5720 TRange.getBegin()) 5721 == DiagnosticsEngine::Ignored) 5722 return; 5723 5724 // Ignore dependent types. 5725 if (T->isDependentType() || Op->getType()->isDependentType()) 5726 return; 5727 5728 // Require that the destination be a pointer type. 5729 const PointerType *DestPtr = T->getAs<PointerType>(); 5730 if (!DestPtr) return; 5731 5732 // If the destination has alignment 1, we're done. 5733 QualType DestPointee = DestPtr->getPointeeType(); 5734 if (DestPointee->isIncompleteType()) return; 5735 CharUnits DestAlign = Context.getTypeAlignInChars(DestPointee); 5736 if (DestAlign.isOne()) return; 5737 5738 // Require that the source be a pointer type. 5739 const PointerType *SrcPtr = Op->getType()->getAs<PointerType>(); 5740 if (!SrcPtr) return; 5741 QualType SrcPointee = SrcPtr->getPointeeType(); 5742 5743 // Whitelist casts from cv void*. We already implicitly 5744 // whitelisted casts to cv void*, since they have alignment 1. 5745 // Also whitelist casts involving incomplete types, which implicitly 5746 // includes 'void'. 5747 if (SrcPointee->isIncompleteType()) return; 5748 5749 CharUnits SrcAlign = Context.getTypeAlignInChars(SrcPointee); 5750 if (SrcAlign >= DestAlign) return; 5751 5752 Diag(TRange.getBegin(), diag::warn_cast_align) 5753 << Op->getType() << T 5754 << static_cast<unsigned>(SrcAlign.getQuantity()) 5755 << static_cast<unsigned>(DestAlign.getQuantity()) 5756 << TRange << Op->getSourceRange(); 5757} 5758 5759static const Type* getElementType(const Expr *BaseExpr) { 5760 const Type* EltType = BaseExpr->getType().getTypePtr(); 5761 if (EltType->isAnyPointerType()) 5762 return EltType->getPointeeType().getTypePtr(); 5763 else if (EltType->isArrayType()) 5764 return EltType->getBaseElementTypeUnsafe(); 5765 return EltType; 5766} 5767 5768/// \brief Check whether this array fits the idiom of a size-one tail padded 5769/// array member of a struct. 5770/// 5771/// We avoid emitting out-of-bounds access warnings for such arrays as they are 5772/// commonly used to emulate flexible arrays in C89 code. 5773static bool IsTailPaddedMemberArray(Sema &S, llvm::APInt Size, 5774 const NamedDecl *ND) { 5775 if (Size != 1 || !ND) return false; 5776 5777 const FieldDecl *FD = dyn_cast<FieldDecl>(ND); 5778 if (!FD) return false; 5779 5780 // Don't consider sizes resulting from macro expansions or template argument 5781 // substitution to form C89 tail-padded arrays. 5782 5783 TypeSourceInfo *TInfo = FD->getTypeSourceInfo(); 5784 while (TInfo) { 5785 TypeLoc TL = TInfo->getTypeLoc(); 5786 // Look through typedefs. 5787 if (TypedefTypeLoc TTL = TL.getAs<TypedefTypeLoc>()) { 5788 const TypedefNameDecl *TDL = TTL.getTypedefNameDecl(); 5789 TInfo = TDL->getTypeSourceInfo(); 5790 continue; 5791 } 5792 if (ConstantArrayTypeLoc CTL = TL.getAs<ConstantArrayTypeLoc>()) { 5793 const Expr *SizeExpr = dyn_cast<IntegerLiteral>(CTL.getSizeExpr()); 5794 if (!SizeExpr || SizeExpr->getExprLoc().isMacroID()) 5795 return false; 5796 } 5797 break; 5798 } 5799 5800 const RecordDecl *RD = dyn_cast<RecordDecl>(FD->getDeclContext()); 5801 if (!RD) return false; 5802 if (RD->isUnion()) return false; 5803 if (const CXXRecordDecl *CRD = dyn_cast<CXXRecordDecl>(RD)) { 5804 if (!CRD->isStandardLayout()) return false; 5805 } 5806 5807 // See if this is the last field decl in the record. 5808 const Decl *D = FD; 5809 while ((D = D->getNextDeclInContext())) 5810 if (isa<FieldDecl>(D)) 5811 return false; 5812 return true; 5813} 5814 5815void Sema::CheckArrayAccess(const Expr *BaseExpr, const Expr *IndexExpr, 5816 const ArraySubscriptExpr *ASE, 5817 bool AllowOnePastEnd, bool IndexNegated) { 5818 IndexExpr = IndexExpr->IgnoreParenImpCasts(); 5819 if (IndexExpr->isValueDependent()) 5820 return; 5821 5822 const Type *EffectiveType = getElementType(BaseExpr); 5823 BaseExpr = BaseExpr->IgnoreParenCasts(); 5824 const ConstantArrayType *ArrayTy = 5825 Context.getAsConstantArrayType(BaseExpr->getType()); 5826 if (!ArrayTy) 5827 return; 5828 5829 llvm::APSInt index; 5830 if (!IndexExpr->EvaluateAsInt(index, Context)) 5831 return; 5832 if (IndexNegated) 5833 index = -index; 5834 5835 const NamedDecl *ND = NULL; 5836 if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(BaseExpr)) 5837 ND = dyn_cast<NamedDecl>(DRE->getDecl()); 5838 if (const MemberExpr *ME = dyn_cast<MemberExpr>(BaseExpr)) 5839 ND = dyn_cast<NamedDecl>(ME->getMemberDecl()); 5840 5841 if (index.isUnsigned() || !index.isNegative()) { 5842 llvm::APInt size = ArrayTy->getSize(); 5843 if (!size.isStrictlyPositive()) 5844 return; 5845 5846 const Type* BaseType = getElementType(BaseExpr); 5847 if (BaseType != EffectiveType) { 5848 // Make sure we're comparing apples to apples when comparing index to size 5849 uint64_t ptrarith_typesize = Context.getTypeSize(EffectiveType); 5850 uint64_t array_typesize = Context.getTypeSize(BaseType); 