SemaChecking.cpp revision b07d448a625d7399d871079d55491c6b10316e6f
1//===--- SemaChecking.cpp - Extra Semantic Checking -----------------------===// 2// 3// The LLVM Compiler Infrastructure 4// 5// This file is distributed under the University of Illinois Open Source 6// License. See LICENSE.TXT for details. 7// 8//===----------------------------------------------------------------------===// 9// 10// This file implements extra semantic analysis beyond what is enforced 11// by the C type system. 12// 13//===----------------------------------------------------------------------===// 14 15#include "clang/Sema/SemaInternal.h" 16#include "clang/AST/ASTContext.h" 17#include "clang/AST/CharUnits.h" 18#include "clang/AST/DeclCXX.h" 19#include "clang/AST/DeclObjC.h" 20#include "clang/AST/EvaluatedExprVisitor.h" 21#include "clang/AST/Expr.h" 22#include "clang/AST/ExprCXX.h" 23#include "clang/AST/ExprObjC.h" 24#include "clang/AST/StmtCXX.h" 25#include "clang/AST/StmtObjC.h" 26#include "clang/Analysis/Analyses/FormatString.h" 27#include "clang/Basic/CharInfo.h" 28#include "clang/Basic/TargetBuiltins.h" 29#include "clang/Basic/TargetInfo.h" 30#include "clang/Lex/Preprocessor.h" 31#include "clang/Sema/Initialization.h" 32#include "clang/Sema/Lookup.h" 33#include "clang/Sema/ScopeInfo.h" 34#include "clang/Sema/Sema.h" 35#include "llvm/ADT/BitVector.h" 36#include "llvm/ADT/STLExtras.h" 37#include "llvm/ADT/SmallString.h" 38#include "llvm/Support/ConvertUTF.h" 39#include "llvm/Support/raw_ostream.h" 40#include <limits> 41using namespace clang; 42using namespace sema; 43 44SourceLocation Sema::getLocationOfStringLiteralByte(const StringLiteral *SL, 45 unsigned ByteNo) const { 46 return SL->getLocationOfByte(ByteNo, PP.getSourceManager(), 47 PP.getLangOpts(), PP.getTargetInfo()); 48} 49 50/// Checks that a call expression's argument count is the desired number. 51/// This is useful when doing custom type-checking. Returns true on error. 52static bool checkArgCount(Sema &S, CallExpr *call, unsigned desiredArgCount) { 53 unsigned argCount = call->getNumArgs(); 54 if (argCount == desiredArgCount) return false; 55 56 if (argCount < desiredArgCount) 57 return S.Diag(call->getLocEnd(), diag::err_typecheck_call_too_few_args) 58 << 0 /*function call*/ << desiredArgCount << argCount 59 << call->getSourceRange(); 60 61 // Highlight all the excess arguments. 62 SourceRange range(call->getArg(desiredArgCount)->getLocStart(), 63 call->getArg(argCount - 1)->getLocEnd()); 64 65 return S.Diag(range.getBegin(), diag::err_typecheck_call_too_many_args) 66 << 0 /*function call*/ << desiredArgCount << argCount 67 << call->getArg(1)->getSourceRange(); 68} 69 70/// Check that the first argument to __builtin_annotation is an integer 71/// and the second argument is a non-wide string literal. 72static bool SemaBuiltinAnnotation(Sema &S, CallExpr *TheCall) { 73 if (checkArgCount(S, TheCall, 2)) 74 return true; 75 76 // First argument should be an integer. 77 Expr *ValArg = TheCall->getArg(0); 78 QualType Ty = ValArg->getType(); 79 if (!Ty->isIntegerType()) { 80 S.Diag(ValArg->getLocStart(), diag::err_builtin_annotation_first_arg) 81 << ValArg->getSourceRange(); 82 return true; 83 } 84 85 // Second argument should be a constant string. 86 Expr *StrArg = TheCall->getArg(1)->IgnoreParenCasts(); 87 StringLiteral *Literal = dyn_cast<StringLiteral>(StrArg); 88 if (!Literal || !Literal->isAscii()) { 89 S.Diag(StrArg->getLocStart(), diag::err_builtin_annotation_second_arg) 90 << StrArg->getSourceRange(); 91 return true; 92 } 93 94 TheCall->setType(Ty); 95 return false; 96} 97 98ExprResult 99Sema::CheckBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) { 100 ExprResult TheCallResult(Owned(TheCall)); 101 102 // Find out if any arguments are required to be integer constant expressions. 103 unsigned ICEArguments = 0; 104 ASTContext::GetBuiltinTypeError Error; 105 Context.GetBuiltinType(BuiltinID, Error, &ICEArguments); 106 if (Error != ASTContext::GE_None) 107 ICEArguments = 0; // Don't diagnose previously diagnosed errors. 108 109 // If any arguments are required to be ICE's, check and diagnose. 110 for (unsigned ArgNo = 0; ICEArguments != 0; ++ArgNo) { 111 // Skip arguments not required to be ICE's. 112 if ((ICEArguments & (1 << ArgNo)) == 0) continue; 113 114 llvm::APSInt Result; 115 if (SemaBuiltinConstantArg(TheCall, ArgNo, Result)) 116 return true; 117 ICEArguments &= ~(1 << ArgNo); 118 } 119 120 switch (BuiltinID) { 121 case Builtin::BI__builtin___CFStringMakeConstantString: 122 assert(TheCall->getNumArgs() == 1 && 123 "Wrong # arguments to builtin CFStringMakeConstantString"); 124 if (CheckObjCString(TheCall->getArg(0))) 125 return ExprError(); 126 break; 127 case Builtin::BI__builtin_stdarg_start: 128 case Builtin::BI__builtin_va_start: 129 if (SemaBuiltinVAStart(TheCall)) 130 return ExprError(); 131 break; 132 case Builtin::BI__builtin_isgreater: 133 case Builtin::BI__builtin_isgreaterequal: 134 case Builtin::BI__builtin_isless: 135 case Builtin::BI__builtin_islessequal: 136 case Builtin::BI__builtin_islessgreater: 137 case Builtin::BI__builtin_isunordered: 138 if (SemaBuiltinUnorderedCompare(TheCall)) 139 return ExprError(); 140 break; 141 case Builtin::BI__builtin_fpclassify: 142 if (SemaBuiltinFPClassification(TheCall, 6)) 143 return ExprError(); 144 break; 145 case Builtin::BI__builtin_isfinite: 146 case Builtin::BI__builtin_isinf: 147 case Builtin::BI__builtin_isinf_sign: 148 case Builtin::BI__builtin_isnan: 149 case Builtin::BI__builtin_isnormal: 150 if (SemaBuiltinFPClassification(TheCall, 1)) 151 return ExprError(); 152 break; 153 case Builtin::BI__builtin_shufflevector: 154 return SemaBuiltinShuffleVector(TheCall); 155 // TheCall will be freed by the smart pointer here, but that's fine, since 156 // SemaBuiltinShuffleVector guts it, but then doesn't release it. 157 case Builtin::BI__builtin_prefetch: 158 if (SemaBuiltinPrefetch(TheCall)) 159 return ExprError(); 160 break; 161 case Builtin::BI__builtin_object_size: 162 if (SemaBuiltinObjectSize(TheCall)) 163 return ExprError(); 164 break; 165 case Builtin::BI__builtin_longjmp: 166 if (SemaBuiltinLongjmp(TheCall)) 167 return ExprError(); 168 break; 169 170 case Builtin::BI__builtin_classify_type: 171 if (checkArgCount(*this, TheCall, 1)) return true; 172 TheCall->setType(Context.IntTy); 173 break; 174 case Builtin::BI__builtin_constant_p: 175 if (checkArgCount(*this, TheCall, 1)) return true; 176 TheCall->setType(Context.IntTy); 177 break; 178 case Builtin::BI__sync_fetch_and_add: 179 case Builtin::BI__sync_fetch_and_add_1: 180 case Builtin::BI__sync_fetch_and_add_2: 181 case Builtin::BI__sync_fetch_and_add_4: 182 case Builtin::BI__sync_fetch_and_add_8: 183 case Builtin::BI__sync_fetch_and_add_16: 184 case Builtin::BI__sync_fetch_and_sub: 185 case Builtin::BI__sync_fetch_and_sub_1: 186 case Builtin::BI__sync_fetch_and_sub_2: 187 case Builtin::BI__sync_fetch_and_sub_4: 188 case Builtin::BI__sync_fetch_and_sub_8: 189 case Builtin::BI__sync_fetch_and_sub_16: 190 case Builtin::BI__sync_fetch_and_or: 191 case Builtin::BI__sync_fetch_and_or_1: 192 case Builtin::BI__sync_fetch_and_or_2: 193 case Builtin::BI__sync_fetch_and_or_4: 194 case Builtin::BI__sync_fetch_and_or_8: 195 case Builtin::BI__sync_fetch_and_or_16: 196 case Builtin::BI__sync_fetch_and_and: 197 case Builtin::BI__sync_fetch_and_and_1: 198 case Builtin::BI__sync_fetch_and_and_2: 199 case Builtin::BI__sync_fetch_and_and_4: 200 case Builtin::BI__sync_fetch_and_and_8: 201 case Builtin::BI__sync_fetch_and_and_16: 202 case Builtin::BI__sync_fetch_and_xor: 203 case Builtin::BI__sync_fetch_and_xor_1: 204 case Builtin::BI__sync_fetch_and_xor_2: 205 case Builtin::BI__sync_fetch_and_xor_4: 206 case Builtin::BI__sync_fetch_and_xor_8: 207 case Builtin::BI__sync_fetch_and_xor_16: 208 case Builtin::BI__sync_add_and_fetch: 209 case Builtin::BI__sync_add_and_fetch_1: 210 case Builtin::BI__sync_add_and_fetch_2: 211 case Builtin::BI__sync_add_and_fetch_4: 212 case Builtin::BI__sync_add_and_fetch_8: 213 case Builtin::BI__sync_add_and_fetch_16: 214 case Builtin::BI__sync_sub_and_fetch: 215 case Builtin::BI__sync_sub_and_fetch_1: 216 case Builtin::BI__sync_sub_and_fetch_2: 217 case Builtin::BI__sync_sub_and_fetch_4: 218 case Builtin::BI__sync_sub_and_fetch_8: 219 case Builtin::BI__sync_sub_and_fetch_16: 220 case Builtin::BI__sync_and_and_fetch: 221 case Builtin::BI__sync_and_and_fetch_1: 222 case Builtin::BI__sync_and_and_fetch_2: 223 case Builtin::BI__sync_and_and_fetch_4: 224 case Builtin::BI__sync_and_and_fetch_8: 225 case Builtin::BI__sync_and_and_fetch_16: 226 case Builtin::BI__sync_or_and_fetch: 227 case Builtin::BI__sync_or_and_fetch_1: 228 case Builtin::BI__sync_or_and_fetch_2: 229 case Builtin::BI__sync_or_and_fetch_4: 230 case Builtin::BI__sync_or_and_fetch_8: 231 case Builtin::BI__sync_or_and_fetch_16: 232 case Builtin::BI__sync_xor_and_fetch: 233 case Builtin::BI__sync_xor_and_fetch_1: 234 case Builtin::BI__sync_xor_and_fetch_2: 235 case Builtin::BI__sync_xor_and_fetch_4: 236 case Builtin::BI__sync_xor_and_fetch_8: 237 case Builtin::BI__sync_xor_and_fetch_16: 238 case Builtin::BI__sync_val_compare_and_swap: 239 case Builtin::BI__sync_val_compare_and_swap_1: 240 case Builtin::BI__sync_val_compare_and_swap_2: 241 case Builtin::BI__sync_val_compare_and_swap_4: 242 case Builtin::BI__sync_val_compare_and_swap_8: 243 case Builtin::BI__sync_val_compare_and_swap_16: 244 case Builtin::BI__sync_bool_compare_and_swap: 245 case Builtin::BI__sync_bool_compare_and_swap_1: 246 case Builtin::BI__sync_bool_compare_and_swap_2: 247 case Builtin::BI__sync_bool_compare_and_swap_4: 248 case Builtin::BI__sync_bool_compare_and_swap_8: 249 case Builtin::BI__sync_bool_compare_and_swap_16: 250 case Builtin::BI__sync_lock_test_and_set: 251 case Builtin::BI__sync_lock_test_and_set_1: 252 case Builtin::BI__sync_lock_test_and_set_2: 253 case Builtin::BI__sync_lock_test_and_set_4: 254 case Builtin::BI__sync_lock_test_and_set_8: 255 case Builtin::BI__sync_lock_test_and_set_16: 256 case Builtin::BI__sync_lock_release: 257 case Builtin::BI__sync_lock_release_1: 258 case Builtin::BI__sync_lock_release_2: 259 case Builtin::BI__sync_lock_release_4: 260 case Builtin::BI__sync_lock_release_8: 261 case Builtin::BI__sync_lock_release_16: 262 case Builtin::BI__sync_swap: 263 case Builtin::BI__sync_swap_1: 264 case Builtin::BI__sync_swap_2: 265 case Builtin::BI__sync_swap_4: 266 case Builtin::BI__sync_swap_8: 267 case Builtin::BI__sync_swap_16: 268 return SemaBuiltinAtomicOverloaded(TheCallResult); 269#define BUILTIN(ID, TYPE, ATTRS) 270#define ATOMIC_BUILTIN(ID, TYPE, ATTRS) \ 271 case Builtin::BI##ID: \ 272 return SemaAtomicOpsOverloaded(TheCallResult, AtomicExpr::AO##ID); 273#include "clang/Basic/Builtins.def" 274 case Builtin::BI__builtin_annotation: 275 if (SemaBuiltinAnnotation(*this, TheCall)) 276 return ExprError(); 277 break; 278 } 279 280 // Since the target specific builtins for each arch overlap, only check those 281 // of the arch we are compiling for. 282 if (BuiltinID >= Builtin::FirstTSBuiltin) { 283 switch (Context.getTargetInfo().getTriple().getArch()) { 284 case llvm::Triple::arm: 285 case llvm::Triple::thumb: 286 if (CheckARMBuiltinFunctionCall(BuiltinID, TheCall)) 287 return ExprError(); 288 break; 289 case llvm::Triple::mips: 290 case llvm::Triple::mipsel: 291 case llvm::Triple::mips64: 292 case llvm::Triple::mips64el: 293 if (CheckMipsBuiltinFunctionCall(BuiltinID, TheCall)) 294 return ExprError(); 295 break; 296 default: 297 break; 298 } 299 } 300 301 return TheCallResult; 302} 303 304// Get the valid immediate range for the specified NEON type code. 305static unsigned RFT(unsigned t, bool shift = false) { 306 NeonTypeFlags Type(t); 307 int IsQuad = Type.isQuad(); 308 switch (Type.getEltType()) { 309 case NeonTypeFlags::Int8: 310 case NeonTypeFlags::Poly8: 311 return shift ? 7 : (8 << IsQuad) - 1; 312 case NeonTypeFlags::Int16: 313 case NeonTypeFlags::Poly16: 314 return shift ? 15 : (4 << IsQuad) - 1; 315 case NeonTypeFlags::Int32: 316 return shift ? 31 : (2 << IsQuad) - 1; 317 case NeonTypeFlags::Int64: 318 return shift ? 63 : (1 << IsQuad) - 1; 319 case NeonTypeFlags::Float16: 320 assert(!shift && "cannot shift float types!"); 321 return (4 << IsQuad) - 1; 322 case NeonTypeFlags::Float32: 323 assert(!shift && "cannot shift float types!"); 324 return (2 << IsQuad) - 1; 325 } 326 llvm_unreachable("Invalid NeonTypeFlag!"); 327} 328 329/// getNeonEltType - Return the QualType corresponding to the elements of 330/// the vector type specified by the NeonTypeFlags. This is used to check 331/// the pointer arguments for Neon load/store intrinsics. 332static QualType getNeonEltType(NeonTypeFlags Flags, ASTContext &Context) { 333 switch (Flags.getEltType()) { 334 case NeonTypeFlags::Int8: 335 return Flags.isUnsigned() ? Context.UnsignedCharTy : Context.SignedCharTy; 336 case NeonTypeFlags::Int16: 337 return Flags.isUnsigned() ? Context.UnsignedShortTy : Context.ShortTy; 338 case NeonTypeFlags::Int32: 339 return Flags.isUnsigned() ? Context.UnsignedIntTy : Context.IntTy; 340 case NeonTypeFlags::Int64: 341 return Flags.isUnsigned() ? Context.UnsignedLongLongTy : Context.LongLongTy; 342 case NeonTypeFlags::Poly8: 343 return Context.SignedCharTy; 344 case NeonTypeFlags::Poly16: 345 return Context.ShortTy; 346 case NeonTypeFlags::Float16: 347 return Context.UnsignedShortTy; 348 case NeonTypeFlags::Float32: 349 return Context.FloatTy; 350 } 351 llvm_unreachable("Invalid NeonTypeFlag!"); 352} 353 354bool Sema::CheckARMBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) { 355 llvm::APSInt Result; 356 357 uint64_t mask = 0; 358 unsigned TV = 0; 359 int PtrArgNum = -1; 360 bool HasConstPtr = false; 361 switch (BuiltinID) { 362#define GET_NEON_OVERLOAD_CHECK 363#include "clang/Basic/arm_neon.inc" 364#undef GET_NEON_OVERLOAD_CHECK 365 } 366 367 // For NEON intrinsics which are overloaded on vector element type, validate 368 // the immediate which specifies which variant to emit. 369 unsigned ImmArg = TheCall->getNumArgs()-1; 370 if (mask) { 371 if (SemaBuiltinConstantArg(TheCall, ImmArg, Result)) 372 return true; 373 374 TV = Result.getLimitedValue(64); 375 if ((TV > 63) || (mask & (1ULL << TV)) == 0) 376 return Diag(TheCall->getLocStart(), diag::err_invalid_neon_type_code) 377 << TheCall->getArg(ImmArg)->getSourceRange(); 378 } 379 380 if (PtrArgNum >= 0) { 381 // Check that pointer arguments have the specified type. 382 Expr *Arg = TheCall->getArg(PtrArgNum); 383 if (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(Arg)) 384 Arg = ICE->getSubExpr(); 385 ExprResult RHS = DefaultFunctionArrayLvalueConversion(Arg); 386 QualType RHSTy = RHS.get()->getType(); 387 QualType EltTy = getNeonEltType(NeonTypeFlags(TV), Context); 388 if (HasConstPtr) 389 EltTy = EltTy.withConst(); 390 QualType LHSTy = Context.getPointerType(EltTy); 391 AssignConvertType ConvTy; 392 ConvTy = CheckSingleAssignmentConstraints(LHSTy, RHS); 393 if (RHS.isInvalid()) 394 return true; 395 if (DiagnoseAssignmentResult(ConvTy, Arg->getLocStart(), LHSTy, RHSTy, 396 RHS.get(), AA_Assigning)) 397 return true; 398 } 399 400 // For NEON intrinsics which take an immediate value as part of the 401 // instruction, range check them here. 402 unsigned i = 0, l = 0, u = 0; 403 switch (BuiltinID) { 404 default: return false; 405 case ARM::BI__builtin_arm_ssat: i = 1; l = 1; u = 31; break; 406 case ARM::BI__builtin_arm_usat: i = 1; u = 31; break; 407 case ARM::BI__builtin_arm_vcvtr_f: 408 case ARM::BI__builtin_arm_vcvtr_d: i = 1; u = 1; break; 409#define GET_NEON_IMMEDIATE_CHECK 410#include "clang/Basic/arm_neon.inc" 411#undef GET_NEON_IMMEDIATE_CHECK 412 }; 413 414 // We can't check the value of a dependent argument. 