5851 // Handle ptrarith_typesize being zero, such as when casting to void* 5852 if (!ptrarith_typesize) ptrarith_typesize = 1; 5853 if (ptrarith_typesize != array_typesize) { 5854 // There's a cast to a different size type involved 5855 uint64_t ratio = array_typesize / ptrarith_typesize; 5856 // TODO: Be smarter about handling cases where array_typesize is not a 5857 // multiple of ptrarith_typesize 5858 if (ptrarith_typesize * ratio == array_typesize) 5859 size *= llvm::APInt(size.getBitWidth(), ratio); 5860 } 5861 } 5862 5863 if (size.getBitWidth() > index.getBitWidth()) 5864 index = index.zext(size.getBitWidth()); 5865 else if (size.getBitWidth() < index.getBitWidth()) 5866 size = size.zext(index.getBitWidth()); 5867 5868 // For array subscripting the index must be less than size, but for pointer 5869 // arithmetic also allow the index (offset) to be equal to size since 5870 // computing the next address after the end of the array is legal and 5871 // commonly done e.g. in C++ iterators and range-based for loops. 5872 if (AllowOnePastEnd ? index.ule(size) : index.ult(size)) 5873 return; 5874 5875 // Also don't warn for arrays of size 1 which are members of some 5876 // structure. These are often used to approximate flexible arrays in C89 5877 // code. 5878 if (IsTailPaddedMemberArray(*this, size, ND)) 5879 return; 5880 5881 // Suppress the warning if the subscript expression (as identified by the 5882 // ']' location) and the index expression are both from macro expansions 5883 // within a system header. 5884 if (ASE) { 5885 SourceLocation RBracketLoc = SourceMgr.getSpellingLoc( 5886 ASE->getRBracketLoc()); 5887 if (SourceMgr.isInSystemHeader(RBracketLoc)) { 5888 SourceLocation IndexLoc = SourceMgr.getSpellingLoc( 5889 IndexExpr->getLocStart()); 5890 if (SourceMgr.isFromSameFile(RBracketLoc, IndexLoc)) 5891 return; 5892 } 5893 } 5894 5895 unsigned DiagID = diag::warn_ptr_arith_exceeds_bounds; 5896 if (ASE) 5897 DiagID = diag::warn_array_index_exceeds_bounds; 5898 5899 DiagRuntimeBehavior(BaseExpr->getLocStart(), BaseExpr, 5900 PDiag(DiagID) << index.toString(10, true) 5901 << size.toString(10, true) 5902 << (unsigned)size.getLimitedValue(~0U) 5903 << IndexExpr->getSourceRange()); 5904 } else { 5905 unsigned DiagID = diag::warn_array_index_precedes_bounds; 5906 if (!ASE) { 5907 DiagID = diag::warn_ptr_arith_precedes_bounds; 5908 if (index.isNegative()) index = -index; 5909 } 5910 5911 DiagRuntimeBehavior(BaseExpr->getLocStart(), BaseExpr, 5912 PDiag(DiagID) << index.toString(10, true) 5913 << IndexExpr->getSourceRange()); 5914 } 5915 5916 if (!ND) { 5917 // Try harder to find a NamedDecl to point at in the note. 5918 while (const ArraySubscriptExpr *ASE = 5919 dyn_cast<ArraySubscriptExpr>(BaseExpr)) 5920 BaseExpr = ASE->getBase()->IgnoreParenCasts(); 5921 if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(BaseExpr)) 5922 ND = dyn_cast<NamedDecl>(DRE->getDecl()); 5923 if (const MemberExpr *ME = dyn_cast<MemberExpr>(BaseExpr)) 5924 ND = dyn_cast<NamedDecl>(ME->getMemberDecl()); 5925 } 5926 5927 if (ND) 5928 DiagRuntimeBehavior(ND->getLocStart(), BaseExpr, 5929 PDiag(diag::note_array_index_out_of_bounds) 5930 << ND->getDeclName()); 5931} 5932 5933void Sema::CheckArrayAccess(const Expr *expr) { 5934 int AllowOnePastEnd = 0; 5935 while (expr) { 5936 expr = expr->IgnoreParenImpCasts(); 5937 switch (expr->getStmtClass()) { 5938 case Stmt::ArraySubscriptExprClass: { 5939 const ArraySubscriptExpr *ASE = cast<ArraySubscriptExpr>(expr); 5940 CheckArrayAccess(ASE->getBase(), ASE->getIdx(), ASE, 5941 AllowOnePastEnd > 0); 5942 return; 5943 } 5944 case Stmt::UnaryOperatorClass: { 5945 // Only unwrap the * and & unary operators 5946 const UnaryOperator *UO = cast<UnaryOperator>(expr); 5947 expr = UO->getSubExpr(); 5948 switch (UO->getOpcode()) { 5949 case UO_AddrOf: 5950 AllowOnePastEnd++; 5951 break; 5952 case UO_Deref: 5953 AllowOnePastEnd--; 5954 break; 5955 default: 5956 return; 5957 } 5958 break; 5959 } 5960 case Stmt::ConditionalOperatorClass: { 5961 const ConditionalOperator *cond = cast<ConditionalOperator>(expr); 5962 if (const Expr *lhs = cond->getLHS()) 5963 CheckArrayAccess(lhs); 5964 if (const Expr *rhs = cond->getRHS()) 5965 CheckArrayAccess(rhs); 5966 return; 5967 } 5968 default: 5969 return; 5970 } 5971 } 5972} 5973 5974//===--- CHECK: Objective-C retain cycles ----------------------------------// 5975 5976namespace { 5977 struct RetainCycleOwner { 5978 RetainCycleOwner() : Variable(0), Indirect(false) {} 5979 VarDecl *Variable; 5980 SourceRange Range; 5981 SourceLocation Loc; 5982 bool Indirect; 5983 5984 void setLocsFrom(Expr *e) { 5985 Loc = e->getExprLoc(); 5986 Range = e->getSourceRange(); 5987 } 5988 }; 5989} 5990 5991/// Consider whether capturing the given variable can possibly lead to 5992/// a retain cycle. 5993static bool considerVariable(VarDecl *var, Expr *ref, RetainCycleOwner &owner) { 5994 // In ARC, it's captured strongly iff the variable has __strong 5995 // lifetime. In MRR, it's captured strongly if the variable is 5996 // __block and has an appropriate type. 5997 if (var->getType().getObjCLifetime() != Qualifiers::OCL_Strong) 5998 return false; 5999 6000 owner.Variable = var; 6001 if (ref) 6002 owner.setLocsFrom(ref); 6003 return true; 6004} 6005 6006static bool findRetainCycleOwner(Sema &S, Expr *e, RetainCycleOwner &owner) { 6007 while (true) { 6008 e = e->IgnoreParens(); 6009 if (CastExpr *cast = dyn_cast<CastExpr>(e)) { 6010 switch (cast->getCastKind()) { 6011 case CK_BitCast: 6012 case CK_LValueBitCast: 6013 case CK_LValueToRValue: 6014 case CK_ARCReclaimReturnedObject: 6015 e = cast->getSubExpr(); 6016 continue; 6017 6018 default: 6019 return false; 6020 } 6021 } 6022 6023 if (ObjCIvarRefExpr *ref = dyn_cast<ObjCIvarRefExpr>(e)) { 6024 ObjCIvarDecl *ivar = ref->getDecl(); 6025 if (ivar->getType().getObjCLifetime() != Qualifiers::OCL_Strong) 6026 return false; 6027 6028 // Try to find a retain cycle in the base. 6029 if (!findRetainCycleOwner(S, ref->getBase(), owner)) 6030 return false; 6031 6032 if (ref->isFreeIvar()) owner.setLocsFrom(ref); 6033 owner.Indirect = true; 6034 return true; 6035 } 6036 6037 if (DeclRefExpr *ref = dyn_cast<DeclRefExpr>(e)) { 6038 VarDecl *var = dyn_cast<VarDecl>(ref->getDecl()); 6039 if (!var) return false; 6040 return considerVariable(var, ref, owner); 6041 } 6042 6043 if (MemberExpr *member = dyn_cast<MemberExpr>(e)) { 6044 if (member->isArrow()) return false; 6045 6046 // Don't count this as an indirect ownership. 6047 e = member->getBase(); 6048 continue; 6049 } 6050 6051 if (PseudoObjectExpr *pseudo = dyn_cast<PseudoObjectExpr>(e)) { 6052 // Only pay attention to pseudo-objects on property references. 6053 ObjCPropertyRefExpr *pre 6054 = dyn_cast<ObjCPropertyRefExpr>(pseudo->getSyntacticForm() 6055 ->IgnoreParens()); 6056 if (!pre) return false; 6057 if (pre->isImplicitProperty()) return false; 6058 ObjCPropertyDecl *property = pre->getExplicitProperty(); 6059 if (!property->isRetaining() && 6060 !(property->getPropertyIvarDecl() && 6061 property->getPropertyIvarDecl()->getType() 6062 .getObjCLifetime() == Qualifiers::OCL_Strong)) 6063 return false; 6064 6065 owner.Indirect = true; 6066 if (pre->isSuperReceiver()) { 6067 owner.Variable = S.getCurMethodDecl()->getSelfDecl(); 6068 if (!owner.Variable) 6069 return false; 6070 owner.Loc = pre->getLocation(); 6071 owner.Range = pre->getSourceRange(); 6072 return true; 6073 } 6074 e = const_cast<Expr*>(cast<OpaqueValueExpr>(pre->getBase()) 6075 ->getSourceExpr()); 6076 continue; 6077 } 6078 6079 // Array ivars? 6080 6081 return false; 6082 } 6083} 6084 6085namespace { 6086 struct FindCaptureVisitor : EvaluatedExprVisitor<FindCaptureVisitor> { 6087 FindCaptureVisitor(ASTContext &Context, VarDecl *variable) 6088 : EvaluatedExprVisitor<FindCaptureVisitor>(Context), 6089 Variable(variable), Capturer(0) {} 6090 6091 VarDecl *Variable; 6092 Expr *Capturer; 6093 6094 void VisitDeclRefExpr(DeclRefExpr *ref) { 6095 if (ref->getDecl() == Variable && !Capturer) 6096 Capturer = ref; 6097 } 6098 6099 void VisitObjCIvarRefExpr(ObjCIvarRefExpr *ref) { 6100 if (Capturer) return; 6101 Visit(ref->getBase()); 6102 if (Capturer && ref->isFreeIvar()) 6103 Capturer = ref; 6104 } 6105 6106 void VisitBlockExpr(BlockExpr *block) { 6107 // Look inside nested blocks 6108 if (block->getBlockDecl()->capturesVariable(Variable)) 6109 Visit(block->getBlockDecl()->getBody()); 6110 } 6111 6112 void VisitOpaqueValueExpr(OpaqueValueExpr *OVE) { 6113 if (Capturer) return; 6114 if (OVE->getSourceExpr()) 6115 Visit(OVE->getSourceExpr()); 6116 } 6117 }; 6118} 6119 6120/// Check whether the given argument is a block which captures a 6121/// variable. 6122static Expr *findCapturingExpr(Sema &S, Expr *e, RetainCycleOwner &owner) { 6123 assert(owner.Variable && owner.Loc.isValid()); 6124 6125 e = e->IgnoreParenCasts(); 6126 6127 // Look through [^{...} copy] and Block_copy(^{...}). 6128 if (ObjCMessageExpr *ME = dyn_cast<ObjCMessageExpr>(e)) { 6129 Selector Cmd = ME->getSelector(); 6130 if (Cmd.isUnarySelector() && Cmd.getNameForSlot(0) == "copy") { 6131 e = ME->getInstanceReceiver(); 6132 if (!e) 6133 return 0; 6134 e = e->IgnoreParenCasts(); 6135 } 6136 } else if (CallExpr *CE = dyn_cast<CallExpr>(e)) { 6137 if (CE->getNumArgs() == 1) { 6138 FunctionDecl *Fn = dyn_cast_or_null<FunctionDecl>(CE->getCalleeDecl()); 6139 if (Fn) { 6140 const IdentifierInfo *FnI = Fn->getIdentifier(); 6141 if (FnI && FnI->isStr("_Block_copy")) { 6142 e = CE->getArg(0)->IgnoreParenCasts(); 6143 } 6144 } 6145 } 6146 } 6147 6148 BlockExpr *block = dyn_cast<BlockExpr>(e); 6149 if (!block || !block->getBlockDecl()->capturesVariable(owner.Variable)) 6150 return 0; 6151 6152 FindCaptureVisitor visitor(S.Context, owner.Variable); 6153 visitor.Visit(block->getBlockDecl()->getBody()); 6154 return visitor.Capturer; 6155} 6156 6157static void diagnoseRetainCycle(Sema &S, Expr *capturer, 6158 RetainCycleOwner &owner) { 6159 assert(capturer); 6160 assert(owner.Variable && owner.Loc.isValid()); 6161 6162 S.Diag(capturer->getExprLoc(), diag::warn_arc_retain_cycle) 6163 << owner.Variable << capturer->getSourceRange(); 6164 S.Diag(owner.Loc, diag::note_arc_retain_cycle_owner) 6165 << owner.Indirect << owner.Range; 6166} 6167 6168/// Check for a keyword selector that starts with the word 'add' or 6169/// 'set'. 