415 if (TheCall->getArg(i)->isTypeDependent() || 416 TheCall->getArg(i)->isValueDependent()) 417 return false; 418 419 // Check that the immediate argument is actually a constant. 420 if (SemaBuiltinConstantArg(TheCall, i, Result)) 421 return true; 422 423 // Range check against the upper/lower values for this isntruction. 424 unsigned Val = Result.getZExtValue(); 425 if (Val < l || Val > (u + l)) 426 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 427 << l << u+l << TheCall->getArg(i)->getSourceRange(); 428 429 // FIXME: VFP Intrinsics should error if VFP not present. 430 return false; 431} 432 433bool Sema::CheckMipsBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) { 434 unsigned i = 0, l = 0, u = 0; 435 switch (BuiltinID) { 436 default: return false; 437 case Mips::BI__builtin_mips_wrdsp: i = 1; l = 0; u = 63; break; 438 case Mips::BI__builtin_mips_rddsp: i = 0; l = 0; u = 63; break; 439 case Mips::BI__builtin_mips_append: i = 2; l = 0; u = 31; break; 440 case Mips::BI__builtin_mips_balign: i = 2; l = 0; u = 3; break; 441 case Mips::BI__builtin_mips_precr_sra_ph_w: i = 2; l = 0; u = 31; break; 442 case Mips::BI__builtin_mips_precr_sra_r_ph_w: i = 2; l = 0; u = 31; break; 443 case Mips::BI__builtin_mips_prepend: i = 2; l = 0; u = 31; break; 444 }; 445 446 // We can't check the value of a dependent argument. 447 if (TheCall->getArg(i)->isTypeDependent() || 448 TheCall->getArg(i)->isValueDependent()) 449 return false; 450 451 // Check that the immediate argument is actually a constant. 452 llvm::APSInt Result; 453 if (SemaBuiltinConstantArg(TheCall, i, Result)) 454 return true; 455 456 // Range check against the upper/lower values for this instruction. 457 unsigned Val = Result.getZExtValue(); 458 if (Val < l || Val > u) 459 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 460 << l << u << TheCall->getArg(i)->getSourceRange(); 461 462 return false; 463} 464 465/// Given a FunctionDecl's FormatAttr, attempts to populate the FomatStringInfo 466/// parameter with the FormatAttr's correct format_idx and firstDataArg. 467/// Returns true when the format fits the function and the FormatStringInfo has 468/// been populated. 469bool Sema::getFormatStringInfo(const FormatAttr *Format, bool IsCXXMember, 470 FormatStringInfo *FSI) { 471 FSI->HasVAListArg = Format->getFirstArg() == 0; 472 FSI->FormatIdx = Format->getFormatIdx() - 1; 473 FSI->FirstDataArg = FSI->HasVAListArg ? 0 : Format->getFirstArg() - 1; 474 475 // The way the format attribute works in GCC, the implicit this argument 476 // of member functions is counted. However, it doesn't appear in our own 477 // lists, so decrement format_idx in that case. 478 if (IsCXXMember) { 479 if(FSI->FormatIdx == 0) 480 return false; 481 --FSI->FormatIdx; 482 if (FSI->FirstDataArg != 0) 483 --FSI->FirstDataArg; 484 } 485 return true; 486} 487 488/// Handles the checks for format strings, non-POD arguments to vararg 489/// functions, and NULL arguments passed to non-NULL parameters. 490void Sema::checkCall(NamedDecl *FDecl, 491 ArrayRef<const Expr *> Args, 492 unsigned NumProtoArgs, 493 bool IsMemberFunction, 494 SourceLocation Loc, 495 SourceRange Range, 496 VariadicCallType CallType) { 497 if (CurContext->isDependentContext()) 498 return; 499 500 // Printf and scanf checking. 501 bool HandledFormatString = false; 502 for (specific_attr_iterator<FormatAttr> 503 I = FDecl->specific_attr_begin<FormatAttr>(), 504 E = FDecl->specific_attr_end<FormatAttr>(); I != E ; ++I) 505 if (CheckFormatArguments(*I, Args, IsMemberFunction, CallType, Loc, Range)) 506 HandledFormatString = true; 507 508 // Refuse POD arguments that weren't caught by the format string 509 // checks above. 510 if (!HandledFormatString && CallType != VariadicDoesNotApply) 511 for (unsigned ArgIdx = NumProtoArgs; ArgIdx < Args.size(); ++ArgIdx) { 512 // Args[ArgIdx] can be null in malformed code. 513 if (const Expr *Arg = Args[ArgIdx]) 514 variadicArgumentPODCheck(Arg, CallType); 515 } 516 517 for (specific_attr_iterator<NonNullAttr> 518 I = FDecl->specific_attr_begin<NonNullAttr>(), 519 E = FDecl->specific_attr_end<NonNullAttr>(); I != E; ++I) 520 CheckNonNullArguments(*I, Args.data(), Loc); 521 522 // Type safety checking. 523 for (specific_attr_iterator<ArgumentWithTypeTagAttr> 524 i = FDecl->specific_attr_begin<ArgumentWithTypeTagAttr>(), 525 e = FDecl->specific_attr_end<ArgumentWithTypeTagAttr>(); i != e; ++i) { 526 CheckArgumentWithTypeTag(*i, Args.data()); 527 } 528} 529 530/// CheckConstructorCall - Check a constructor call for correctness and safety 531/// properties not enforced by the C type system. 532void Sema::CheckConstructorCall(FunctionDecl *FDecl, 533 ArrayRef<const Expr *> Args, 534 const FunctionProtoType *Proto, 535 SourceLocation Loc) { 536 VariadicCallType CallType = 537 Proto->isVariadic() ? VariadicConstructor : VariadicDoesNotApply; 538 checkCall(FDecl, Args, Proto->getNumArgs(), 539 /*IsMemberFunction=*/true, Loc, SourceRange(), CallType); 540} 541 542/// CheckFunctionCall - Check a direct function call for various correctness 543/// and safety properties not strictly enforced by the C type system. 544bool Sema::CheckFunctionCall(FunctionDecl *FDecl, CallExpr *TheCall, 545 const FunctionProtoType *Proto) { 546 bool IsMemberOperatorCall = isa<CXXOperatorCallExpr>(TheCall) && 547 isa<CXXMethodDecl>(FDecl); 548 bool IsMemberFunction = isa<CXXMemberCallExpr>(TheCall) || 549 IsMemberOperatorCall; 550 VariadicCallType CallType = getVariadicCallType(FDecl, Proto, 551 TheCall->getCallee()); 552 unsigned NumProtoArgs = Proto ? Proto->getNumArgs() : 0; 553 Expr** Args = TheCall->getArgs(); 554 unsigned NumArgs = TheCall->getNumArgs(); 555 if (IsMemberOperatorCall) { 556 // If this is a call to a member operator, hide the first argument 557 // from checkCall. 558 // FIXME: Our choice of AST representation here is less than ideal. 559 ++Args; 560 --NumArgs; 561 } 562 checkCall(FDecl, llvm::makeArrayRef<const Expr *>(Args, NumArgs), 563 NumProtoArgs, 564 IsMemberFunction, TheCall->getRParenLoc(), 565 TheCall->getCallee()->getSourceRange(), CallType); 566 567 IdentifierInfo *FnInfo = FDecl->getIdentifier(); 568 // None of the checks below are needed for functions that don't have 569 // simple names (e.g., C++ conversion functions). 570 if (!FnInfo) 571 return false; 572 573 unsigned CMId = FDecl->getMemoryFunctionKind(); 574 if (CMId == 0) 575 return false; 576 577 // Handle memory setting and copying functions. 578 if (CMId == Builtin::BIstrlcpy || CMId == Builtin::BIstrlcat) 579 CheckStrlcpycatArguments(TheCall, FnInfo); 580 else if (CMId == Builtin::BIstrncat) 581 CheckStrncatArguments(TheCall, FnInfo); 582 else 583 CheckMemaccessArguments(TheCall, CMId, FnInfo); 584 585 return false; 586} 587 588bool Sema::CheckObjCMethodCall(ObjCMethodDecl *Method, SourceLocation lbrac, 589 ArrayRef<const Expr *> Args) { 590 VariadicCallType CallType = 591 Method->isVariadic() ? VariadicMethod : VariadicDoesNotApply; 592 593 checkCall(Method, Args, Method->param_size(), 594 /*IsMemberFunction=*/false, 595 lbrac, Method->getSourceRange(), CallType); 596 597 return false; 598} 599 600bool Sema::CheckBlockCall(NamedDecl *NDecl, CallExpr *TheCall, 601 const FunctionProtoType *Proto) { 602 const VarDecl *V = dyn_cast<VarDecl>(NDecl); 603 if (!V) 604 return false; 605 606 QualType Ty = V->getType(); 607 if (!Ty->isBlockPointerType()) 608 return false; 609 610 VariadicCallType CallType = 611 Proto && Proto->isVariadic() ? VariadicBlock : VariadicDoesNotApply ; 612 unsigned NumProtoArgs = Proto ? Proto->getNumArgs() : 0; 613 614 checkCall(NDecl, 615 llvm::makeArrayRef<const Expr *>(TheCall->getArgs(), 616 TheCall->getNumArgs()), 617 NumProtoArgs, /*IsMemberFunction=*/false, 618 TheCall->getRParenLoc(), 619 TheCall->getCallee()->getSourceRange(), CallType); 620 621 return false; 622} 623 624ExprResult Sema::SemaAtomicOpsOverloaded(ExprResult TheCallResult, 625 AtomicExpr::AtomicOp Op) { 626 CallExpr *TheCall = cast<CallExpr>(TheCallResult.get()); 627 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 628 629 // All these operations take one of the following forms: 630 enum { 631 // C __c11_atomic_init(A *, C) 632 Init, 633 // C __c11_atomic_load(A *, int) 634 Load, 635 // void __atomic_load(A *, CP, int) 636 Copy, 637 // C __c11_atomic_add(A *, M, int) 638 Arithmetic, 639 // C __atomic_exchange_n(A *, CP, int) 640 Xchg, 641 // void __atomic_exchange(A *, C *, CP, int) 642 GNUXchg, 643 // bool __c11_atomic_compare_exchange_strong(A *, C *, CP, int, int) 644 C11CmpXchg, 645 // bool __atomic_compare_exchange(A *, C *, CP, bool, int, int) 646 GNUCmpXchg 647 } Form = Init; 648 const unsigned NumArgs[] = { 2, 2, 3, 3, 3, 4, 5, 6 }; 649 const unsigned NumVals[] = { 1, 0, 1, 1, 1, 2, 2, 3 }; 650 // where: 651 // C is an appropriate type, 652 // A is volatile _Atomic(C) for __c11 builtins and is C for GNU builtins, 653 // CP is C for __c11 builtins and GNU _n builtins and is C * otherwise, 654 // M is C if C is an integer, and ptrdiff_t if C is a pointer, and 655 // the int parameters are for orderings. 656 657 assert(AtomicExpr::AO__c11_atomic_init == 0 && 658 AtomicExpr::AO__c11_atomic_fetch_xor + 1 == AtomicExpr::AO__atomic_load 659 && "need to update code for modified C11 atomics"); 660 bool IsC11 = Op >= AtomicExpr::AO__c11_atomic_init && 661 Op <= AtomicExpr::AO__c11_atomic_fetch_xor; 662 bool IsN = Op == AtomicExpr::AO__atomic_load_n || 663 Op == AtomicExpr::AO__atomic_store_n || 664 Op == AtomicExpr::AO__atomic_exchange_n || 665 Op == AtomicExpr::AO__atomic_compare_exchange_n; 666 bool IsAddSub = false; 667 668 switch (Op) { 669 case AtomicExpr::AO__c11_atomic_init: 670 Form = Init; 671 break; 672 673 case AtomicExpr::AO__c11_atomic_load: 674 case AtomicExpr::AO__atomic_load_n: 675 Form = Load; 676 break; 677 678 case AtomicExpr::AO__c11_atomic_store: 679 case AtomicExpr::AO__atomic_load: 680 case AtomicExpr::AO__atomic_store: 681 case AtomicExpr::AO__atomic_store_n: 682 Form = Copy; 683 break; 684 685 case AtomicExpr::AO__c11_atomic_fetch_add: 686 case AtomicExpr::AO__c11_atomic_fetch_sub: 687 case AtomicExpr::AO__atomic_fetch_add: 688 case AtomicExpr::AO__atomic_fetch_sub: 689 case AtomicExpr::AO__atomic_add_fetch: 690 case AtomicExpr::AO__atomic_sub_fetch: 691 IsAddSub = true; 692 // Fall through. 693 case AtomicExpr::AO__c11_atomic_fetch_and: 694 case AtomicExpr::AO__c11_atomic_fetch_or: 695 case AtomicExpr::AO__c11_atomic_fetch_xor: 696 case AtomicExpr::AO__atomic_fetch_and: 697 case AtomicExpr::AO__atomic_fetch_or: 698 case AtomicExpr::AO__atomic_fetch_xor: 699 case AtomicExpr::AO__atomic_fetch_nand: 700 case AtomicExpr::AO__atomic_and_fetch: 701 case AtomicExpr::AO__atomic_or_fetch: 702 case AtomicExpr::AO__atomic_xor_fetch: 703 case AtomicExpr::AO__atomic_nand_fetch: 704 Form = Arithmetic; 705 break; 706 707 case AtomicExpr::AO__c11_atomic_exchange: 708 case AtomicExpr::AO__atomic_exchange_n: 709 Form = Xchg; 710 break; 711 712 case AtomicExpr::AO__atomic_exchange: 713 Form = GNUXchg; 714 break; 715 716 case AtomicExpr::AO__c11_atomic_compare_exchange_strong: 717 case AtomicExpr::AO__c11_atomic_compare_exchange_weak: 718 Form = C11CmpXchg; 719 break; 720 721 case AtomicExpr::AO__atomic_compare_exchange: 722 case AtomicExpr::AO__atomic_compare_exchange_n: 723 Form = GNUCmpXchg; 724 break; 725 } 726 727 // Check we have the right number of arguments. 728 if (TheCall->getNumArgs() < NumArgs[Form]) { 729 Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 730 << 0 << NumArgs[Form] << TheCall->getNumArgs() 731 << TheCall->getCallee()->getSourceRange(); 732 return ExprError(); 733 } else if (TheCall->getNumArgs() > NumArgs[Form]) { 734 Diag(TheCall->getArg(NumArgs[Form])->getLocStart(), 735 diag::err_typecheck_call_too_many_args) 736 << 0 << NumArgs[Form] << TheCall->getNumArgs() 737 << TheCall->getCallee()->getSourceRange(); 738 return ExprError(); 739 } 740 741 // Inspect the first argument of the atomic operation. 742 Expr *Ptr = TheCall->getArg(0); 743 Ptr = DefaultFunctionArrayLvalueConversion(Ptr).get(); 744 const PointerType *pointerType = Ptr->getType()->getAs<PointerType>(); 745 if (!pointerType) { 746 Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer) 747 << Ptr->getType() << Ptr->getSourceRange(); 748 return ExprError(); 749 } 750 751 // For a __c11 builtin, this should be a pointer to an _Atomic type. 752 QualType AtomTy = pointerType->getPointeeType(); // 'A' 753 QualType ValType = AtomTy; // 'C' 754 if (IsC11) { 755 if (!AtomTy->isAtomicType()) { 756 Diag(DRE->getLocStart(), diag::err_atomic_op_needs_atomic) 757 << Ptr->getType() << Ptr->getSourceRange(); 758 return ExprError(); 759 } 760 if (AtomTy.isConstQualified()) { 761 Diag(DRE->getLocStart(), diag::err_atomic_op_needs_non_const_atomic) 762 << Ptr->getType() << Ptr->getSourceRange(); 763 return ExprError(); 764 } 765 ValType = AtomTy->getAs<AtomicType>()->getValueType(); 766 } 767 768 // For an arithmetic operation, the implied arithmetic must be well-formed. 769 if (Form == Arithmetic) { 770 // gcc does not enforce these rules for GNU atomics, but we do so for sanity. 771 if (IsAddSub && !ValType->isIntegerType() && !ValType->isPointerType()) { 772 Diag(DRE->getLocStart(), diag::err_atomic_op_needs_atomic_int_or_ptr) 773 << IsC11 << Ptr->getType() << Ptr->getSourceRange(); 774 return ExprError(); 775 } 776 if (!IsAddSub && !ValType->isIntegerType()) { 777 Diag(DRE->getLocStart(), diag::err_atomic_op_bitwise_needs_atomic_int) 778 << IsC11 << Ptr->getType() << Ptr->getSourceRange(); 779 return ExprError(); 780 } 781 } else if (IsN && !ValType->isIntegerType() && !ValType->isPointerType()) { 782 // For __atomic_*_n operations, the value type must be a scalar integral or 783 // pointer type which is 1, 2, 4, 8 or 16 bytes in length. 784 Diag(DRE->getLocStart(), diag::err_atomic_op_needs_atomic_int_or_ptr) 785 << IsC11 << Ptr->getType() << Ptr->getSourceRange(); 786 return ExprError(); 787 } 788 789 if (!IsC11 && !AtomTy.isTriviallyCopyableType(Context)) { 790 // For GNU atomics, require a trivially-copyable type. This is not part of 791 // the GNU atomics specification, but we enforce it for sanity. 792 Diag(DRE->getLocStart(), diag::err_atomic_op_needs_trivial_copy) 793 << Ptr->getType() << Ptr->getSourceRange(); 794 return ExprError(); 795 } 796 797 // FIXME: For any builtin other than a load, the ValType must not be 798 // const-qualified. 799 800 switch (ValType.getObjCLifetime()) { 801 case Qualifiers::OCL_None: 802 case Qualifiers::OCL_ExplicitNone: 803 // okay 804 break; 805 806 case Qualifiers::OCL_Weak: 807 case Qualifiers::OCL_Strong: 808 case Qualifiers::OCL_Autoreleasing: 809 // FIXME: Can this happen? By this point, ValType should be known 810 // to be trivially copyable. 811 Diag(DRE->getLocStart(), diag::err_arc_atomic_ownership) 812 << ValType << Ptr->getSourceRange(); 813 return ExprError(); 814 } 815 816 QualType ResultType = ValType; 817 if (Form == Copy || Form == GNUXchg || Form == Init) 818 ResultType = Context.VoidTy; 819 else if (Form == C11CmpXchg || Form == GNUCmpXchg) 820 ResultType = Context.BoolTy; 821 822 // The type of a parameter passed 'by value'. In the GNU atomics, such 823 // arguments are actually passed as pointers. 824 QualType ByValType = ValType; // 'CP' 825 if (!IsC11 && !IsN) 826 ByValType = Ptr->getType(); 827 828 // The first argument --- the pointer --- has a fixed type; we 829 // deduce the types of the rest of the arguments accordingly. Walk 830 // the remaining arguments, converting them to the deduced value type. 831 for (unsigned i = 1; i != NumArgs[Form]; ++i) { 832 QualType Ty; 833 if (i < NumVals[Form] + 1) { 834 switch (i) { 835 case 1: 836 // The second argument is the non-atomic operand. For arithmetic, this 837 // is always passed by value, and for a compare_exchange it is always 838 // passed by address. For the rest, GNU uses by-address and C11 uses 839 // by-value. 840 assert(Form != Load); 841 if (Form == Init || (Form == Arithmetic && ValType->isIntegerType())) 842 Ty = ValType; 843 else if (Form == Copy || Form == Xchg) 844 Ty = ByValType; 845 else if (Form == Arithmetic) 846 Ty = Context.getPointerDiffType(); 847 else 848 Ty = Context.getPointerType(ValType.getUnqualifiedType()); 849 break; 850 case 2: 851 // The third argument to compare_exchange / GNU exchange is a 852 // (pointer to a) desired value. 853 Ty = ByValType; 854 break; 855 case 3: 856 // The fourth argument to GNU compare_exchange is a 'weak' flag. 857 Ty = Context.BoolTy; 858 break; 859 } 860 } else { 861 // The order(s) are always converted to int. 862 Ty = Context.IntTy; 863 } 864 865 InitializedEntity Entity = 866 InitializedEntity::InitializeParameter(Context, Ty, false); 867 ExprResult Arg = TheCall->getArg(i); 868 Arg = PerformCopyInitialization(Entity, SourceLocation(), Arg); 869 if (Arg.isInvalid()) 870 return true; 871 TheCall->setArg(i, Arg.get()); 872 } 873 874 // Permute the arguments into a 'consistent' order. 875 SmallVector<Expr*, 5> SubExprs; 876 SubExprs.push_back(Ptr); 877 switch (Form) { 878 case Init: 879 // Note, AtomicExpr::getVal1() has a special case for this atomic. 880 SubExprs.push_back(TheCall->getArg(1)); // Val1 881 break; 882 case Load: 883 SubExprs.push_back(TheCall->getArg(1)); // Order 884 break; 885 case Copy: 886 case Arithmetic: 887 case Xchg: 888 SubExprs.push_back(TheCall->getArg(2)); // Order 889 SubExprs.push_back(TheCall->getArg(1)); // Val1 890 break; 891 case GNUXchg: 892 // Note, AtomicExpr::getVal2() has a special case for this atomic. 893 SubExprs.push_back(TheCall->getArg(3)); // Order 894 SubExprs.push_back(TheCall->getArg(1)); // Val1 895 SubExprs.push_back(TheCall->getArg(2)); // Val2 896 break; 897 case C11CmpXchg: 898 SubExprs.push_back(TheCall->getArg(3)); // Order 899 SubExprs.push_back(TheCall->getArg(1)); // Val1 900 SubExprs.push_back(TheCall->getArg(4)); // OrderFail 901 SubExprs.push_back(TheCall->getArg(2)); // Val2 902 break; 903 case GNUCmpXchg: 904 SubExprs.push_back(TheCall->getArg(4)); // Order 905 SubExprs.push_back(TheCall->getArg(1)); // Val1 906 SubExprs.push_back(TheCall->getArg(5)); // OrderFail 907 SubExprs.push_back(TheCall->getArg(2)); // Val2 908 SubExprs.push_back(TheCall->getArg(3)); // Weak 909 break; 910 } 911 912 return Owned(new (Context) AtomicExpr(TheCall->getCallee()->getLocStart(), 913 SubExprs, ResultType, Op, 914 TheCall->getRParenLoc())); 915} 916 917 918/// checkBuiltinArgument - Given a call to a builtin function, perform 919/// normal type-checking on the given argument, updating the call in 920/// place. This is useful when a builtin function requires custom 921/// type-checking for some of its arguments but not necessarily all of 922/// them. 923/// 924/// Returns true on error. 925static bool checkBuiltinArgument(Sema &S, CallExpr *E, unsigned ArgIndex) { 926 FunctionDecl *Fn = E->getDirectCallee(); 927 assert(Fn && "builtin call without direct callee!"); 928 929 ParmVarDecl *Param = Fn->getParamDecl(ArgIndex); 930 InitializedEntity Entity = 931 InitializedEntity::InitializeParameter(S.Context, Param); 932 933 ExprResult Arg = E->getArg(0); 934 Arg = S.PerformCopyInitialization(Entity, SourceLocation(), Arg); 935 if (Arg.isInvalid()) 936 return true; 937 938 E->setArg(ArgIndex, Arg.take()); 939 return false; 940} 941 942/// SemaBuiltinAtomicOverloaded - We have a call to a function like 943/// __sync_fetch_and_add, which is an overloaded function based on the pointer 944/// type of its first argument. The main ActOnCallExpr routines have already 945/// promoted the types of arguments because all of these calls are prototyped as 946/// void(...). 947/// 948/// This function goes through and does final semantic checking for these 949/// builtins, 950ExprResult 951Sema::SemaBuiltinAtomicOverloaded(ExprResult TheCallResult) { 952 CallExpr *TheCall = (CallExpr *)TheCallResult.get(); 953 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 954 FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl()); 955 956 // Ensure that we have at least one argument to do type inference from. 957 if (TheCall->getNumArgs() < 1) { 958 Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args_at_least) 959 << 0 << 1 << TheCall->getNumArgs() 960 << TheCall->getCallee()->getSourceRange(); 961 return ExprError(); 962 } 963 964 // Inspect the first argument of the atomic builtin. This should always be 965 // a pointer type, whose element is an integral scalar or pointer type. 966 // Because it is a pointer type, we don't have to worry about any implicit 967 // casts here. 968 // FIXME: We don't allow floating point scalars as input. 969 Expr *FirstArg = TheCall->getArg(0); 970 ExprResult FirstArgResult = DefaultFunctionArrayLvalueConversion(FirstArg); 971 if (FirstArgResult.isInvalid()) 972 return ExprError(); 973 FirstArg = FirstArgResult.take(); 974 TheCall->setArg(0, FirstArg); 975 976 const PointerType *pointerType = FirstArg->getType()->getAs<PointerType>(); 977 if (!pointerType) { 978 Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer) 979 << FirstArg->getType() << FirstArg->getSourceRange(); 980 return ExprError(); 981 } 982 983 QualType ValType = pointerType->getPointeeType(); 984 if (!ValType->isIntegerType() && !ValType->isAnyPointerType() && 985 !ValType->isBlockPointerType()) { 986 Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer_intptr) 987 << FirstArg->getType() << FirstArg->getSourceRange(); 988 return ExprError(); 989 } 990 991 switch (ValType.getObjCLifetime()) { 992 case Qualifiers::OCL_None: 993 case Qualifiers::OCL_ExplicitNone: 994 // okay 995 break; 996 997 case Qualifiers::OCL_Weak: 998 case Qualifiers::OCL_Strong: 999 case Qualifiers::OCL_Autoreleasing: 1000 Diag(DRE->getLocStart(), diag::err_arc_atomic_ownership) 1001 << ValType << FirstArg->getSourceRange(); 1002 return ExprError(); 1003 } 1004 1005 // Strip any qualifiers off ValType. 1006 ValType = ValType.getUnqualifiedType(); 1007 1008 // The majority of builtins return a value, but a few have special return 1009 // types, so allow them to override appropriately below. 1010 QualType ResultType = ValType; 1011 1012 // We need to figure out which concrete builtin this maps onto. For example, 1013 // __sync_fetch_and_add with a 2 byte object turns into 1014 // __sync_fetch_and_add_2. 1015#define BUILTIN_ROW(x) \ 1016 { Builtin::BI##x##_1, Builtin::BI##x##_2, Builtin::BI##x##_4, \ 1017 Builtin::BI##x##_8, Builtin::BI##x##_16 } 1018 1019 static const unsigned BuiltinIndices[][5] = { 1020 BUILTIN_ROW(__sync_fetch_and_add), 1021 BUILTIN_ROW(__sync_fetch_and_sub), 1022 BUILTIN_ROW(__sync_fetch_and_or), 1023 BUILTIN_ROW(__sync_fetch_and_and), 1024 BUILTIN_ROW(__sync_fetch_and_xor), 1025 1026 BUILTIN_ROW(__sync_add_and_fetch), 1027 BUILTIN_ROW(__sync_sub_and_fetch), 1028 BUILTIN_ROW(__sync_and_and_fetch), 1029 BUILTIN_ROW(__sync_or_and_fetch), 1030 BUILTIN_ROW(__sync_xor_and_fetch), 1031 1032 BUILTIN_ROW(__sync_val_compare_and_swap), 1033 BUILTIN_ROW(__sync_bool_compare_and_swap), 1034 BUILTIN_ROW(__sync_lock_test_and_set), 1035 BUILTIN_ROW(__sync_lock_release), 1036 BUILTIN_ROW(__sync_swap) 1037 }; 1038#undef BUILTIN_ROW 1039 1040 // Determine the index of the size. 1041 unsigned SizeIndex; 1042 switch (Context.getTypeSizeInChars(ValType).getQuantity()) { 1043 case 1: SizeIndex = 0; break; 1044 case 2: SizeIndex = 1; break; 1045 case 4: SizeIndex = 2; break; 1046 case 8: SizeIndex = 3; break; 1047 case 16: SizeIndex = 4; break; 1048 default: 1049 Diag(DRE->getLocStart(), diag::err_atomic_builtin_pointer_size) 1050 << FirstArg->getType() << FirstArg->getSourceRange(); 1051 return ExprError(); 1052 } 1053 1054 // Each of these builtins has one pointer argument, followed by some number of 1055 // values (0, 1 or 2) followed by a potentially empty varags list of stuff 1056 // that we ignore. Find out which row of BuiltinIndices to read from as well 1057 // as the number of fixed args. 1058 unsigned BuiltinID = FDecl->getBuiltinID(); 1059 unsigned BuiltinIndex, NumFixed = 1; 1060 switch (BuiltinID) { 1061 default: llvm_unreachable("Unknown overloaded atomic builtin!"); 1062 case Builtin::BI__sync_fetch_and_add: 1063 case Builtin::BI__sync_fetch_and_add_1: 1064 case Builtin::BI__sync_fetch_and_add_2: 1065 case Builtin::BI__sync_fetch_and_add_4: 1066 case Builtin::BI__sync_fetch_and_add_8: 1067 case Builtin::BI__sync_fetch_and_add_16: 1068 BuiltinIndex = 0; 1069 break; 1070 1071 case Builtin::BI__sync_fetch_and_sub: 1072 case Builtin::BI__sync_fetch_and_sub_1: 1073 case Builtin::BI__sync_fetch_and_sub_2: 1074 case Builtin::BI__sync_fetch_and_sub_4: 1075 case Builtin::BI__sync_fetch_and_sub_8: 1076 case Builtin::BI__sync_fetch_and_sub_16: 1077 BuiltinIndex = 1; 1078 break; 1079 1080 case Builtin::BI__sync_fetch_and_or: 1081 case Builtin::BI__sync_fetch_and_or_1: 1082 case Builtin::BI__sync_fetch_and_or_2: 1083 case Builtin::BI__sync_fetch_and_or_4: 1084 case Builtin::BI__sync_fetch_and_or_8: 1085 case Builtin::BI__sync_fetch_and_or_16: 1086 BuiltinIndex = 2; 1087 break; 1088 1089 case Builtin::BI__sync_fetch_and_and: 1090 case Builtin::BI__sync_fetch_and_and_1: 1091 case Builtin::BI__sync_fetch_and_and_2: 1092 case Builtin::BI__sync_fetch_and_and_4: 1093 case Builtin::BI__sync_fetch_and_and_8: 1094 case Builtin::BI__sync_fetch_and_and_16: 1095 BuiltinIndex = 3; 1096 break; 1097 1098 case Builtin::BI__sync_fetch_and_xor: 1099 case Builtin::BI__sync_fetch_and_xor_1: 1100 case Builtin::BI__sync_fetch_and_xor_2: 1101 case Builtin::BI__sync_fetch_and_xor_4: 1102 case Builtin::BI__sync_fetch_and_xor_8: 1103 case Builtin::BI__sync_fetch_and_xor_16: 1104 BuiltinIndex = 4; 1105 break; 1106 1107 case Builtin::BI__sync_add_and_fetch: 1108 case Builtin::BI__sync_add_and_fetch_1: 1109 case Builtin::BI__sync_add_and_fetch_2: 1110 case Builtin::BI__sync_add_and_fetch_4: 1111 case Builtin::BI__sync_add_and_fetch_8: 1112 case Builtin::BI__sync_add_and_fetch_16: 1113 BuiltinIndex = 5; 1114 break; 1115 1116 case Builtin::BI__sync_sub_and_fetch: 1117 case Builtin::BI__sync_sub_and_fetch_1: 1118 case Builtin::BI__sync_sub_and_fetch_2: 1119 case Builtin::BI__sync_sub_and_fetch_4: 1120 case Builtin::BI__sync_sub_and_fetch_8: 1121 case Builtin::BI__sync_sub_and_fetch_16: 1122 BuiltinIndex = 6; 1123 break; 1124 1125 case Builtin::BI__sync_and_and_fetch: 1126 case Builtin::BI__sync_and_and_fetch_1: 1127 case Builtin::BI__sync_and_and_fetch_2: 1128 case Builtin::BI__sync_and_and_fetch_4: 1129 case Builtin::BI__sync_and_and_fetch_8: 1130 case Builtin::BI__sync_and_and_fetch_16: 1131 BuiltinIndex = 7; 1132 break; 1133 1134 case Builtin::BI__sync_or_and_fetch: 1135 case Builtin::BI__sync_or_and_fetch_1: 1136 case Builtin::BI__sync_or_and_fetch_2: 1137 case Builtin::BI__sync_or_and_fetch_4: 1138 case Builtin::BI__sync_or_and_fetch_8: 1139 case Builtin::BI__sync_or_and_fetch_16: 1140 BuiltinIndex = 8; 1141 break; 1142 1143 case Builtin::BI__sync_xor_and_fetch: 1144 case Builtin::BI__sync_xor_and_fetch_1: 1145 case Builtin::BI__sync_xor_and_fetch_2: 1146 case Builtin::BI__sync_xor_and_fetch_4: 1147 case Builtin::BI__sync_xor_and_fetch_8: 1148 case Builtin::BI__sync_xor_and_fetch_16: 1149 BuiltinIndex = 9; 1150 break; 1151 1152 case Builtin::BI__sync_val_compare_and_swap: 1153 case Builtin::BI__sync_val_compare_and_swap_1: 1154 case Builtin::BI__sync_val_compare_and_swap_2: 1155 case Builtin::BI__sync_val_compare_and_swap_4: 1156 case Builtin::BI__sync_val_compare_and_swap_8: 1157 case Builtin::BI__sync_val_compare_and_swap_16: 1158 BuiltinIndex = 10; 1159 NumFixed = 2; 1160 break; 1161 1162 case Builtin::BI__sync_bool_compare_and_swap: 1163 case Builtin::BI__sync_bool_compare_and_swap_1: 1164 case Builtin::BI__sync_bool_compare_and_swap_2: 1165 case Builtin::BI__sync_bool_compare_and_swap_4: 1166 case Builtin::BI__sync_bool_compare_and_swap_8: 1167 case Builtin::BI__sync_bool_compare_and_swap_16: 1168 BuiltinIndex = 11; 1169 NumFixed = 2; 1170 ResultType = Context.BoolTy; 1171 break; 1172 1173 case Builtin::BI__sync_lock_test_and_set: 1174 case Builtin::BI__sync_lock_test_and_set_1: 1175 