6170static bool isSetterLikeSelector(Selector sel) { 6171 if (sel.isUnarySelector()) return false; 6172 6173 StringRef str = sel.getNameForSlot(0); 6174 while (!str.empty() && str.front() == '_') str = str.substr(1); 6175 if (str.startswith("set")) 6176 str = str.substr(3); 6177 else if (str.startswith("add")) { 6178 // Specially whitelist 'addOperationWithBlock:'. 6179 if (sel.getNumArgs() == 1 && str.startswith("addOperationWithBlock")) 6180 return false; 6181 str = str.substr(3); 6182 } 6183 else 6184 return false; 6185 6186 if (str.empty()) return true; 6187 return !isLowercase(str.front()); 6188} 6189 6190/// Check a message send to see if it's likely to cause a retain cycle. 6191void Sema::checkRetainCycles(ObjCMessageExpr *msg) { 6192 // Only check instance methods whose selector looks like a setter. 6193 if (!msg->isInstanceMessage() || !isSetterLikeSelector(msg->getSelector())) 6194 return; 6195 6196 // Try to find a variable that the receiver is strongly owned by. 6197 RetainCycleOwner owner; 6198 if (msg->getReceiverKind() == ObjCMessageExpr::Instance) { 6199 if (!findRetainCycleOwner(*this, msg->getInstanceReceiver(), owner)) 6200 return; 6201 } else { 6202 assert(msg->getReceiverKind() == ObjCMessageExpr::SuperInstance); 6203 owner.Variable = getCurMethodDecl()->getSelfDecl(); 6204 owner.Loc = msg->getSuperLoc(); 6205 owner.Range = msg->getSuperLoc(); 6206 } 6207 6208 // Check whether the receiver is captured by any of the arguments. 6209 for (unsigned i = 0, e = msg->getNumArgs(); i != e; ++i) 6210 if (Expr *capturer = findCapturingExpr(*this, msg->getArg(i), owner)) 6211 return diagnoseRetainCycle(*this, capturer, owner); 6212} 6213 6214/// Check a property assign to see if it's likely to cause a retain cycle. 6215void Sema::checkRetainCycles(Expr *receiver, Expr *argument) { 6216 RetainCycleOwner owner; 6217 if (!findRetainCycleOwner(*this, receiver, owner)) 6218 return; 6219 6220 if (Expr *capturer = findCapturingExpr(*this, argument, owner)) 6221 diagnoseRetainCycle(*this, capturer, owner); 6222} 6223 6224void Sema::checkRetainCycles(VarDecl *Var, Expr *Init) { 6225 RetainCycleOwner Owner; 6226 if (!considerVariable(Var, /*DeclRefExpr=*/0, Owner)) 6227 return; 6228 6229 // Because we don't have an expression for the variable, we have to set the 6230 // location explicitly here. 6231 Owner.Loc = Var->getLocation(); 6232 Owner.Range = Var->getSourceRange(); 6233 6234 if (Expr *Capturer = findCapturingExpr(*this, Init, Owner)) 6235 diagnoseRetainCycle(*this, Capturer, Owner); 6236} 6237 6238static bool checkUnsafeAssignLiteral(Sema &S, SourceLocation Loc, 6239 Expr *RHS, bool isProperty) { 6240 // Check if RHS is an Objective-C object literal, which also can get 6241 // immediately zapped in a weak reference. Note that we explicitly 6242 // allow ObjCStringLiterals, since those are designed to never really die. 6243 RHS = RHS->IgnoreParenImpCasts(); 6244 6245 // This enum needs to match with the 'select' in 6246 // warn_objc_arc_literal_assign (off-by-1). 6247 Sema::ObjCLiteralKind Kind = S.CheckLiteralKind(RHS); 6248 if (Kind == Sema::LK_String || Kind == Sema::LK_None) 6249 return false; 6250 6251 S.Diag(Loc, diag::warn_arc_literal_assign) 6252 << (unsigned) Kind 6253 << (isProperty ? 0 : 1) 6254 << RHS->getSourceRange(); 6255 6256 return true; 6257} 6258 6259static bool checkUnsafeAssignObject(Sema &S, SourceLocation Loc, 6260 Qualifiers::ObjCLifetime LT, 6261 Expr *RHS, bool isProperty) { 6262 // Strip off any implicit cast added to get to the one ARC-specific. 6263 while (ImplicitCastExpr *cast = dyn_cast<ImplicitCastExpr>(RHS)) { 6264 if (cast->getCastKind() == CK_ARCConsumeObject) { 6265 S.Diag(Loc, diag::warn_arc_retained_assign) 6266 << (LT == Qualifiers::OCL_ExplicitNone) 6267 << (isProperty ? 