case Builtin::BI__sync_lock_test_and_set_2: 1176 case Builtin::BI__sync_lock_test_and_set_4: 1177 case Builtin::BI__sync_lock_test_and_set_8: 1178 case Builtin::BI__sync_lock_test_and_set_16: 1179 BuiltinIndex = 12; 1180 break; 1181 1182 case Builtin::BI__sync_lock_release: 1183 case Builtin::BI__sync_lock_release_1: 1184 case Builtin::BI__sync_lock_release_2: 1185 case Builtin::BI__sync_lock_release_4: 1186 case Builtin::BI__sync_lock_release_8: 1187 case Builtin::BI__sync_lock_release_16: 1188 BuiltinIndex = 13; 1189 NumFixed = 0; 1190 ResultType = Context.VoidTy; 1191 break; 1192 1193 case Builtin::BI__sync_swap: 1194 case Builtin::BI__sync_swap_1: 1195 case Builtin::BI__sync_swap_2: 1196 case Builtin::BI__sync_swap_4: 1197 case Builtin::BI__sync_swap_8: 1198 case Builtin::BI__sync_swap_16: 1199 BuiltinIndex = 14; 1200 break; 1201 } 1202 1203 // Now that we know how many fixed arguments we expect, first check that we 1204 // have at least that many. 1205 if (TheCall->getNumArgs() < 1+NumFixed) { 1206 Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args_at_least) 1207 << 0 << 1+NumFixed << TheCall->getNumArgs() 1208 << TheCall->getCallee()->getSourceRange(); 1209 return ExprError(); 1210 } 1211 1212 // Get the decl for the concrete builtin from this, we can tell what the 1213 // concrete integer type we should convert to is. 1214 unsigned NewBuiltinID = BuiltinIndices[BuiltinIndex][SizeIndex]; 1215 const char *NewBuiltinName = Context.BuiltinInfo.GetName(NewBuiltinID); 1216 FunctionDecl *NewBuiltinDecl; 1217 if (NewBuiltinID == BuiltinID) 1218 NewBuiltinDecl = FDecl; 1219 else { 1220 // Perform builtin lookup to avoid redeclaring it. 1221 DeclarationName DN(&Context.Idents.get(NewBuiltinName)); 1222 LookupResult Res(*this, DN, DRE->getLocStart(), LookupOrdinaryName); 1223 LookupName(Res, TUScope, /*AllowBuiltinCreation=*/true); 1224 assert(Res.getFoundDecl()); 1225 NewBuiltinDecl = dyn_cast<FunctionDecl>(Res.getFoundDecl()); 1226 if (NewBuiltinDecl == 0) 1227 return ExprError(); 1228 } 1229 1230 // The first argument --- the pointer --- has a fixed type; we 1231 // deduce the types of the rest of the arguments accordingly. Walk 1232 // the remaining arguments, converting them to the deduced value type. 1233 for (unsigned i = 0; i != NumFixed; ++i) { 1234 ExprResult Arg = TheCall->getArg(i+1); 1235 1236 // GCC does an implicit conversion to the pointer or integer ValType. This 1237 // can fail in some cases (1i -> int**), check for this error case now. 1238 // Initialize the argument. 1239 InitializedEntity Entity = InitializedEntity::InitializeParameter(Context, 1240 ValType, /*consume*/ false); 1241 Arg = PerformCopyInitialization(Entity, SourceLocation(), Arg); 1242 if (Arg.isInvalid()) 1243 return ExprError(); 1244 1245 // Okay, we have something that *can* be converted to the right type. Check 1246 // to see if there is a potentially weird extension going on here. This can 1247 // happen when you do an atomic operation on something like an char* and 1248 // pass in 42. The 42 gets converted to char. This is even more strange 1249 // for things like 45.123 -> char, etc. 1250 // FIXME: Do this check. 1251 TheCall->setArg(i+1, Arg.take()); 1252 } 1253 1254 ASTContext& Context = this->getASTContext(); 1255 1256 // Create a new DeclRefExpr to refer to the new decl. 1257 DeclRefExpr* NewDRE = DeclRefExpr::Create( 1258 Context, 1259 DRE->getQualifierLoc(), 1260 SourceLocation(), 1261 NewBuiltinDecl, 1262 /*enclosing*/ false, 1263 DRE->getLocation(), 1264 Context.BuiltinFnTy, 1265 DRE->getValueKind()); 1266 1267 // Set the callee in the CallExpr. 1268 // FIXME: This loses syntactic information. 1269 QualType CalleePtrTy = Context.getPointerType(NewBuiltinDecl->getType()); 1270 ExprResult PromotedCall = ImpCastExprToType(NewDRE, CalleePtrTy, 1271 CK_BuiltinFnToFnPtr); 1272 TheCall->setCallee(PromotedCall.take()); 1273 1274 // Change the result type of the call to match the original value type. This 1275 // is arbitrary, but the codegen for these builtins ins design to handle it 1276 // gracefully. 1277 TheCall->setType(ResultType); 1278 1279 return TheCallResult; 1280} 1281 1282/// CheckObjCString - Checks that the argument to the builtin 1283/// CFString constructor is correct 1284/// Note: It might also make sense to do the UTF-16 conversion here (would 1285/// simplify the backend). 1286bool Sema::CheckObjCString(Expr *Arg) { 1287 Arg = Arg->IgnoreParenCasts(); 1288 StringLiteral *Literal = dyn_cast<StringLiteral>(Arg); 1289 1290 if (!Literal || !Literal->isAscii()) { 1291 Diag(Arg->getLocStart(), diag::err_cfstring_literal_not_string_constant) 1292 << Arg->getSourceRange(); 1293 return true; 1294 } 1295 1296 if (Literal->containsNonAsciiOrNull()) { 1297 StringRef String = Literal->getString(); 1298 unsigned NumBytes = String.size(); 1299 SmallVector<UTF16, 128> ToBuf(NumBytes); 1300 const UTF8 *FromPtr = (const UTF8 *)String.data(); 1301 UTF16 *ToPtr = &ToBuf[0]; 1302 1303 ConversionResult Result = ConvertUTF8toUTF16(&FromPtr, FromPtr + NumBytes, 1304 &ToPtr, ToPtr + NumBytes, 1305 strictConversion); 1306 // Check for conversion failure. 1307 if (Result != conversionOK) 1308 Diag(Arg->getLocStart(), 1309 diag::warn_cfstring_truncated) << Arg->getSourceRange(); 1310 } 1311 return false; 1312} 1313 1314/// SemaBuiltinVAStart - Check the arguments to __builtin_va_start for validity. 1315/// Emit an error and return true on failure, return false on success. 1316bool Sema::SemaBuiltinVAStart(CallExpr *TheCall) { 1317 Expr *Fn = TheCall->getCallee(); 1318 if (TheCall->getNumArgs() > 2) { 1319 Diag(TheCall->getArg(2)->getLocStart(), 1320 diag::err_typecheck_call_too_many_args) 1321 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 1322 << Fn->getSourceRange() 1323 << SourceRange(TheCall->getArg(2)->getLocStart(), 1324 (*(TheCall->arg_end()-1))->getLocEnd()); 1325 return true; 1326 } 1327 1328 if (TheCall->getNumArgs() < 2) { 1329 return Diag(TheCall->getLocEnd(), 1330 diag::err_typecheck_call_too_few_args_at_least) 1331 << 0 /*function call*/ << 2 << TheCall->getNumArgs(); 1332 } 1333 1334 // Type-check the first argument normally. 1335 if (checkBuiltinArgument(*this, TheCall, 0)) 1336 return true; 1337 1338 // Determine whether the current function is variadic or not. 1339 BlockScopeInfo *CurBlock = getCurBlock(); 1340 bool isVariadic; 1341 if (CurBlock) 1342 isVariadic = CurBlock->TheDecl->isVariadic(); 1343 else if (FunctionDecl *FD = getCurFunctionDecl()) 1344 isVariadic = FD->isVariadic(); 1345 else 1346 isVariadic = getCurMethodDecl()->isVariadic(); 1347 1348 if (!isVariadic) { 1349 Diag(Fn->getLocStart(), diag::err_va_start_used_in_non_variadic_function); 1350 return true; 1351 } 1352 1353 // Verify that the second argument to the builtin is the last argument of the 1354 // current function or method. 1355 bool SecondArgIsLastNamedArgument = false; 1356 const Expr *Arg = TheCall->getArg(1)->IgnoreParenCasts(); 1357 1358 // These are valid if SecondArgIsLastNamedArgument is false after the next 1359 // block. 1360 QualType Type; 1361 SourceLocation ParamLoc; 1362 1363 if (const DeclRefExpr *DR = dyn_cast<DeclRefExpr>(Arg)) { 1364 if (const ParmVarDecl *PV = dyn_cast<ParmVarDecl>(DR->getDecl())) { 1365 // FIXME: This isn't correct for methods (results in bogus warning). 1366 // Get the last formal in the current function. 1367 const ParmVarDecl *LastArg; 1368 if (CurBlock) 1369 LastArg = *(CurBlock->TheDecl->param_end()-1); 1370 else if (FunctionDecl *FD = getCurFunctionDecl()) 1371 LastArg = *(FD->param_end()-1); 1372 else 1373 LastArg = *(getCurMethodDecl()->param_end()-1); 1374 SecondArgIsLastNamedArgument = PV == LastArg; 1375 1376 Type = PV->getType(); 1377 ParamLoc = PV->getLocation(); 1378 } 1379 } 1380 1381 if (!SecondArgIsLastNamedArgument) 1382 Diag(TheCall->getArg(1)->getLocStart(), 1383 diag::warn_second_parameter_of_va_start_not_last_named_argument); 1384 else if (Type->isReferenceType()) { 1385 Diag(Arg->getLocStart(), 1386 diag::warn_va_start_of_reference_type_is_undefined); 1387 Diag(ParamLoc, diag::note_parameter_type) << Type; 1388 } 1389 1390 return false; 1391} 1392 1393/// SemaBuiltinUnorderedCompare - Handle functions like __builtin_isgreater and 1394/// friends. This is declared to take (...), so we have to check everything. 1395bool Sema::SemaBuiltinUnorderedCompare(CallExpr *TheCall) { 1396 if (TheCall->getNumArgs() < 2) 1397 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 1398 << 0 << 2 << TheCall->getNumArgs()/*function call*/; 1399 if (TheCall->getNumArgs() > 2) 1400 return Diag(TheCall->getArg(2)->getLocStart(), 1401 diag::err_typecheck_call_too_many_args) 1402 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 1403 << SourceRange(TheCall->getArg(2)->getLocStart(), 1404 (*(TheCall->arg_end()-1))->getLocEnd()); 1405 1406 ExprResult OrigArg0 = TheCall->getArg(0); 1407 ExprResult OrigArg1 = TheCall->getArg(1); 1408 1409 // Do standard promotions between the two arguments, returning their common 1410 // type. 1411 QualType Res = UsualArithmeticConversions(OrigArg0, OrigArg1, false); 1412 if (OrigArg0.isInvalid() || OrigArg1.isInvalid()) 1413 return true; 1414 1415 // Make sure any conversions are pushed back into the call; this is 1416 // type safe since unordered compare builtins are declared as "_Bool 1417 // foo(...)". 1418 TheCall->setArg(0, OrigArg0.get()); 1419 TheCall->setArg(1, OrigArg1.get()); 1420 1421 if (OrigArg0.get()->isTypeDependent() || OrigArg1.get()->isTypeDependent()) 1422 return false; 1423 1424 // If the common type isn't a real floating type, then the arguments were 1425 // invalid for this operation. 1426 if (Res.isNull() || !Res->isRealFloatingType()) 1427 return Diag(OrigArg0.get()->getLocStart(), 1428 diag::err_typecheck_call_invalid_ordered_compare) 1429 << OrigArg0.get()->getType() << OrigArg1.get()->getType() 1430 << SourceRange(OrigArg0.get()->getLocStart(), OrigArg1.get()->getLocEnd()); 1431 1432 return false; 1433} 1434 1435/// SemaBuiltinSemaBuiltinFPClassification - Handle functions like 1436/// __builtin_isnan and friends. This is declared to take (...), so we have 1437/// to check everything. We expect the last argument to be a floating point 1438/// value. 1439bool Sema::SemaBuiltinFPClassification(CallExpr *TheCall, unsigned NumArgs) { 1440 if (TheCall->getNumArgs() < NumArgs) 1441 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 1442 << 0 << NumArgs << TheCall->getNumArgs()/*function call*/; 1443 if (TheCall->getNumArgs() > NumArgs) 1444 return Diag(TheCall->getArg(NumArgs)->getLocStart(), 1445 diag::err_typecheck_call_too_many_args) 1446 << 0 /*function call*/ << NumArgs << TheCall->getNumArgs() 1447 << SourceRange(TheCall->getArg(NumArgs)->getLocStart(), 1448 (*(TheCall->arg_end()-1))->getLocEnd()); 1449 1450 Expr *OrigArg = TheCall->getArg(NumArgs-1); 1451 1452 if (OrigArg->isTypeDependent()) 1453 return false; 1454 1455 // This operation requires a non-_Complex floating-point number. 1456 if (!OrigArg->getType()->isRealFloatingType()) 1457 return Diag(OrigArg->getLocStart(), 1458 diag::err_typecheck_call_invalid_unary_fp) 1459 << OrigArg->getType() << OrigArg->getSourceRange(); 1460 1461 // If this is an implicit conversion from float -> double, remove it. 1462 if (ImplicitCastExpr *Cast = dyn_cast<ImplicitCastExpr>(OrigArg)) { 1463 Expr *CastArg = Cast->getSubExpr(); 1464 if (CastArg->getType()->isSpecificBuiltinType(BuiltinType::Float)) { 1465 assert(Cast->getType()->isSpecificBuiltinType(BuiltinType::Double) && 1466 "promotion from float to double is the only expected cast here"); 1467 Cast->setSubExpr(0); 1468 TheCall->setArg(NumArgs-1, CastArg); 1469 } 1470 } 1471 1472 return false; 1473} 1474 1475/// SemaBuiltinShuffleVector - Handle __builtin_shufflevector. 1476// This is declared to take (...), so we have to check everything. 1477ExprResult Sema::SemaBuiltinShuffleVector(CallExpr *TheCall) { 1478 if (TheCall->getNumArgs() < 2) 1479 return ExprError(Diag(TheCall->getLocEnd(), 1480 diag::err_typecheck_call_too_few_args_at_least) 1481 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 1482 << TheCall->getSourceRange()); 1483 1484 // Determine which of the following types of shufflevector we're checking: 1485 // 1) unary, vector mask: (lhs, mask) 1486 // 2) binary, vector mask: (lhs, rhs, mask) 1487 // 3) binary, scalar mask: (lhs, rhs, index, ..., index) 1488 QualType resType = TheCall->getArg(0)->getType(); 1489 unsigned numElements = 0; 1490 1491 if (!TheCall->getArg(0)->isTypeDependent() && 1492 !TheCall->getArg(1)->isTypeDependent()) { 1493 QualType LHSType = TheCall->getArg(0)->getType(); 1494 QualType RHSType = TheCall->getArg(1)->getType(); 1495 1496 if (!LHSType->isVectorType() || !RHSType->isVectorType()) { 1497 Diag(TheCall->getLocStart(), diag::err_shufflevector_non_vector) 1498 << SourceRange(TheCall->getArg(0)->getLocStart(), 1499 TheCall->getArg(1)->getLocEnd()); 1500 return ExprError(); 1501 } 1502 1503 numElements = LHSType->getAs<VectorType>()->getNumElements(); 1504 unsigned numResElements = TheCall->getNumArgs() - 2; 1505 1506 // Check to see if we have a call with 2 vector arguments, the unary shuffle 1507 // with mask. If so, verify that RHS is an integer vector type with the 1508 // same number of elts as lhs. 1509 if (TheCall->getNumArgs() == 2) { 1510 if (!RHSType->hasIntegerRepresentation() || 1511 RHSType->getAs<VectorType>()->getNumElements() != numElements) 1512 Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector) 1513 << SourceRange(TheCall->getArg(1)->getLocStart(), 1514 TheCall->getArg(1)->getLocEnd()); 1515 numResElements = numElements; 1516 } 1517 else if (!Context.hasSameUnqualifiedType(LHSType, RHSType)) { 1518 Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector) 1519 << SourceRange(TheCall->getArg(0)->getLocStart(), 1520 TheCall->getArg(1)->getLocEnd()); 1521 return ExprError(); 1522 } else if (numElements != numResElements) { 1523 QualType eltType = LHSType->getAs<VectorType>()->getElementType(); 1524 resType = Context.getVectorType(eltType, numResElements, 1525 VectorType::GenericVector); 1526 } 1527 } 1528 1529 for (unsigned i = 2; i < TheCall->getNumArgs(); i++) { 1530 if (TheCall->getArg(i)->isTypeDependent() || 1531 TheCall->getArg(i)->isValueDependent()) 1532 continue; 1533 1534 llvm::APSInt Result(32); 1535 if (!TheCall->getArg(i)->isIntegerConstantExpr(Result, Context)) 1536 return ExprError(Diag(TheCall->getLocStart(), 1537 diag::err_shufflevector_nonconstant_argument) 1538 << TheCall->getArg(i)->getSourceRange()); 1539 1540 if (Result.getActiveBits() > 64 || Result.getZExtValue() >= numElements*2) 1541 return ExprError(Diag(TheCall->getLocStart(), 1542 diag::err_shufflevector_argument_too_large) 1543 << TheCall->getArg(i)->getSourceRange()); 1544 } 1545 1546 SmallVector<Expr*, 32> exprs; 1547 1548 for (unsigned i = 0, e = TheCall->getNumArgs(); i != e; i++) { 1549 exprs.push_back(TheCall->getArg(i)); 1550 TheCall->setArg(i, 0); 1551 } 1552 1553 return Owned(new (Context) ShuffleVectorExpr(Context, exprs, resType, 1554 TheCall->getCallee()->getLocStart(), 1555 TheCall->getRParenLoc())); 1556} 1557 1558/// SemaBuiltinPrefetch - Handle __builtin_prefetch. 1559// This is declared to take (const void*, ...) and can take two 1560// optional constant int args. 1561bool Sema::SemaBuiltinPrefetch(CallExpr *TheCall) { 1562 unsigned NumArgs = TheCall->getNumArgs(); 1563 1564 if (NumArgs > 3) 1565 return Diag(TheCall->getLocEnd(), 1566 diag::err_typecheck_call_too_many_args_at_most) 1567 << 0 /*function call*/ << 3 << NumArgs 1568 << TheCall->getSourceRange(); 1569 1570 // Argument 0 is checked for us and the remaining arguments must be 1571 // constant integers. 1572 for (unsigned i = 1; i != NumArgs; ++i) { 1573 Expr *Arg = TheCall->getArg(i); 1574 1575 // We can't check the value of a dependent argument. 1576 if (Arg->isTypeDependent() || Arg->isValueDependent()) 1577 continue; 1578 1579 llvm::APSInt Result; 1580 if (SemaBuiltinConstantArg(TheCall, i, Result)) 1581 return true; 1582 1583 // FIXME: gcc issues a warning and rewrites these to 0. These 1584 // seems especially odd for the third argument since the default 1585 // is 3. 1586 if (i == 1) { 1587 if (Result.getLimitedValue() > 1) 1588 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 1589 << "0" << "1" << Arg->getSourceRange(); 1590 } else { 1591 if (Result.getLimitedValue() > 3) 1592 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 1593 << "0" << "3" << Arg->getSourceRange(); 1594 } 1595 } 1596 1597 return false; 1598} 1599 1600/// SemaBuiltinConstantArg - Handle a check if argument ArgNum of CallExpr 1601/// TheCall is a constant expression. 1602bool Sema::SemaBuiltinConstantArg(CallExpr *TheCall, int ArgNum, 1603 llvm::APSInt &Result) { 1604 Expr *Arg = TheCall->getArg(ArgNum); 1605 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 1606 FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl()); 1607 1608 if (Arg->isTypeDependent() || Arg->isValueDependent()) return false; 1609 1610 if (!Arg->isIntegerConstantExpr(Result, Context)) 1611 return Diag(TheCall->getLocStart(), diag::err_constant_integer_arg_type) 1612 << FDecl->getDeclName() << Arg->getSourceRange(); 1613 1614 return false; 1615} 1616 1617/// SemaBuiltinObjectSize - Handle __builtin_object_size(void *ptr, 1618/// int type). This simply type checks that type is one of the defined 1619/// constants (0-3). 1620// For compatibility check 0-3, llvm only handles 0 and 2. 1621bool Sema::SemaBuiltinObjectSize(CallExpr *TheCall) { 1622 llvm::APSInt Result; 1623 1624 // We can't check the value of a dependent argument. 1625 if (TheCall->getArg(1)->isTypeDependent() || 1626 TheCall->getArg(1)->isValueDependent()) 1627 return false; 1628 1629 // Check constant-ness first. 1630 if (SemaBuiltinConstantArg(TheCall, 1, Result)) 1631 return true; 1632 1633 Expr *Arg = TheCall->getArg(1); 1634 if (Result.getSExtValue() < 0 || Result.getSExtValue() > 3) { 1635 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 1636 << "0" << "3" << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 1637 } 1638 1639 return false; 1640} 1641 1642/// SemaBuiltinLongjmp - Handle __builtin_longjmp(void *env[5], int val). 1643/// This checks that val is a constant 1. 1644bool Sema::SemaBuiltinLongjmp(CallExpr *TheCall) { 1645 Expr *Arg = TheCall->getArg(1); 1646 llvm::APSInt Result; 1647 1648 // TODO: This is less than ideal. Overload this to take a value. 1649 if (SemaBuiltinConstantArg(TheCall, 1, Result)) 1650 return true; 1651 1652 if (Result != 1) 1653 return Diag(TheCall->getLocStart(), diag::err_builtin_longjmp_invalid_val) 1654 << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 1655 1656 return false; 1657} 1658 1659// Determine if an expression is a string literal or constant string. 1660// If this function returns false on the arguments to a function expecting a 1661// format string, we will usually need to emit a warning. 1662// True string literals are then checked by CheckFormatString. 1663Sema::StringLiteralCheckType 1664Sema::checkFormatStringExpr(const Expr *E, ArrayRef<const Expr *> Args, 1665 bool HasVAListArg, 1666 unsigned format_idx, unsigned firstDataArg, 1667 FormatStringType Type, VariadicCallType CallType, 1668 bool inFunctionCall) { 1669 tryAgain: 1670 if (E->isTypeDependent() || E->isValueDependent()) 1671 return SLCT_NotALiteral; 1672 1673 E = E->IgnoreParenCasts(); 1674 1675 if (E->isNullPointerConstant(Context, Expr::NPC_ValueDependentIsNotNull)) 1676 // Technically -Wformat-nonliteral does not warn about this case. 1677 // The behavior of printf and friends in this case is implementation 1678 // dependent. Ideally if the format string cannot be null then 1679 // it should have a 'nonnull' attribute in the function prototype. 1680 return SLCT_CheckedLiteral; 1681 1682 switch (E->getStmtClass()) { 1683 case Stmt::BinaryConditionalOperatorClass: 1684 case Stmt::ConditionalOperatorClass: { 1685 // The expression is a literal if both sub-expressions were, and it was 1686 // completely checked only if both sub-expressions were checked. 1687 const AbstractConditionalOperator *C = 1688 cast<AbstractConditionalOperator>(E); 1689 StringLiteralCheckType Left = 1690 checkFormatStringExpr(C->getTrueExpr(), Args, 1691 HasVAListArg, format_idx, firstDataArg, 1692 Type, CallType, inFunctionCall); 1693 if (Left == SLCT_NotALiteral) 1694 return SLCT_NotALiteral; 1695 StringLiteralCheckType Right = 1696 checkFormatStringExpr(C->getFalseExpr(), Args, 1697 HasVAListArg, format_idx, firstDataArg, 1698 Type, CallType, inFunctionCall); 1699 return Left < Right ? Left : Right; 1700 } 1701 1702 case Stmt::ImplicitCastExprClass: { 1703 E = cast<ImplicitCastExpr>(E)->getSubExpr(); 1704 goto tryAgain; 1705 } 1706 1707 case Stmt::OpaqueValueExprClass: 1708 if (const Expr *src = cast<OpaqueValueExpr>(E)->getSourceExpr()) { 1709 E = src; 1710 goto tryAgain; 1711 } 1712 return SLCT_NotALiteral; 1713 1714 case Stmt::PredefinedExprClass: 1715 // While __func__, etc., are technically not string literals, they 1716 // cannot contain format specifiers and thus are not a security 1717 // liability. 1718 return SLCT_UncheckedLiteral; 1719 1720 case Stmt::DeclRefExprClass: { 1721 const DeclRefExpr *DR = cast<DeclRefExpr>(E); 1722 1723 // As an exception, do not flag errors for variables binding to 1724 // const string literals. 1725 if (const VarDecl *VD = dyn_cast<VarDecl>(DR->getDecl())) { 1726 bool isConstant = false; 1727 QualType T = DR->getType(); 1728 1729 if (const ArrayType *AT = Context.getAsArrayType(T)) { 1730 isConstant = AT->getElementType().isConstant(Context); 1731 } else if (const PointerType *PT = T->getAs<PointerType>()) { 1732 isConstant = T.isConstant(Context) && 1733 PT->getPointeeType().isConstant(Context); 1734 } else if (T->isObjCObjectPointerType()) { 1735 // In ObjC, there is usually no "const ObjectPointer" type, 1736 // so don't check if the pointee type is constant. 1737 isConstant = T.isConstant(Context); 1738 } 1739 1740 if (isConstant) { 1741 if (const Expr *Init = VD->getAnyInitializer()) { 1742 // Look through initializers like const char c[] = { "foo" } 1743 if (const InitListExpr *InitList = dyn_cast<InitListExpr>(Init)) { 1744 if (InitList->isStringLiteralInit()) 1745 Init = InitList->getInit(0)->IgnoreParenImpCasts(); 1746 } 1747 return checkFormatStringExpr(Init, Args, 1748 HasVAListArg, format_idx, 1749 firstDataArg, Type, CallType, 1750 /*inFunctionCall*/false); 1751 } 1752 } 1753 1754 // For vprintf* functions (i.e., HasVAListArg==true), we add a 1755 // special check to see if the format string is a function parameter 1756 // of the function calling the printf function. If the function 1757 // has an attribute indicating it is a printf-like function, then we 1758 // should suppress warnings concerning non-literals being used in a call 1759 // to a vprintf function. For example: 1760 // 1761 // void 1762 // logmessage(char const *fmt __attribute__ (format (printf, 1, 2)), ...){ 1763 // va_list ap; 1764 // va_start(ap, fmt); 1765 // vprintf(fmt, ap); // Do NOT emit a warning about "fmt". 1766 // ... 1767 // 1768 if (HasVAListArg) { 1769 if (const ParmVarDecl *PV = dyn_cast<ParmVarDecl>(VD)) { 1770 if (const NamedDecl *ND = dyn_cast<NamedDecl>(PV->getDeclContext())) { 1771 int PVIndex = PV->getFunctionScopeIndex() + 1; 1772 for (specific_attr_iterator<FormatAttr> 1773 i = ND->specific_attr_begin<FormatAttr>(), 1774 e = ND->specific_attr_end<FormatAttr>(); i != e ; ++i) { 1775 FormatAttr *PVFormat = *i; 1776 // adjust for implicit parameter 1777 if (const CXXMethodDecl *MD = dyn_cast<CXXMethodDecl>(ND)) 1778 if (MD->isInstance()) 1779 ++PVIndex; 1780 // We also check if the formats are compatible. 1781 // We can't pass a 'scanf' string to a 'printf' function. 1782 if (PVIndex == PVFormat->getFormatIdx() && 1783 Type == GetFormatStringType(PVFormat)) 1784 return SLCT_UncheckedLiteral; 1785 } 1786 } 1787 } 1788 } 1789 } 1790 1791 return SLCT_NotALiteral; 1792 } 1793 1794 case Stmt::CallExprClass: 1795 case Stmt::CXXMemberCallExprClass: { 1796 const CallExpr *CE = cast<CallExpr>(E); 1797 if (const NamedDecl *ND = dyn_cast_or_null<NamedDecl>(CE->getCalleeDecl())) { 1798 if (const FormatArgAttr *FA = ND->getAttr<FormatArgAttr>()) { 1799 unsigned ArgIndex = FA->getFormatIdx(); 1800 if (const CXXMethodDecl *MD = dyn_cast<CXXMethodDecl>(ND)) 1801 if (MD->isInstance()) 1802 --ArgIndex; 1803 const Expr *Arg = CE->getArg(ArgIndex - 1); 1804 1805 return checkFormatStringExpr(Arg, Args, 1806 HasVAListArg, format_idx, firstDataArg, 1807 Type, CallType, inFunctionCall); 1808 } else if (const FunctionDecl *FD = dyn_cast<FunctionDecl>(ND)) { 1809 unsigned BuiltinID = FD->getBuiltinID(); 1810 if (BuiltinID == Builtin::BI__builtin___CFStringMakeConstantString || 1811 BuiltinID == Builtin::BI__builtin___NSStringMakeConstantString) { 1812 const Expr *Arg = CE->getArg(0); 1813 return checkFormatStringExpr(Arg, Args, 1814 HasVAListArg, format_idx, 1815 firstDataArg, Type, CallType, 1816 inFunctionCall); 1817 } 1818 } 1819 } 1820 1821 return SLCT_NotALiteral; 1822 } 1823 case Stmt::ObjCStringLiteralClass: 1824 case Stmt::StringLiteralClass: { 1825 const StringLiteral *StrE = NULL; 1826 1827 if (const ObjCStringLiteral *ObjCFExpr = dyn_cast<ObjCStringLiteral>(E)) 1828 StrE = ObjCFExpr->getString(); 1829 else 1830 StrE = cast<StringLiteral>(E); 1831 1832 if (StrE) { 1833 CheckFormatString(StrE, E, Args, HasVAListArg, format_idx, 1834 firstDataArg, Type, inFunctionCall, CallType); 1835 return SLCT_CheckedLiteral; 1836 } 1837 1838 return SLCT_NotALiteral; 1839 } 1840 1841 default: 1842 return SLCT_NotALiteral; 1843 } 1844} 1845 1846void 1847Sema::CheckNonNullArguments(const NonNullAttr *NonNull, 1848 const Expr * const *ExprArgs, 1849 SourceLocation CallSiteLoc) { 1850 for (NonNullAttr::args_iterator i = NonNull->args_begin(), 1851 e = NonNull->args_end(); 1852 i != e; ++i) { 1853 const Expr *ArgExpr = ExprArgs[*i]; 1854 1855 // As a special case, transparent unions initialized with zero are 1856 // considered null for the purposes of the nonnull attribute. 1857 if (const RecordType *UT = ArgExpr->getType()->getAsUnionType()) { 1858 if (UT->getDecl()->hasAttr<TransparentUnionAttr>()) 1859 if (const CompoundLiteralExpr *CLE = 1860 dyn_cast<CompoundLiteralExpr>(ArgExpr)) 1861 if (const InitListExpr *ILE = 1862 dyn_cast<InitListExpr>(CLE->getInitializer())) 1863 ArgExpr = ILE->getInit(0); 1864 } 1865 1866 bool Result; 1867 if (ArgExpr->EvaluateAsBooleanCondition(Result, Context) && !Result) 1868 Diag(CallSiteLoc, diag::warn_null_arg) << ArgExpr->getSourceRange(); 1869 } 1870} 1871 1872Sema::FormatStringType Sema::GetFormatStringType(const FormatAttr *Format) { 1873 return llvm::StringSwitch<FormatStringType>(Format->getType()) 1874 .Case("scanf", FST_Scanf) 1875 .Cases("printf", "printf0", FST_Printf) 1876 .Cases("NSString", "CFString", FST_NSString) 1877 .Case("strftime", FST_Strftime) 1878 .Case("strfmon", FST_Strfmon) 1879 .Cases("kprintf", "cmn_err", "vcmn_err", "zcmn_err", FST_Kprintf) 1880 .Default(FST_Unknown); 1881} 1882 1883/// CheckFormatArguments - Check calls to printf and scanf (and similar 1884/// functions) for correct use of format strings. 1885/// Returns true if a format string has been fully checked. 1886bool Sema::CheckFormatArguments(const FormatAttr *Format, 1887 ArrayRef<const Expr *> Args, 1888 bool IsCXXMember, 1889 VariadicCallType CallType, 1890 SourceLocation Loc, SourceRange Range) { 1891 FormatStringInfo FSI; 1892 if (getFormatStringInfo(Format, IsCXXMember, &FSI)) 1893 return CheckFormatArguments(Args, FSI.HasVAListArg, FSI.FormatIdx, 1894 FSI.FirstDataArg, GetFormatStringType(Format), 1895 CallType, Loc, Range); 1896 return false; 1897} 1898 1899bool Sema::CheckFormatArguments(ArrayRef<const Expr *> Args, 1900 bool HasVAListArg, unsigned format_idx, 1901 unsigned firstDataArg, FormatStringType Type, 1902 VariadicCallType CallType, 1903 SourceLocation Loc, SourceRange Range) { 1904 // CHECK: printf/scanf-like function is called with no format string. 1905 if (format_idx >= Args.size()) { 1906 Diag(Loc, diag::warn_missing_format_string) << Range; 1907 return false; 1908 } 1909 1910 const Expr *OrigFormatExpr = Args[format_idx]->IgnoreParenCasts(); 1911 1912 // CHECK: format string is not a string literal. 1913 // 1914 // Dynamically generated format strings are difficult to 1915 // automatically vet at compile time. Requiring that format strings 1916 // are string literals: (1) permits the checking of format strings by 1917 // the compiler and thereby (2) can practically remove the source of 1918 // many format string exploits. 1919 1920 // Format string can be either ObjC string (e.g. @"%d") or 1921 // C string (e.g. "%d") 1922 // ObjC string uses the same format specifiers as C string, so we can use 1923 // the same format string checking logic for both ObjC and C strings. 1924 StringLiteralCheckType CT = 1925 checkFormatStringExpr(OrigFormatExpr, Args, HasVAListArg, 1926 format_idx, firstDataArg, Type, CallType); 1927 if (CT != SLCT_NotALiteral) 1928 // Literal format string found, check done! 1929 return CT == SLCT_CheckedLiteral; 1930 1931 // Strftime is particular as it always uses a single 'time' argument, 1932 // so it is safe to pass a non-literal string. 