0 : 1) 6268 << RHS->getSourceRange(); 6269 return true; 6270 } 6271 RHS = cast->getSubExpr(); 6272 } 6273 6274 if (LT == Qualifiers::OCL_Weak && 6275 checkUnsafeAssignLiteral(S, Loc, RHS, isProperty)) 6276 return true; 6277 6278 return false; 6279} 6280 6281bool Sema::checkUnsafeAssigns(SourceLocation Loc, 6282 QualType LHS, Expr *RHS) { 6283 Qualifiers::ObjCLifetime LT = LHS.getObjCLifetime(); 6284 6285 if (LT != Qualifiers::OCL_Weak && LT != Qualifiers::OCL_ExplicitNone) 6286 return false; 6287 6288 if (checkUnsafeAssignObject(*this, Loc, LT, RHS, false)) 6289 return true; 6290 6291 return false; 6292} 6293 6294void Sema::checkUnsafeExprAssigns(SourceLocation Loc, 6295 Expr *LHS, Expr *RHS) { 6296 QualType LHSType; 6297 // PropertyRef on LHS type need be directly obtained from 6298 // its declaration as it has a PsuedoType. 6299 ObjCPropertyRefExpr *PRE 6300 = dyn_cast<ObjCPropertyRefExpr>(LHS->IgnoreParens()); 6301 if (PRE && !PRE->isImplicitProperty()) { 6302 const ObjCPropertyDecl *PD = PRE->getExplicitProperty(); 6303 if (PD) 6304 LHSType = PD->getType(); 6305 } 6306 6307 if (LHSType.isNull()) 6308 LHSType = LHS->getType(); 6309 6310 Qualifiers::ObjCLifetime LT = LHSType.getObjCLifetime(); 6311 6312 if (LT == Qualifiers::OCL_Weak) { 6313 DiagnosticsEngine::Level Level = 6314 Diags.getDiagnosticLevel(diag::warn_arc_repeated_use_of_weak, Loc); 6315 if (Level != DiagnosticsEngine::Ignored) 6316 getCurFunction()->markSafeWeakUse(LHS); 6317 } 6318 6319 if (checkUnsafeAssigns(Loc, LHSType, RHS)) 6320 return; 6321 6322 // FIXME. Check for other life times. 6323 if (LT != Qualifiers::OCL_None) 6324 return; 6325 6326 if (PRE) { 6327 if (PRE->isImplicitProperty()) 6328 return; 6329 const ObjCPropertyDecl *PD = PRE->getExplicitProperty(); 6330 if (!PD) 6331 return; 6332 6333 unsigned Attributes = PD->getPropertyAttributes(); 6334 if (Attributes & ObjCPropertyDecl::OBJC_PR_assign) { 6335 // when 'assign' attribute was not explicitly specified 6336 // by user, ignore it and rely on property type itself 6337 // for lifetime info. 6338 unsigned AsWrittenAttr = PD->getPropertyAttributesAsWritten(); 6339 if (!(AsWrittenAttr & ObjCPropertyDecl::OBJC_PR_assign) && 6340 LHSType->isObjCRetainableType()) 6341 return; 6342 6343 while (ImplicitCastExpr *cast = dyn_cast<ImplicitCastExpr>(RHS)) { 6344 if (cast->getCastKind() == CK_ARCConsumeObject) { 6345 Diag(Loc, diag::warn_arc_retained_property_assign) 6346 << RHS->getSourceRange(); 6347 return; 6348 } 6349 RHS = cast->getSubExpr(); 6350 } 6351 } 6352 else if (Attributes & ObjCPropertyDecl::OBJC_PR_weak) { 6353 if (checkUnsafeAssignObject(*this, Loc, Qualifiers::OCL_Weak, RHS, true)) 6354 return; 6355 } 6356 } 6357} 6358 6359//===--- CHECK: Empty statement body (-Wempty-body) ---------------------===// 6360 6361namespace { 6362bool ShouldDiagnoseEmptyStmtBody(const SourceManager &SourceMgr, 6363 SourceLocation StmtLoc, 6364 const NullStmt *Body) { 6365 // Do not warn if the body is a macro that expands to nothing, e.g: 6366 // 6367 // #define CALL(x) 6368 // if (condition) 6369 // CALL(0); 6370 // 6371 if (Body->hasLeadingEmptyMacro()) 6372 return false; 6373 6374 // Get line numbers of statement and body. 6375 bool StmtLineInvalid; 6376 unsigned StmtLine = SourceMgr.getSpellingLineNumber(StmtLoc, 6377 &StmtLineInvalid); 6378 if (StmtLineInvalid) 6379 return false; 6380 6381 bool BodyLineInvalid; 6382 unsigned BodyLine = SourceMgr.getSpellingLineNumber(Body->getSemiLoc(), 6383 &BodyLineInvalid); 6384 if (BodyLineInvalid) 6385 return false; 6386 6387 // Warn if null statement and body are on the same line. 6388 if (StmtLine != BodyLine) 6389 return false; 6390 6391 return true; 6392} 6393} // Unnamed namespace 6394 6395void Sema::DiagnoseEmptyStmtBody(SourceLocation StmtLoc, 6396 const Stmt *Body, 6397 unsigned DiagID) { 6398 // Since this is a syntactic check, don't emit diagnostic for template 6399 // instantiations, this just adds noise. 6400 if (CurrentInstantiationScope) 6401 return; 6402 6403 // The body should be a null statement. 6404 const NullStmt *NBody = dyn_cast<NullStmt>(Body); 6405 if (!NBody) 6406 return; 6407 6408 // Do the usual checks. 6409 if (!ShouldDiagnoseEmptyStmtBody(SourceMgr, StmtLoc, NBody)) 6410 return; 6411 6412 Diag(NBody->getSemiLoc(), DiagID); 6413 Diag(NBody->getSemiLoc(), diag::note_empty_body_on_separate_line); 6414} 6415 6416void Sema::DiagnoseEmptyLoopBody(const Stmt *S, 6417 const Stmt *PossibleBody) { 6418 assert(!CurrentInstantiationScope); // Ensured by caller 6419 6420 SourceLocation StmtLoc; 6421 const Stmt *Body; 6422 unsigned DiagID; 6423 if (const ForStmt *FS = dyn_cast<ForStmt>(S)) { 6424 StmtLoc = FS->getRParenLoc(); 6425 Body = FS->getBody(); 6426 DiagID = diag::warn_empty_for_body; 6427 } else if (const WhileStmt *WS = dyn_cast<WhileStmt>(S)) { 6428 StmtLoc = WS->getCond()->getSourceRange().getEnd(); 6429 Body = WS->getBody(); 6430 DiagID = diag::warn_empty_while_body; 6431 } else 6432 return; // Neither `for' nor `while'. 6433 6434 // The body should be a null statement. 6435 const NullStmt *NBody = dyn_cast<NullStmt>(Body); 6436 if (!NBody) 6437 return; 6438 6439 // Skip expensive checks if diagnostic is disabled. 