1933 if (Type == FST_Strftime) 1934 return false; 1935 1936 // Do not emit diag when the string param is a macro expansion and the 1937 // format is either NSString or CFString. This is a hack to prevent 1938 // diag when using the NSLocalizedString and CFCopyLocalizedString macros 1939 // which are usually used in place of NS and CF string literals. 1940 if (Type == FST_NSString && 1941 SourceMgr.isInSystemMacro(Args[format_idx]->getLocStart())) 1942 return false; 1943 1944 // If there are no arguments specified, warn with -Wformat-security, otherwise 1945 // warn only with -Wformat-nonliteral. 1946 if (Args.size() == format_idx+1) 1947 Diag(Args[format_idx]->getLocStart(), 1948 diag::warn_format_nonliteral_noargs) 1949 << OrigFormatExpr->getSourceRange(); 1950 else 1951 Diag(Args[format_idx]->getLocStart(), 1952 diag::warn_format_nonliteral) 1953 << OrigFormatExpr->getSourceRange(); 1954 return false; 1955} 1956 1957namespace { 1958class CheckFormatHandler : public analyze_format_string::FormatStringHandler { 1959protected: 1960 Sema &S; 1961 const StringLiteral *FExpr; 1962 const Expr *OrigFormatExpr; 1963 const unsigned FirstDataArg; 1964 const unsigned NumDataArgs; 1965 const char *Beg; // Start of format string. 1966 const bool HasVAListArg; 1967 ArrayRef<const Expr *> Args; 1968 unsigned FormatIdx; 1969 llvm::BitVector CoveredArgs; 1970 bool usesPositionalArgs; 1971 bool atFirstArg; 1972 bool inFunctionCall; 1973 Sema::VariadicCallType CallType; 1974public: 1975 CheckFormatHandler(Sema &s, const StringLiteral *fexpr, 1976 const Expr *origFormatExpr, unsigned firstDataArg, 1977 unsigned numDataArgs, const char *beg, bool hasVAListArg, 1978 ArrayRef<const Expr *> Args, 1979 unsigned formatIdx, bool inFunctionCall, 1980 Sema::VariadicCallType callType) 1981 : S(s), FExpr(fexpr), OrigFormatExpr(origFormatExpr), 1982 FirstDataArg(firstDataArg), NumDataArgs(numDataArgs), 1983 Beg(beg), HasVAListArg(hasVAListArg), 1984 Args(Args), FormatIdx(formatIdx), 1985 usesPositionalArgs(false), atFirstArg(true), 1986 inFunctionCall(inFunctionCall), CallType(callType) { 1987 CoveredArgs.resize(numDataArgs); 1988 CoveredArgs.reset(); 1989 } 1990 1991 void DoneProcessing(); 1992 1993 void HandleIncompleteSpecifier(const char *startSpecifier, 1994 unsigned specifierLen); 1995 1996 void HandleInvalidLengthModifier( 1997 const analyze_format_string::FormatSpecifier &FS, 1998 const analyze_format_string::ConversionSpecifier &CS, 1999 const char *startSpecifier, unsigned specifierLen, unsigned DiagID); 2000 2001 void HandleNonStandardLengthModifier( 2002 const analyze_format_string::FormatSpecifier &FS, 2003 const char *startSpecifier, unsigned specifierLen); 2004 2005 void HandleNonStandardConversionSpecifier( 2006 const analyze_format_string::ConversionSpecifier &CS, 2007 const char *startSpecifier, unsigned specifierLen); 2008 2009 virtual void HandlePosition(const char *startPos, unsigned posLen); 2010 2011 virtual void HandleInvalidPosition(const char *startSpecifier, 2012 unsigned specifierLen, 2013 analyze_format_string::PositionContext p); 2014 2015 virtual void HandleZeroPosition(const char *startPos, unsigned posLen); 2016 2017 void HandleNullChar(const char *nullCharacter); 2018 2019 template <typename Range> 2020 static void EmitFormatDiagnostic(Sema &S, bool inFunctionCall, 2021 const Expr *ArgumentExpr, 2022 PartialDiagnostic PDiag, 2023 SourceLocation StringLoc, 2024 bool IsStringLocation, Range StringRange, 2025 ArrayRef<FixItHint> Fixit = None); 2026 2027protected: 2028 bool HandleInvalidConversionSpecifier(unsigned argIndex, SourceLocation Loc, 2029 const char *startSpec, 2030 unsigned specifierLen, 2031 const char *csStart, unsigned csLen); 2032 2033 void HandlePositionalNonpositionalArgs(SourceLocation Loc, 2034 const char *startSpec, 2035 unsigned specifierLen); 2036 2037 SourceRange getFormatStringRange(); 2038 CharSourceRange getSpecifierRange(const char *startSpecifier, 2039 unsigned specifierLen); 2040 SourceLocation getLocationOfByte(const char *x); 2041 2042 const Expr *getDataArg(unsigned i) const; 2043 2044 bool CheckNumArgs(const analyze_format_string::FormatSpecifier &FS, 2045 const analyze_format_string::ConversionSpecifier &CS, 2046 const char *startSpecifier, unsigned specifierLen, 2047 unsigned argIndex); 2048 2049 template <typename Range> 2050 void EmitFormatDiagnostic(PartialDiagnostic PDiag, SourceLocation StringLoc, 2051 bool IsStringLocation, Range StringRange, 2052 ArrayRef<FixItHint> Fixit = None); 2053 2054 void CheckPositionalAndNonpositionalArgs( 2055 const analyze_format_string::FormatSpecifier *FS); 2056}; 2057} 2058 2059SourceRange CheckFormatHandler::getFormatStringRange() { 2060 return OrigFormatExpr->getSourceRange(); 2061} 2062 2063CharSourceRange CheckFormatHandler:: 2064getSpecifierRange(const char *startSpecifier, unsigned specifierLen) { 2065 SourceLocation Start = getLocationOfByte(startSpecifier); 2066 SourceLocation End = getLocationOfByte(startSpecifier + specifierLen - 1); 2067 2068 // Advance the end SourceLocation by one due to half-open ranges. 2069 End = End.getLocWithOffset(1); 2070 2071 return CharSourceRange::getCharRange(Start, End); 2072} 2073 2074SourceLocation CheckFormatHandler::getLocationOfByte(const char *x) { 2075 return S.getLocationOfStringLiteralByte(FExpr, x - Beg); 2076} 2077 2078void CheckFormatHandler::HandleIncompleteSpecifier(const char *startSpecifier, 2079 unsigned specifierLen){ 2080 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_incomplete_specifier), 2081 getLocationOfByte(startSpecifier), 2082 /*IsStringLocation*/true, 2083 getSpecifierRange(startSpecifier, specifierLen)); 2084} 2085 2086void CheckFormatHandler::HandleInvalidLengthModifier( 2087 const analyze_format_string::FormatSpecifier &FS, 2088 const analyze_format_string::ConversionSpecifier &CS, 2089 const char *startSpecifier, unsigned specifierLen, unsigned DiagID) { 2090 using namespace analyze_format_string; 2091 2092 const LengthModifier &LM = FS.getLengthModifier(); 2093 CharSourceRange LMRange = getSpecifierRange(LM.getStart(), LM.getLength()); 2094 2095 // See if we know how to fix this length modifier. 2096 Optional<LengthModifier> FixedLM = FS.getCorrectedLengthModifier(); 2097 if (FixedLM) { 2098 EmitFormatDiagnostic(S.PDiag(DiagID) << LM.toString() << CS.toString(), 2099 getLocationOfByte(LM.getStart()), 2100 /*IsStringLocation*/true, 2101 getSpecifierRange(startSpecifier, specifierLen)); 2102 2103 S.Diag(getLocationOfByte(LM.getStart()), diag::note_format_fix_specifier) 2104 << FixedLM->toString() 2105 << FixItHint::CreateReplacement(LMRange, FixedLM->toString()); 2106 2107 } else { 2108 FixItHint Hint; 2109 if (DiagID == diag::warn_format_nonsensical_length) 2110 Hint = FixItHint::CreateRemoval(LMRange); 2111 2112 EmitFormatDiagnostic(S.PDiag(DiagID) << LM.toString() << CS.toString(), 2113 getLocationOfByte(LM.getStart()), 2114 /*IsStringLocation*/true, 2115 getSpecifierRange(startSpecifier, specifierLen), 2116 Hint); 2117 } 2118} 2119 2120void CheckFormatHandler::HandleNonStandardLengthModifier( 2121 const analyze_format_string::FormatSpecifier &FS, 2122 const char *startSpecifier, unsigned specifierLen) { 2123 using namespace analyze_format_string; 2124 2125 const LengthModifier &LM = FS.getLengthModifier(); 2126 CharSourceRange LMRange = getSpecifierRange(LM.getStart(), LM.getLength()); 2127 2128 // See if we know how to fix this length modifier. 2129 Optional<LengthModifier> FixedLM = FS.getCorrectedLengthModifier(); 2130 if (FixedLM) { 2131 EmitFormatDiagnostic(S.PDiag(diag::warn_format_non_standard) 2132 << LM.toString() << 0, 2133 getLocationOfByte(LM.getStart()), 2134 /*IsStringLocation*/true, 2135 getSpecifierRange(startSpecifier, specifierLen)); 2136 2137 S.Diag(getLocationOfByte(LM.getStart()), diag::note_format_fix_specifier) 2138 << FixedLM->toString() 2139 << FixItHint::CreateReplacement(LMRange, FixedLM->toString()); 2140 2141 } else { 2142 EmitFormatDiagnostic(S.PDiag(diag::warn_format_non_standard) 2143 << LM.toString() << 0, 2144 getLocationOfByte(LM.getStart()), 2145 /*IsStringLocation*/true, 2146 getSpecifierRange(startSpecifier, specifierLen)); 2147 } 2148} 2149 2150void CheckFormatHandler::HandleNonStandardConversionSpecifier( 2151 const analyze_format_string::ConversionSpecifier &CS, 2152 const char *startSpecifier, unsigned specifierLen) { 2153 using namespace analyze_format_string; 2154 2155 // See if we know how to fix this conversion specifier. 2156 Optional<ConversionSpecifier> FixedCS = CS.getStandardSpecifier(); 2157 if (FixedCS) { 2158 EmitFormatDiagnostic(S.PDiag(diag::warn_format_non_standard) 2159 << CS.toString() << /*conversion specifier*/1, 2160 getLocationOfByte(CS.getStart()), 2161 /*IsStringLocation*/true, 2162 getSpecifierRange(startSpecifier, specifierLen)); 2163 2164 CharSourceRange CSRange = getSpecifierRange(CS.getStart(), CS.getLength()); 2165 S.Diag(getLocationOfByte(CS.getStart()), diag::note_format_fix_specifier) 2166 << FixedCS->toString() 2167 << FixItHint::CreateReplacement(CSRange, FixedCS->toString()); 2168 } else { 2169 EmitFormatDiagnostic(S.PDiag(diag::warn_format_non_standard) 2170 << CS.toString() << /*conversion specifier*/1, 2171 getLocationOfByte(CS.getStart()), 2172 /*IsStringLocation*/true, 2173 getSpecifierRange(startSpecifier, specifierLen)); 2174 } 2175} 2176 2177void CheckFormatHandler::HandlePosition(const char *startPos, 2178 unsigned posLen) { 2179 EmitFormatDiagnostic(S.PDiag(diag::warn_format_non_standard_positional_arg), 2180 getLocationOfByte(startPos), 2181 /*IsStringLocation*/true, 2182 getSpecifierRange(startPos, posLen)); 2183} 2184 2185void 2186CheckFormatHandler::HandleInvalidPosition(const char *startPos, unsigned posLen, 2187 analyze_format_string::PositionContext p) { 2188 EmitFormatDiagnostic(S.PDiag(diag::warn_format_invalid_positional_specifier) 2189 << (unsigned) p, 2190 getLocationOfByte(startPos), /*IsStringLocation*/true, 2191 getSpecifierRange(startPos, posLen)); 2192} 2193 2194void CheckFormatHandler::HandleZeroPosition(const char *startPos, 2195 unsigned posLen) { 2196 EmitFormatDiagnostic(S.PDiag(diag::warn_format_zero_positional_specifier), 2197 getLocationOfByte(startPos), 2198 /*IsStringLocation*/true, 2199 getSpecifierRange(startPos, posLen)); 2200} 2201 2202void CheckFormatHandler::HandleNullChar(const char *nullCharacter) { 2203 if (!isa<ObjCStringLiteral>(OrigFormatExpr)) { 2204 // The presence of a null character is likely an error. 2205 EmitFormatDiagnostic( 2206 S.PDiag(diag::warn_printf_format_string_contains_null_char), 2207 getLocationOfByte(nullCharacter), /*IsStringLocation*/true, 2208 getFormatStringRange()); 2209 } 2210} 2211 2212// Note that this may return NULL if there was an error parsing or building 2213// one of the argument expressions. 2214const Expr *CheckFormatHandler::getDataArg(unsigned i) const { 2215 return Args[FirstDataArg + i]; 2216} 2217 2218void CheckFormatHandler::DoneProcessing() { 2219 // Does the number of data arguments exceed the number of 2220 // format conversions in the format string? 2221 if (!HasVAListArg) { 2222 // Find any arguments that weren't covered. 2223 CoveredArgs.flip(); 2224 signed notCoveredArg = CoveredArgs.find_first(); 2225 if (notCoveredArg >= 0) { 2226 assert((unsigned)notCoveredArg < NumDataArgs); 2227 if (const Expr *E = getDataArg((unsigned) notCoveredArg)) { 2228 SourceLocation Loc = E->getLocStart(); 2229 if (!S.getSourceManager().isInSystemMacro(Loc)) { 2230 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_data_arg_not_used), 2231 Loc, /*IsStringLocation*/false, 2232 getFormatStringRange()); 2233 } 2234 } 2235 } 2236 } 2237} 2238 2239bool 2240CheckFormatHandler::HandleInvalidConversionSpecifier(unsigned argIndex, 2241 SourceLocation Loc, 2242 const char *startSpec, 2243 unsigned specifierLen, 2244 const char *csStart, 2245 unsigned csLen) { 2246 2247 bool keepGoing = true; 2248 if (argIndex < NumDataArgs) { 2249 // Consider the argument coverered, even though the specifier doesn't 2250 // make sense. 2251 CoveredArgs.set(argIndex); 2252 } 2253 else { 2254 // If argIndex exceeds the number of data arguments we 2255 // don't issue a warning because that is just a cascade of warnings (and 2256 // they may have intended '%%' anyway). We don't want to continue processing 2257 // the format string after this point, however, as we will like just get 2258 // gibberish when trying to match arguments. 2259 keepGoing = false; 2260 } 2261 2262 EmitFormatDiagnostic(S.PDiag(diag::warn_format_invalid_conversion) 2263 << StringRef(csStart, csLen), 2264 Loc, /*IsStringLocation*/true, 2265 getSpecifierRange(startSpec, specifierLen)); 2266 2267 return keepGoing; 2268} 2269 2270void 2271CheckFormatHandler::HandlePositionalNonpositionalArgs(SourceLocation Loc, 2272 const char *startSpec, 2273 unsigned specifierLen) { 2274 EmitFormatDiagnostic( 2275 S.PDiag(diag::warn_format_mix_positional_nonpositional_args), 2276 Loc, /*isStringLoc*/true, getSpecifierRange(startSpec, specifierLen)); 2277} 2278 2279bool 2280CheckFormatHandler::CheckNumArgs( 2281 const analyze_format_string::FormatSpecifier &FS, 2282 const analyze_format_string::ConversionSpecifier &CS, 2283 const char *startSpecifier, unsigned specifierLen, unsigned argIndex) { 2284 2285 if (argIndex >= NumDataArgs) { 2286 PartialDiagnostic PDiag = FS.usesPositionalArg() 2287 ? (S.PDiag(diag::warn_printf_positional_arg_exceeds_data_args) 2288 << (argIndex+1) << NumDataArgs) 2289 : S.PDiag(diag::warn_printf_insufficient_data_args); 2290 EmitFormatDiagnostic( 2291 PDiag, getLocationOfByte(CS.getStart()), /*IsStringLocation*/true, 2292 getSpecifierRange(startSpecifier, specifierLen)); 2293 return false; 2294 } 2295 return true; 2296} 2297 2298template<typename Range> 2299void CheckFormatHandler::EmitFormatDiagnostic(PartialDiagnostic PDiag, 2300 SourceLocation Loc, 2301 bool IsStringLocation, 2302 Range StringRange, 2303 ArrayRef<FixItHint> FixIt) { 2304 EmitFormatDiagnostic(S, inFunctionCall, Args[FormatIdx], PDiag, 2305 Loc, IsStringLocation, StringRange, FixIt); 2306} 2307 2308/// \brief If the format string is not within the funcion call, emit a note 2309/// so that the function call and string are in diagnostic messages. 2310/// 2311/// \param InFunctionCall if true, the format string is within the function 2312/// call and only one diagnostic message will be produced. Otherwise, an 2313/// extra note will be emitted pointing to location of the format string. 2314/// 2315/// \param ArgumentExpr the expression that is passed as the format string 2316/// argument in the function call. Used for getting locations when two 2317/// diagnostics are emitted. 2318/// 2319/// \param PDiag the callee should already have provided any strings for the 2320/// diagnostic message. This function only adds locations and fixits 2321/// to diagnostics. 2322/// 2323/// \param Loc primary location for diagnostic. If two diagnostics are 2324/// required, one will be at Loc and a new SourceLocation will be created for 2325/// the other one. 2326/// 2327/// \param IsStringLocation if true, Loc points to the format string should be 2328/// used for the note. Otherwise, Loc points to the argument list and will 2329/// be used with PDiag. 2330/// 2331/// \param StringRange some or all of the string to highlight. This is 2332/// templated so it can accept either a CharSourceRange or a SourceRange. 2333/// 2334/// \param FixIt optional fix it hint for the format string. 2335template<typename Range> 2336void CheckFormatHandler::EmitFormatDiagnostic(Sema &S, bool InFunctionCall, 2337 const Expr *ArgumentExpr, 2338 PartialDiagnostic PDiag, 2339 SourceLocation Loc, 2340 bool IsStringLocation, 2341 Range StringRange, 2342 ArrayRef<FixItHint> FixIt) { 2343 if (InFunctionCall) { 2344 const Sema::SemaDiagnosticBuilder &D = S.Diag(Loc, PDiag); 2345 D << StringRange; 2346 for (ArrayRef<FixItHint>::iterator I = FixIt.begin(), E = FixIt.end(); 2347 I != E; ++I) { 2348 D << *I; 2349 } 2350 } else { 2351 S.Diag(IsStringLocation ? ArgumentExpr->getExprLoc() : Loc, PDiag) 2352 << ArgumentExpr->getSourceRange(); 2353 2354 const Sema::SemaDiagnosticBuilder &Note = 2355 S.Diag(IsStringLocation ? Loc : StringRange.getBegin(), 2356 diag::note_format_string_defined); 2357 2358 Note << StringRange; 2359 for (ArrayRef<FixItHint>::iterator I = FixIt.begin(), E = FixIt.end(); 2360 I != E; ++I) { 2361 Note << *I; 2362 } 2363 } 2364} 2365 2366//===--- CHECK: Printf format string checking ------------------------------===// 2367 2368namespace { 2369class CheckPrintfHandler : public CheckFormatHandler { 2370 bool ObjCContext; 2371public: 2372 CheckPrintfHandler(Sema &s, const StringLiteral *fexpr, 2373 const Expr *origFormatExpr, unsigned firstDataArg, 2374 unsigned numDataArgs, bool isObjC, 2375 const char *beg, bool hasVAListArg, 2376 ArrayRef<const Expr *> Args, 2377 unsigned formatIdx, bool inFunctionCall, 2378 Sema::VariadicCallType CallType) 2379 : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg, 2380 numDataArgs, beg, hasVAListArg, Args, 2381 formatIdx, inFunctionCall, CallType), ObjCContext(isObjC) 2382 {} 2383 2384 2385 bool HandleInvalidPrintfConversionSpecifier( 2386 const analyze_printf::PrintfSpecifier &FS, 2387 const char *startSpecifier, 2388 unsigned specifierLen); 2389 2390 bool HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier &FS, 2391 const char *startSpecifier, 2392 unsigned specifierLen); 2393 bool checkFormatExpr(const analyze_printf::PrintfSpecifier &FS, 2394 const char *StartSpecifier, 2395 unsigned SpecifierLen, 2396 const Expr *E); 2397 2398 bool HandleAmount(const analyze_format_string::OptionalAmount &Amt, unsigned k, 2399 const char *startSpecifier, unsigned specifierLen); 2400 void HandleInvalidAmount(const analyze_printf::PrintfSpecifier &FS, 2401 const analyze_printf::OptionalAmount &Amt, 2402 unsigned type, 2403 const char *startSpecifier, unsigned specifierLen); 2404 void HandleFlag(const analyze_printf::PrintfSpecifier &FS, 2405 const analyze_printf::OptionalFlag &flag, 2406 const char *startSpecifier, unsigned specifierLen); 2407 void HandleIgnoredFlag(const analyze_printf::PrintfSpecifier &FS, 2408 const analyze_printf::OptionalFlag &ignoredFlag, 2409 const analyze_printf::OptionalFlag &flag, 2410 const char *startSpecifier, unsigned specifierLen); 2411 bool checkForCStrMembers(const analyze_printf::ArgType &AT, 2412 const Expr *E, const CharSourceRange &CSR); 2413 2414}; 2415} 2416 2417bool CheckPrintfHandler::HandleInvalidPrintfConversionSpecifier( 2418 const analyze_printf::PrintfSpecifier &FS, 2419 const char *startSpecifier, 2420 unsigned specifierLen) { 2421 const analyze_printf::PrintfConversionSpecifier &CS = 2422 FS.getConversionSpecifier(); 2423 2424 return HandleInvalidConversionSpecifier(FS.getArgIndex(), 2425 getLocationOfByte(CS.getStart()), 2426 startSpecifier, specifierLen, 2427 CS.getStart(), CS.getLength()); 2428} 2429 2430bool CheckPrintfHandler::HandleAmount( 2431 const analyze_format_string::OptionalAmount &Amt, 2432 unsigned k, const char *startSpecifier, 2433 unsigned specifierLen) { 2434 2435 if (Amt.hasDataArgument()) { 2436 if (!HasVAListArg) { 2437 unsigned argIndex = Amt.getArgIndex(); 2438 if (argIndex >= NumDataArgs) { 2439 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_asterisk_missing_arg) 2440 << k, 2441 getLocationOfByte(Amt.getStart()), 2442 /*IsStringLocation*/true, 2443 getSpecifierRange(startSpecifier, specifierLen)); 2444 // Don't do any more checking. We will just emit 2445 // spurious errors. 2446 return false; 2447 } 2448 2449 // Type check the data argument. It should be an 'int'. 2450 // Although not in conformance with C99, we also allow the argument to be 2451 // an 'unsigned int' as that is a reasonably safe case. GCC also 2452 // doesn't emit a warning for that case. 2453 CoveredArgs.set(argIndex); 2454 const Expr *Arg = getDataArg(argIndex); 2455 if (!Arg) 2456 return false; 2457 2458 QualType T = Arg->getType(); 2459 2460 const analyze_printf::ArgType &AT = Amt.getArgType(S.Context); 2461 assert(AT.isValid()); 2462 2463 if (!AT.matchesType(S.Context, T)) { 2464 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_asterisk_wrong_type) 2465 << k << AT.getRepresentativeTypeName(S.Context) 2466 << T << Arg->getSourceRange(), 2467 getLocationOfByte(Amt.getStart()), 2468 /*IsStringLocation*/true, 2469 getSpecifierRange(startSpecifier, specifierLen)); 2470 // Don't do any more checking. We will just emit 2471 // spurious errors. 2472 return false; 2473 } 2474 } 2475 } 2476 return true; 2477} 2478 2479void CheckPrintfHandler::HandleInvalidAmount( 2480 const analyze_printf::PrintfSpecifier &FS, 2481 const analyze_printf::OptionalAmount &Amt, 2482 unsigned type, 2483 const char *startSpecifier, 2484 unsigned specifierLen) { 2485 const analyze_printf::PrintfConversionSpecifier &CS = 2486 FS.getConversionSpecifier(); 2487 2488 FixItHint fixit = 2489 Amt.getHowSpecified() == analyze_printf::OptionalAmount::Constant 2490 ? FixItHint::CreateRemoval(getSpecifierRange(Amt.getStart(), 2491 Amt.getConstantLength())) 2492 : FixItHint(); 2493 2494 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_nonsensical_optional_amount) 2495 << type << CS.toString(), 2496 getLocationOfByte(Amt.getStart()), 2497 /*IsStringLocation*/true, 2498 getSpecifierRange(startSpecifier, specifierLen), 2499 fixit); 2500} 2501 2502void CheckPrintfHandler::HandleFlag(const analyze_printf::PrintfSpecifier &FS, 2503 const analyze_printf::OptionalFlag &flag, 2504 const char *startSpecifier, 2505 unsigned specifierLen) { 2506 // Warn about pointless flag with a fixit removal. 2507 const analyze_printf::PrintfConversionSpecifier &CS = 2508 FS.getConversionSpecifier(); 2509 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_nonsensical_flag) 2510 << flag.toString() << CS.toString(), 2511 getLocationOfByte(flag.getPosition()), 2512 /*IsStringLocation*/true, 2513 getSpecifierRange(startSpecifier, specifierLen), 2514 FixItHint::CreateRemoval( 2515 getSpecifierRange(flag.getPosition(), 1))); 2516} 2517 2518void CheckPrintfHandler::HandleIgnoredFlag( 2519 const analyze_printf::PrintfSpecifier &FS, 2520 const analyze_printf::OptionalFlag &ignoredFlag, 2521 const analyze_printf::OptionalFlag &flag, 2522 const char *startSpecifier, 2523 unsigned specifierLen) { 2524 // Warn about ignored flag with a fixit removal. 2525 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_ignored_flag) 2526 << ignoredFlag.toString() << flag.toString(), 2527 getLocationOfByte(ignoredFlag.getPosition()), 2528 /*IsStringLocation*/true, 2529 getSpecifierRange(startSpecifier, specifierLen), 2530 FixItHint::CreateRemoval( 2531 getSpecifierRange(ignoredFlag.getPosition(), 1))); 2532} 2533 2534// Determines if the specified is a C++ class or struct containing 2535// a member with the specified name and kind (e.g. a CXXMethodDecl named 2536// "c_str()"). 2537template<typename MemberKind> 2538static llvm::SmallPtrSet<MemberKind*, 1> 2539CXXRecordMembersNamed(StringRef Name, Sema &S, QualType Ty) { 2540 const RecordType *RT = Ty->getAs<RecordType>(); 2541 llvm::SmallPtrSet<MemberKind*, 1> Results; 2542 2543 if (!RT) 2544 return Results; 2545 const CXXRecordDecl *RD = dyn_cast<CXXRecordDecl>(RT->getDecl()); 2546 if (!RD) 2547 return Results; 2548 2549 LookupResult R(S, &S.PP.getIdentifierTable().get(Name), SourceLocation(), 2550 Sema::LookupMemberName); 2551 2552 // We just need to include all members of the right kind turned up by the 2553 // filter, at this point. 2554 if (S.LookupQualifiedName(R, RT->getDecl())) 2555 for (LookupResult::iterator I = R.begin(), E = R.end(); I != E; ++I) { 2556 NamedDecl *decl = (*I)->getUnderlyingDecl(); 2557 if (MemberKind *FK = dyn_cast<MemberKind>(decl)) 2558 Results.insert(FK); 2559 } 2560 return Results; 2561} 2562 2563// Check if a (w)string was passed when a (w)char* was needed, and offer a 2564// better diagnostic if so. AT is assumed to be valid. 2565// Returns true when a c_str() conversion method is found. 2566bool CheckPrintfHandler::checkForCStrMembers( 2567 const analyze_printf::ArgType &AT, const Expr *E, 2568 const CharSourceRange &CSR) { 2569 typedef llvm::SmallPtrSet<CXXMethodDecl*, 1> MethodSet; 2570 2571 MethodSet Results = 2572 CXXRecordMembersNamed<CXXMethodDecl>("c_str", S, E->getType()); 2573 2574 for (MethodSet::iterator MI = Results.begin(), ME = Results.end(); 2575 MI != ME; ++MI) { 2576 const CXXMethodDecl *Method = *MI; 2577 if (Method->getNumParams() == 0 && 2578 AT.matchesType(S.Context, Method->getResultType())) { 2579 // FIXME: Suggest parens if the expression needs them. 2580 SourceLocation EndLoc = 2581 S.getPreprocessor().getLocForEndOfToken(E->getLocEnd()); 2582 S.Diag(E->getLocStart(), diag::note_printf_c_str) 2583 << "c_str()" 2584 << FixItHint::CreateInsertion(EndLoc, ".c_str()"); 2585 return true; 2586 } 2587 } 2588 2589 return false; 2590} 2591 2592bool 2593CheckPrintfHandler::HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier 2594 &FS, 2595 const char *startSpecifier, 2596 unsigned specifierLen) { 2597 2598 using namespace analyze_format_string; 2599 using namespace analyze_printf; 2600 const PrintfConversionSpecifier &CS = FS.getConversionSpecifier(); 2601 2602 if (FS.consumesDataArgument()) { 2603 if (atFirstArg) { 2604 atFirstArg = false; 2605 usesPositionalArgs = FS.usesPositionalArg(); 2606 } 2607 else if (usesPositionalArgs != FS.usesPositionalArg()) { 2608 HandlePositionalNonpositionalArgs(getLocationOfByte(CS.getStart()), 2609 startSpecifier, specifierLen); 2610 return false; 2611 } 2612 } 2613 2614 // First check if the field width, precision, and conversion specifier 2615 // have matching data arguments. 2616 if (!HandleAmount(FS.getFieldWidth(), /* field width */ 0, 2617 startSpecifier, specifierLen)) { 2618 return false; 2619 } 2620 2621 if (!HandleAmount(FS.getPrecision(), /* precision */ 1, 2622 startSpecifier, specifierLen)) { 2623 return false; 2624 } 2625 2626 if (!CS.consumesDataArgument()) { 2627 // FIXME: Technically specifying a precision or field width here 2628 // makes no sense. Worth issuing a warning at some point. 2629 return true; 2630 } 2631 2632 // Consume the argument. 2633 unsigned argIndex = FS.getArgIndex(); 2634 if (argIndex < NumDataArgs) { 2635 // The check to see if the argIndex is valid will come later. 2636 // We set the bit here because we may exit early from this 2637 // function if we encounter some other error. 2638 CoveredArgs.set(argIndex); 2639 } 2640 2641 // Check for using an Objective-C specific conversion specifier 2642 // in a non-ObjC literal. 2643 if (!ObjCContext && CS.isObjCArg()) { 2644 return HandleInvalidPrintfConversionSpecifier(FS, startSpecifier, 2645 specifierLen); 2646 } 2647 2648 // Check for invalid use of field width 2649 if (!FS.hasValidFieldWidth()) { 2650 HandleInvalidAmount(FS, FS.getFieldWidth(), /* field width */ 0, 2651 startSpecifier, specifierLen); 2652 } 2653 2654 // Check for invalid use of precision 2655 if (!FS.hasValidPrecision()) { 2656 HandleInvalidAmount(FS, FS.getPrecision(), /* precision */ 1, 2657 startSpecifier, specifierLen); 2658 } 2659 2660 // Check each flag does not conflict with any other component. 2661 if (!FS.hasValidThousandsGroupingPrefix()) 2662 HandleFlag(FS, FS.hasThousandsGrouping(), startSpecifier, specifierLen); 2663 if (!FS.hasValidLeadingZeros()) 2664 HandleFlag(FS, FS.hasLeadingZeros(), startSpecifier, specifierLen); 2665 if (!FS.hasValidPlusPrefix()) 2666 HandleFlag(FS, FS.hasPlusPrefix(), startSpecifier, specifierLen); 2667 if (!FS.hasValidSpacePrefix()) 2668 HandleFlag(FS, FS.hasSpacePrefix(), startSpecifier, specifierLen); 2669 if (!FS.hasValidAlternativeForm()) 2670 HandleFlag(FS, FS.hasAlternativeForm(), startSpecifier, specifierLen); 2671 if (!FS.hasValidLeftJustified()) 2672 HandleFlag(FS, FS.isLeftJustified(), startSpecifier, specifierLen); 2673 2674 // Check that flags are not ignored by another flag 2675 if (FS.hasSpacePrefix() && FS.hasPlusPrefix()) // ' ' ignored by '+' 2676 HandleIgnoredFlag(FS, FS.hasSpacePrefix(), FS.hasPlusPrefix(), 2677 startSpecifier, specifierLen); 2678 if (FS.hasLeadingZeros() && FS.isLeftJustified()) // '0' ignored by '-' 2679 HandleIgnoredFlag(FS, FS.hasLeadingZeros(), FS.isLeftJustified(), 2680 startSpecifier, specifierLen); 2681 2682 // Check the length modifier is valid with the given conversion specifier. 2683 if (!FS.hasValidLengthModifier(S.getASTContext().getTargetInfo())) 2684 HandleInvalidLengthModifier(FS, CS, startSpecifier, specifierLen, 2685 diag::warn_format_nonsensical_length); 2686 else if (!FS.hasStandardLengthModifier()) 2687 HandleNonStandardLengthModifier(FS, startSpecifier, specifierLen); 2688 else if (!FS.hasStandardLengthConversionCombination()) 2689 HandleInvalidLengthModifier(FS, CS, startSpecifier, specifierLen, 2690 diag::warn_format_non_standard_conversion_spec); 2691 2692 if (!FS.hasStandardConversionSpecifier(S.getLangOpts())) 2693 HandleNonStandardConversionSpecifier(CS, startSpecifier, specifierLen); 2694 2695 // The remaining checks depend on the data arguments. 