6440 if (Diags.getDiagnosticLevel(DiagID, NBody->getSemiLoc()) == 6441 DiagnosticsEngine::Ignored) 6442 return; 6443 6444 // Do the usual checks. 6445 if (!ShouldDiagnoseEmptyStmtBody(SourceMgr, StmtLoc, NBody)) 6446 return; 6447 6448 // `for(...);' and `while(...);' are popular idioms, so in order to keep 6449 // noise level low, emit diagnostics only if for/while is followed by a 6450 // CompoundStmt, e.g.: 6451 // for (int i = 0; i < n; i++); 6452 // { 6453 // a(i); 6454 // } 6455 // or if for/while is followed by a statement with more indentation 6456 // than for/while itself: 6457 // for (int i = 0; i < n; i++); 6458 // a(i); 6459 bool ProbableTypo = isa<CompoundStmt>(PossibleBody); 6460 if (!ProbableTypo) { 6461 bool BodyColInvalid; 6462 unsigned BodyCol = SourceMgr.getPresumedColumnNumber( 6463 PossibleBody->getLocStart(), 6464 &BodyColInvalid); 6465 if (BodyColInvalid) 6466 return; 6467 6468 bool StmtColInvalid; 6469 unsigned StmtCol = SourceMgr.getPresumedColumnNumber( 6470 S->getLocStart(), 6471 &StmtColInvalid); 6472 if (StmtColInvalid) 6473 return; 6474 6475 if (BodyCol > StmtCol) 6476 ProbableTypo = true; 6477 } 6478 6479 if (ProbableTypo) { 6480 Diag(NBody->getSemiLoc(), DiagID); 6481 Diag(NBody->getSemiLoc(), diag::note_empty_body_on_separate_line); 6482 } 6483} 6484 6485//===--- Layout compatibility ----------------------------------------------// 6486 6487namespace { 6488 6489bool isLayoutCompatible(ASTContext &C, QualType T1, QualType T2); 6490 6491/// \brief Check if two enumeration types are layout-compatible. 6492bool isLayoutCompatible(ASTContext &C, EnumDecl *ED1, EnumDecl *ED2) { 6493 // C++11 [dcl.enum] p8: 6494 // Two enumeration types are layout-compatible if they have the same 6495 // underlying type. 6496 return ED1->isComplete() && ED2->isComplete() && 6497 C.hasSameType(ED1->getIntegerType(), ED2->getIntegerType()); 6498} 6499 6500/// \brief Check if two fields are layout-compatible. 6501bool isLayoutCompatible(ASTContext &C, FieldDecl *Field1, FieldDecl *Field2) { 6502 if (!isLayoutCompatible(C, Field1->getType(), Field2->getType())) 6503 return false; 6504 6505 if (Field1->isBitField() != Field2->isBitField()) 6506 return false; 6507 6508 if (Field1->isBitField()) { 6509 // Make sure that the bit-fields are the same length. 6510 unsigned Bits1 = Field1->getBitWidthValue(C); 6511 unsigned Bits2 = Field2->getBitWidthValue(C); 6512 6513 if (Bits1 != Bits2) 6514 return false; 6515 } 6516 6517 return true; 6518} 6519 6520/// \brief Check if two standard-layout structs are layout-compatible. 6521/// (C++11 [class.mem] p17) 6522bool isLayoutCompatibleStruct(ASTContext &C, 6523 RecordDecl *RD1, 6524 RecordDecl *RD2) { 6525 // If both records are C++ classes, check that base classes match. 6526 if (const CXXRecordDecl *D1CXX = dyn_cast<CXXRecordDecl>(RD1)) { 6527 // If one of records is a CXXRecordDecl we are in C++ mode, 6528 // thus the other one is a CXXRecordDecl, too. 6529 const CXXRecordDecl *D2CXX = cast<CXXRecordDecl>(RD2); 6530 // Check number of base classes. 6531 if (D1CXX->getNumBases() != D2CXX->getNumBases()) 6532 return false; 6533 6534 // Check the base classes. 6535 for (CXXRecordDecl::base_class_const_iterator 6536 Base1 = D1CXX->bases_begin(), 6537 BaseEnd1 = D1CXX->bases_end(), 6538 Base2 = D2CXX->bases_begin(); 6539 Base1 != BaseEnd1; 6540 ++Base1, ++Base2) { 6541 if (!isLayoutCompatible(C, Base1->getType(), Base2->getType())) 6542 return false; 6543 } 6544 } else if (const CXXRecordDecl *D2CXX = dyn_cast<CXXRecordDecl>(RD2)) { 6545 // If only RD2 is a C++ class, it should have zero base classes. 6546 if (D2CXX->getNumBases() > 0) 6547 return false; 6548 } 6549 6550 // Check the fields. 6551 RecordDecl::field_iterator Field2 = RD2->field_begin(), 6552 Field2End = RD2->field_end(), 6553 Field1 = RD1->field_begin(), 6554 Field1End = RD1->field_end(); 6555 for ( ; Field1 != Field1End && Field2 != Field2End; ++Field1, ++Field2) { 6556 if (!isLayoutCompatible(C, *Field1, *Field2)) 6557 return false; 6558 } 6559 if (Field1 != Field1End || Field2 != Field2End) 6560 return false; 6561 6562 return true; 6563} 6564 6565/// \brief Check if two standard-layout unions are layout-compatible. 6566/// (C++11 [class.mem] p18) 6567bool isLayoutCompatibleUnion(ASTContext &C, 6568 RecordDecl *RD1, 6569 RecordDecl *RD2) { 6570 llvm::SmallPtrSet<FieldDecl *, 8> UnmatchedFields; 6571 for (RecordDecl::field_iterator Field2 = RD2->field_begin(), 6572 Field2End = RD2->field_end(); 6573 Field2 != Field2End; ++Field2) { 6574 UnmatchedFields.insert(*Field2); 6575 } 6576 6577 for (RecordDecl::field_iterator Field1 = RD1->field_begin(), 6578 Field1End = RD1->field_end(); 6579 Field1 != Field1End; ++Field1) { 6580 llvm::SmallPtrSet<FieldDecl *, 8>::iterator 6581 I = UnmatchedFields.begin(), 6582 E = UnmatchedFields.end(); 6583 6584 for ( ; I != E; ++I) { 6585 if (isLayoutCompatible(C, *Field1, *I)) { 6586 bool Result = UnmatchedFields.erase(*I); 6587 (void) Result; 6588 assert(Result); 6589 break; 6590 } 6591 } 6592 if (I == E) 6593 return false; 6594 } 6595 6596 return UnmatchedFields.empty(); 6597} 6598 6599bool isLayoutCompatible(ASTContext &C, RecordDecl *RD1, RecordDecl *RD2) { 6600 if (RD1->isUnion() != RD2->isUnion()) 6601 return false; 6602 6603 if (RD1->isUnion()) 6604 return isLayoutCompatibleUnion(C, RD1, RD2); 6605 else 6606 return isLayoutCompatibleStruct(C, RD1, RD2); 6607} 6608 6609/// \brief Check if two types are layout-compatible in C++11 sense. 6610bool isLayoutCompatible(ASTContext &C, QualType T1, QualType T2) { 6611 if (T1.isNull() || T2.isNull()) 6612 return false; 6613 6614 // C++11 [basic.types] p11: 6615 // If two types T1 and T2 are the same type, then T1 and T2 are 6616 // layout-compatible types. 6617 if (C.hasSameType(T1, T2)) 6618 return true; 6619 6620 T1 = T1.getCanonicalType().getUnqualifiedType(); 6621 T2 = T2.getCanonicalType().getUnqualifiedType(); 6622 6623 const Type::TypeClass TC1 = T1->getTypeClass(); 6624 const Type::TypeClass TC2 = T2->getTypeClass(); 6625 6626 if (TC1 != TC2) 6627 return false; 6628 6629 if (TC1 == Type::Enum) { 6630 return isLayoutCompatible(C, 6631 cast<EnumType>(T1)->getDecl(), 6632 cast<EnumType>(T2)->getDecl()); 6633 } else if (TC1 == Type::Record) { 6634 if (!T1->isStandardLayoutType() || !T2->isStandardLayoutType()) 6635 return false; 6636 6637 return isLayoutCompatible(C, 6638 cast<RecordType>(T1)->getDecl(), 6639 cast<RecordType>(T2)->getDecl()); 6640 } 6641 6642 return false; 6643} 6644} 6645 6646//===--- CHECK: pointer_with_type_tag attribute: datatypes should match ----// 6647 6648namespace { 6649/// \brief Given a type tag expression find the type tag itself. 6650/// 6651/// \param TypeExpr Type tag expression, as it appears in user's code. 6652/// 6653/// \param VD Declaration of an identifier that appears in a type tag. 6654/// 6655/// \param MagicValue Type tag magic value. 6656bool FindTypeTagExpr(const Expr *TypeExpr, const ASTContext &Ctx, 6657 const ValueDecl **VD, uint64_t *MagicValue) { 6658 while(true) { 6659 if (!TypeExpr) 6660 return false; 6661 6662 TypeExpr = TypeExpr->IgnoreParenImpCasts()->IgnoreParenCasts(); 6663 6664 switch (TypeExpr->getStmtClass()) { 6665 case Stmt::UnaryOperatorClass: { 6666 const UnaryOperator *UO = cast<UnaryOperator>(TypeExpr); 6667 if (UO->getOpcode() == UO_AddrOf || UO->getOpcode() == UO_Deref) { 6668 TypeExpr = UO->getSubExpr(); 6669 continue; 6670 } 6671 return false; 6672 } 6673 6674 case Stmt::DeclRefExprClass: { 6675 const DeclRefExpr *DRE = cast<DeclRefExpr>(TypeExpr); 6676 *VD = DRE->getDecl(); 6677 return true; 6678 } 6679 6680 case Stmt::IntegerLiteralClass: { 6681 const IntegerLiteral *IL = cast<IntegerLiteral>(TypeExpr); 6682 llvm::APInt MagicValueAPInt = IL->getValue(); 6683 if (MagicValueAPInt.getActiveBits() <= 64) { 6684 *MagicValue = MagicValueAPInt.getZExtValue(); 6685 return true; 6686 } else 6687 return false; 6688 } 6689 6690 case Stmt::BinaryConditionalOperatorClass: 6691 case Stmt::ConditionalOperatorClass: { 6692 const AbstractConditionalOperator *ACO = 6693 cast<AbstractConditionalOperator>(TypeExpr); 6694 bool Result; 6695 if (ACO->getCond()->EvaluateAsBooleanCondition(Result, Ctx)) { 6696 if (Result) 6697 TypeExpr = ACO->getTrueExpr(); 6698 else 6699 TypeExpr = ACO->getFalseExpr(); 6700 continue; 6701 } 6702 return false; 6703 } 6704 6705 case Stmt::BinaryOperatorClass: { 6706 const BinaryOperator *BO = cast<BinaryOperator>(TypeExpr); 6707 if (BO->getOpcode() == BO_Comma) { 6708 TypeExpr = BO->getRHS(); 6709 continue; 6710 } 6711 return false; 6712 } 6713 6714 default: 6715 return false; 6716 } 6717 } 6718} 6719 6720/// \brief Retrieve the C type corresponding to type tag TypeExpr. 6721/// 6722/// \param TypeExpr Expression that specifies a type tag. 6723/// 6724/// \param MagicValues Registered magic values. 6725/// 6726/// \param FoundWrongKind Set to true if a type tag was found, but of a wrong 6727/// kind. 6728/// 6729/// \param TypeInfo Information about the corresponding C type. 6730/// 6731/// \returns true if the corresponding C type was found. 6732bool GetMatchingCType( 6733 const IdentifierInfo *ArgumentKind, 6734 const Expr *TypeExpr, const ASTContext &Ctx, 6735 const llvm::DenseMap<Sema::TypeTagMagicValue, 6736 Sema::TypeTagData> *MagicValues, 6737 bool &FoundWrongKind, 6738 Sema::TypeTagData &TypeInfo) { 6739 FoundWrongKind = false; 6740 6741 // Variable declaration that has type_tag_for_datatype attribute. 6742 const ValueDecl *VD = NULL; 6743 6744 uint64_t MagicValue; 6745 6746 if (!FindTypeTagExpr(TypeExpr, Ctx, &VD, &MagicValue)) 6747 return false; 6748 6749 if (VD) { 6750 for (specific_attr_iterator<TypeTagForDatatypeAttr> 6751 I = VD->specific_attr_begin<TypeTagForDatatypeAttr>(), 6752 E = VD->specific_attr_end<TypeTagForDatatypeAttr>(); 6753 I != E; ++I) { 6754 if (I->getArgumentKind() != ArgumentKind) { 6755 FoundWrongKind = true; 6756 return false; 6757 } 6758 TypeInfo.Type = I->getMatchingCType(); 6759 TypeInfo.LayoutCompatible = I->getLayoutCompatible(); 6760 TypeInfo.MustBeNull = I->getMustBeNull(); 6761 return true; 6762 } 6763 return false; 6764 } 6765 6766 if (!MagicValues) 6767 return false; 6768 6769 llvm::DenseMap<Sema::TypeTagMagicValue, 6770 Sema::TypeTagData>::const_iterator I = 6771 MagicValues->find(std::make_pair(ArgumentKind, MagicValue)); 6772 if (I == MagicValues->end()) 6773 return false; 6774 6775 TypeInfo = I->second; 6776 return true; 6777} 6778} // unnamed namespace 6779 6780void Sema::RegisterTypeTagForDatatype(const IdentifierInfo *ArgumentKind, 6781 uint64_t MagicValue, QualType Type, 6782 bool LayoutCompatible, 6783 bool MustBeNull) { 6784 if (!TypeTagForDatatypeMagicValues) 6785 TypeTagForDatatypeMagicValues.reset( 6786 new llvm::DenseMap<TypeTagMagicValue, TypeTagData>); 6787 6788 TypeTagMagicValue Magic(ArgumentKind, MagicValue); 6789 (*TypeTagForDatatypeMagicValues)[Magic] = 6790 TypeTagData(Type, LayoutCompatible, MustBeNull); 6791} 6792 6793namespace { 6794bool IsSameCharType(QualType T1, QualType T2) { 6795 const BuiltinType *BT1 = T1->getAs<BuiltinType>(); 6796 if (!BT1) 6797 return false; 6798 6799 const BuiltinType *BT2 = T2->getAs<BuiltinType>(); 6800 if (!BT2) 6801 return false; 6802 6803 BuiltinType::Kind T1Kind = BT1->getKind(); 6804 BuiltinType::Kind T2Kind = BT2->getKind(); 6805 6806 return (T1Kind == BuiltinType::SChar && T2Kind == BuiltinType::Char_S) || 6807 (T1Kind == BuiltinType::UChar && T2Kind == BuiltinType::Char_U) || 6808 (T1Kind == BuiltinType::Char_U && T2Kind == BuiltinType::UChar) || 6809 (T1Kind == BuiltinType::Char_S && T2Kind == BuiltinType::SChar); 6810} 6811} // unnamed namespace 6812 6813void Sema::CheckArgumentWithTypeTag(const ArgumentWithTypeTagAttr *Attr, 6814 const Expr * const *ExprArgs) { 6815 const IdentifierInfo *ArgumentKind = Attr->getArgumentKind(); 6816 bool IsPointerAttr = Attr->getIsPointer(); 6817 6818 const Expr *TypeTagExpr = ExprArgs[Attr->getTypeTagIdx()]; 6819 bool FoundWrongKind; 6820 TypeTagData TypeInfo; 6821 if (!GetMatchingCType(ArgumentKind, TypeTagExpr, Context, 6822 TypeTagForDatatypeMagicValues.get(), 6823 FoundWrongKind, TypeInfo)) { 6824 if (FoundWrongKind) 6825 Diag(TypeTagExpr->getExprLoc(), 6826 diag::warn_type_tag_for_datatype_wrong_kind) 6827 << TypeTagExpr->getSourceRange(); 6828 return; 6829 } 6830 6831 const Expr *ArgumentExpr = ExprArgs[Attr->getArgumentIdx()]; 6832 if (IsPointerAttr) { 6833 // Skip implicit cast of pointer to `void *' (as a function argument). 6834 if (const ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(ArgumentExpr)) 6835 if (ICE->getType()->isVoidPointerType() && 6836 ICE->getCastKind() == CK_BitCast) 6837 ArgumentExpr = ICE->getSubExpr(); 6838 } 6839 QualType ArgumentType = ArgumentExpr->getType(); 6840 6841 // Passing a `void*' pointer shouldn't trigger a warning. 6842 if (IsPointerAttr && ArgumentType->isVoidPointerType()) 6843 return; 6844 6845 if (TypeInfo.MustBeNull) { 6846 // Type tag with matching void type requires a null pointer. 6847 if (!ArgumentExpr->isNullPointerConstant(Context, 6848 Expr::NPC_ValueDependentIsNotNull)) { 6849 Diag(ArgumentExpr->getExprLoc(), 6850 diag::warn_type_safety_null_pointer_required) 6851 << ArgumentKind->getName() 6852 << ArgumentExpr->getSourceRange() 6853 << TypeTagExpr->getSourceRange(); 6854 } 6855 return; 6856 } 6857 6858 QualType RequiredType = TypeInfo.Type; 6859 if (IsPointerAttr) 6860 RequiredType = Context.getPointerType(RequiredType); 6861 6862 bool mismatch = false; 6863 if (!TypeInfo.LayoutCompatible) { 6864 mismatch = !Context.hasSameType(ArgumentType, RequiredType); 6865 6866 // C++11 [basic.fundamental] p1: 6867 // Plain char, signed char, and unsigned char are three distinct types. 6868 // 6869 // But we treat plain `char' as equivalent to `signed char' or `unsigned 6870 // char' depending on the current char signedness mode. 6871 if (mismatch) 6872 if ((IsPointerAttr && IsSameCharType(ArgumentType->getPointeeType(), 6873 RequiredType->getPointeeType())) || 6874 (!IsPointerAttr && IsSameCharType(ArgumentType, RequiredType))) 6875 mismatch = false; 6876 } else 6877 if (IsPointerAttr) 6878 mismatch = !isLayoutCompatible(Context, 6879 ArgumentType->getPointeeType(), 6880 RequiredType->getPointeeType()); 6881 else 6882 mismatch = !isLayoutCompatible(Context, ArgumentType, RequiredType); 6883 6884 if (mismatch) 6885 Diag(ArgumentExpr->getExprLoc(), diag::warn_type_safety_type_mismatch) 6886 << ArgumentType << ArgumentKind->getName() 6887 << TypeInfo.LayoutCompatible << RequiredType 6888 << ArgumentExpr->getSourceRange() 6889 << TypeTagExpr->getSourceRange(); 6890} 6891