2696 if (HasVAListArg) 2697 return true; 2698 2699 if (!CheckNumArgs(FS, CS, startSpecifier, specifierLen, argIndex)) 2700 return false; 2701 2702 const Expr *Arg = getDataArg(argIndex); 2703 if (!Arg) 2704 return true; 2705 2706 return checkFormatExpr(FS, startSpecifier, specifierLen, Arg); 2707} 2708 2709static bool requiresParensToAddCast(const Expr *E) { 2710 // FIXME: We should have a general way to reason about operator 2711 // precedence and whether parens are actually needed here. 2712 // Take care of a few common cases where they aren't. 2713 const Expr *Inside = E->IgnoreImpCasts(); 2714 if (const PseudoObjectExpr *POE = dyn_cast<PseudoObjectExpr>(Inside)) 2715 Inside = POE->getSyntacticForm()->IgnoreImpCasts(); 2716 2717 switch (Inside->getStmtClass()) { 2718 case Stmt::ArraySubscriptExprClass: 2719 case Stmt::CallExprClass: 2720 case Stmt::CharacterLiteralClass: 2721 case Stmt::CXXBoolLiteralExprClass: 2722 case Stmt::DeclRefExprClass: 2723 case Stmt::FloatingLiteralClass: 2724 case Stmt::IntegerLiteralClass: 2725 case Stmt::MemberExprClass: 2726 case Stmt::ObjCArrayLiteralClass: 2727 case Stmt::ObjCBoolLiteralExprClass: 2728 case Stmt::ObjCBoxedExprClass: 2729 case Stmt::ObjCDictionaryLiteralClass: 2730 case Stmt::ObjCEncodeExprClass: 2731 case Stmt::ObjCIvarRefExprClass: 2732 case Stmt::ObjCMessageExprClass: 2733 case Stmt::ObjCPropertyRefExprClass: 2734 case Stmt::ObjCStringLiteralClass: 2735 case Stmt::ObjCSubscriptRefExprClass: 2736 case Stmt::ParenExprClass: 2737 case Stmt::StringLiteralClass: 2738 case Stmt::UnaryOperatorClass: 2739 return false; 2740 default: 2741 return true; 2742 } 2743} 2744 2745bool 2746CheckPrintfHandler::checkFormatExpr(const analyze_printf::PrintfSpecifier &FS, 2747 const char *StartSpecifier, 2748 unsigned SpecifierLen, 2749 const Expr *E) { 2750 using namespace analyze_format_string; 2751 using namespace analyze_printf; 2752 // Now type check the data expression that matches the 2753 // format specifier. 2754 const analyze_printf::ArgType &AT = FS.getArgType(S.Context, 2755 ObjCContext); 2756 if (!AT.isValid()) 2757 return true; 2758 2759 QualType ExprTy = E->getType(); 2760 while (const TypeOfExprType *TET = dyn_cast<TypeOfExprType>(ExprTy)) { 2761 ExprTy = TET->getUnderlyingExpr()->getType(); 2762 } 2763 2764 if (AT.matchesType(S.Context, ExprTy)) 2765 return true; 2766 2767 // Look through argument promotions for our error message's reported type. 2768 // This includes the integral and floating promotions, but excludes array 2769 // and function pointer decay; seeing that an argument intended to be a 2770 // string has type 'char [6]' is probably more confusing than 'char *'. 2771 if (const ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(E)) { 2772 if (ICE->getCastKind() == CK_IntegralCast || 2773 ICE->getCastKind() == CK_FloatingCast) { 2774 E = ICE->getSubExpr(); 2775 ExprTy = E->getType(); 2776 2777 // Check if we didn't match because of an implicit cast from a 'char' 2778 // or 'short' to an 'int'. This is done because printf is a varargs 2779 // function. 2780 if (ICE->getType() == S.Context.IntTy || 2781 ICE->getType() == S.Context.UnsignedIntTy) { 2782 // All further checking is done on the subexpression. 2783 if (AT.matchesType(S.Context, ExprTy)) 2784 return true; 2785 } 2786 } 2787 } else if (const CharacterLiteral *CL = dyn_cast<CharacterLiteral>(E)) { 2788 // Special case for 'a', which has type 'int' in C. 2789 // Note, however, that we do /not/ want to treat multibyte constants like 2790 // 'MooV' as characters! This form is deprecated but still exists. 2791 if (ExprTy == S.Context.IntTy) 2792 if (llvm::isUIntN(S.Context.getCharWidth(), CL->getValue())) 2793 ExprTy = S.Context.CharTy; 2794 } 2795 2796 // %C in an Objective-C context prints a unichar, not a wchar_t. 2797 // If the argument is an integer of some kind, believe the %C and suggest 2798 // a cast instead of changing the conversion specifier. 2799 QualType IntendedTy = ExprTy; 2800 if (ObjCContext && 2801 FS.getConversionSpecifier().getKind() == ConversionSpecifier::CArg) { 2802 if (ExprTy->isIntegralOrUnscopedEnumerationType() && 2803 !ExprTy->isCharType()) { 2804 // 'unichar' is defined as a typedef of unsigned short, but we should 2805 // prefer using the typedef if it is visible. 2806 IntendedTy = S.Context.UnsignedShortTy; 2807 2808 LookupResult Result(S, &S.Context.Idents.get("unichar"), E->getLocStart(), 2809 Sema::LookupOrdinaryName); 2810 if (S.LookupName(Result, S.getCurScope())) { 2811 NamedDecl *ND = Result.getFoundDecl(); 2812 if (TypedefNameDecl *TD = dyn_cast<TypedefNameDecl>(ND)) 2813 if (TD->getUnderlyingType() == IntendedTy) 2814 IntendedTy = S.Context.getTypedefType(TD); 2815 } 2816 } 2817 } 2818 2819 // Special-case some of Darwin's platform-independence types by suggesting 2820 // casts to primitive types that are known to be large enough. 2821 bool ShouldNotPrintDirectly = false; 2822 if (S.Context.getTargetInfo().getTriple().isOSDarwin()) { 2823 // Use a 'while' to peel off layers of typedefs. 2824 QualType TyTy = IntendedTy; 2825 while (const TypedefType *UserTy = TyTy->getAs<TypedefType>()) { 2826 StringRef Name = UserTy->getDecl()->getName(); 2827 QualType CastTy = llvm::StringSwitch<QualType>(Name) 2828 .Case("NSInteger", S.Context.LongTy) 2829 .Case("NSUInteger", S.Context.UnsignedLongTy) 2830 .Case("SInt32", S.Context.IntTy) 2831 .Case("UInt32", S.Context.UnsignedIntTy) 2832 .Default(QualType()); 2833 2834 if (!CastTy.isNull()) { 2835 ShouldNotPrintDirectly = true; 2836 IntendedTy = CastTy; 2837 break; 2838 } 2839 TyTy = UserTy->desugar(); 2840 } 2841 } 2842 2843 // We may be able to offer a FixItHint if it is a supported type. 2844 PrintfSpecifier fixedFS = FS; 2845 bool success = fixedFS.fixType(IntendedTy, S.getLangOpts(), 2846 S.Context, ObjCContext); 2847 2848 if (success) { 2849 // Get the fix string from the fixed format specifier 2850 SmallString<16> buf; 2851 llvm::raw_svector_ostream os(buf); 2852 fixedFS.toString(os); 2853 2854 CharSourceRange SpecRange = getSpecifierRange(StartSpecifier, SpecifierLen); 2855 2856 if (IntendedTy == ExprTy) { 2857 // In this case, the specifier is wrong and should be changed to match 2858 // the argument. 2859 EmitFormatDiagnostic( 2860 S.PDiag(diag::warn_printf_conversion_argument_type_mismatch) 2861 << AT.getRepresentativeTypeName(S.Context) << IntendedTy 2862 << E->getSourceRange(), 2863 E->getLocStart(), 2864 /*IsStringLocation*/false, 2865 SpecRange, 2866 FixItHint::CreateReplacement(SpecRange, os.str())); 2867 2868 } else { 2869 // The canonical type for formatting this value is different from the 2870 // actual type of the expression. (This occurs, for example, with Darwin's 2871 // NSInteger on 32-bit platforms, where it is typedef'd as 'int', but 2872 // should be printed as 'long' for 64-bit compatibility.) 2873 // Rather than emitting a normal format/argument mismatch, we want to 2874 // add a cast to the recommended type (and correct the format string 2875 // if necessary). 2876 SmallString<16> CastBuf; 2877 llvm::raw_svector_ostream CastFix(CastBuf); 2878 CastFix << "("; 2879 IntendedTy.print(CastFix, S.Context.getPrintingPolicy()); 2880 CastFix << ")"; 2881 2882 SmallVector<FixItHint,4> Hints; 2883 if (!AT.matchesType(S.Context, IntendedTy)) 2884 Hints.push_back(FixItHint::CreateReplacement(SpecRange, os.str())); 2885 2886 if (const CStyleCastExpr *CCast = dyn_cast<CStyleCastExpr>(E)) { 2887 // If there's already a cast present, just replace it. 2888 SourceRange CastRange(CCast->getLParenLoc(), CCast->getRParenLoc()); 2889 Hints.push_back(FixItHint::CreateReplacement(CastRange, CastFix.str())); 2890 2891 } else if (!requiresParensToAddCast(E)) { 2892 // If the expression has high enough precedence, 2893 // just write the C-style cast. 2894 Hints.push_back(FixItHint::CreateInsertion(E->getLocStart(), 2895 CastFix.str())); 2896 } else { 2897 // Otherwise, add parens around the expression as well as the cast. 2898 CastFix << "("; 2899 Hints.push_back(FixItHint::CreateInsertion(E->getLocStart(), 2900 CastFix.str())); 2901 2902 SourceLocation After = S.PP.getLocForEndOfToken(E->getLocEnd()); 2903 Hints.push_back(FixItHint::CreateInsertion(After, ")")); 2904 } 2905 2906 if (ShouldNotPrintDirectly) { 2907 // The expression has a type that should not be printed directly. 2908 // We extract the name from the typedef because we don't want to show 2909 // the underlying type in the diagnostic. 2910 StringRef Name = cast<TypedefType>(ExprTy)->getDecl()->getName(); 2911 2912 EmitFormatDiagnostic(S.PDiag(diag::warn_format_argument_needs_cast) 2913 << Name << IntendedTy 2914 << E->getSourceRange(), 2915 E->getLocStart(), /*IsStringLocation=*/false, 2916 SpecRange, Hints); 2917 } else { 2918 // In this case, the expression could be printed using a different 2919 // specifier, but we've decided that the specifier is probably correct 2920 // and we should cast instead. Just use the normal warning message. 2921 EmitFormatDiagnostic( 2922 S.PDiag(diag::warn_printf_conversion_argument_type_mismatch) 2923 << AT.getRepresentativeTypeName(S.Context) << ExprTy 2924 << E->getSourceRange(), 2925 E->getLocStart(), /*IsStringLocation*/false, 2926 SpecRange, Hints); 2927 } 2928 } 2929 } else { 2930 const CharSourceRange &CSR = getSpecifierRange(StartSpecifier, 2931 SpecifierLen); 2932 // Since the warning for passing non-POD types to variadic functions 2933 // was deferred until now, we emit a warning for non-POD 2934 // arguments here. 2935 if (S.isValidVarArgType(ExprTy) == Sema::VAK_Invalid) { 2936 unsigned DiagKind; 2937 if (ExprTy->isObjCObjectType()) 2938 DiagKind = diag::err_cannot_pass_objc_interface_to_vararg_format; 2939 else 2940 DiagKind = diag::warn_non_pod_vararg_with_format_string; 2941 2942 EmitFormatDiagnostic( 2943 S.PDiag(DiagKind) 2944 << S.getLangOpts().CPlusPlus11 2945 << ExprTy 2946 << CallType 2947 << AT.getRepresentativeTypeName(S.Context) 2948 << CSR 2949 << E->getSourceRange(), 2950 E->getLocStart(), /*IsStringLocation*/false, CSR); 2951 2952 checkForCStrMembers(AT, E, CSR); 2953 } else 2954 EmitFormatDiagnostic( 2955 S.PDiag(diag::warn_printf_conversion_argument_type_mismatch) 2956 << AT.getRepresentativeTypeName(S.Context) << ExprTy 2957 << CSR 2958 << E->getSourceRange(), 2959 E->getLocStart(), /*IsStringLocation*/false, CSR); 2960 } 2961 2962 return true; 2963} 2964 2965//===--- CHECK: Scanf format string checking ------------------------------===// 2966 2967namespace { 2968class CheckScanfHandler : public CheckFormatHandler { 2969public: 2970 CheckScanfHandler(Sema &s, const StringLiteral *fexpr, 2971 const Expr *origFormatExpr, unsigned firstDataArg, 2972 unsigned numDataArgs, const char *beg, bool hasVAListArg, 2973 ArrayRef<const Expr *> Args, 2974 unsigned formatIdx, bool inFunctionCall, 2975 Sema::VariadicCallType CallType) 2976 : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg, 2977 numDataArgs, beg, hasVAListArg, 2978 Args, formatIdx, inFunctionCall, CallType) 2979 {} 2980 2981 bool HandleScanfSpecifier(const analyze_scanf::ScanfSpecifier &FS, 2982 const char *startSpecifier, 2983 unsigned specifierLen); 2984 2985 bool HandleInvalidScanfConversionSpecifier( 2986 const analyze_scanf::ScanfSpecifier &FS, 2987 const char *startSpecifier, 2988 unsigned specifierLen); 2989 2990 void HandleIncompleteScanList(const char *start, const char *end); 2991}; 2992} 2993 2994void CheckScanfHandler::HandleIncompleteScanList(const char *start, 2995 const char *end) { 2996 EmitFormatDiagnostic(S.PDiag(diag::warn_scanf_scanlist_incomplete), 2997 getLocationOfByte(end), /*IsStringLocation*/true, 2998 getSpecifierRange(start, end - start)); 2999} 3000 3001bool CheckScanfHandler::HandleInvalidScanfConversionSpecifier( 3002 const analyze_scanf::ScanfSpecifier &FS, 3003 const char *startSpecifier, 3004 unsigned specifierLen) { 3005 3006 const analyze_scanf::ScanfConversionSpecifier &CS = 3007 FS.getConversionSpecifier(); 3008 3009 return HandleInvalidConversionSpecifier(FS.getArgIndex(), 3010 getLocationOfByte(CS.getStart()), 3011 startSpecifier, specifierLen, 3012 CS.getStart(), CS.getLength()); 3013} 3014 3015bool CheckScanfHandler::HandleScanfSpecifier( 3016 const analyze_scanf::ScanfSpecifier &FS, 3017 const char *startSpecifier, 3018 unsigned specifierLen) { 3019 3020 using namespace analyze_scanf; 3021 using namespace analyze_format_string; 3022 3023 const ScanfConversionSpecifier &CS = FS.getConversionSpecifier(); 3024 3025 // Handle case where '%' and '*' don't consume an argument. These shouldn't 3026 // be used to decide if we are using positional arguments consistently. 3027 if (FS.consumesDataArgument()) { 3028 if (atFirstArg) { 3029 atFirstArg = false; 3030 usesPositionalArgs = FS.usesPositionalArg(); 3031 } 3032 else if (usesPositionalArgs != FS.usesPositionalArg()) { 3033 HandlePositionalNonpositionalArgs(getLocationOfByte(CS.getStart()), 3034 startSpecifier, specifierLen); 3035 return false; 3036 } 3037 } 3038 3039 // Check if the field with is non-zero. 3040 const OptionalAmount &Amt = FS.getFieldWidth(); 3041 if (Amt.getHowSpecified() == OptionalAmount::Constant) { 3042 if (Amt.getConstantAmount() == 0) { 3043 const CharSourceRange &R = getSpecifierRange(Amt.getStart(), 3044 Amt.getConstantLength()); 3045 EmitFormatDiagnostic(S.PDiag(diag::warn_scanf_nonzero_width), 3046 getLocationOfByte(Amt.getStart()), 3047 /*IsStringLocation*/true, R, 3048 FixItHint::CreateRemoval(R)); 3049 } 3050 } 3051 3052 if (!FS.consumesDataArgument()) { 3053 // FIXME: Technically specifying a precision or field width here 3054 // makes no sense. Worth issuing a warning at some point. 3055 return true; 3056 } 3057 3058 // Consume the argument. 3059 unsigned argIndex = FS.getArgIndex(); 3060 if (argIndex < NumDataArgs) { 3061 // The check to see if the argIndex is valid will come later. 3062 // We set the bit here because we may exit early from this 3063 // function if we encounter some other error. 3064 CoveredArgs.set(argIndex); 3065 } 3066 3067 // Check the length modifier is valid with the given conversion specifier. 3068 if (!FS.hasValidLengthModifier(S.getASTContext().getTargetInfo())) 3069 HandleInvalidLengthModifier(FS, CS, startSpecifier, specifierLen, 3070 diag::warn_format_nonsensical_length); 3071 else if (!FS.hasStandardLengthModifier()) 3072 HandleNonStandardLengthModifier(FS, startSpecifier, specifierLen); 3073 else if (!FS.hasStandardLengthConversionCombination()) 3074 HandleInvalidLengthModifier(FS, CS, startSpecifier, specifierLen, 3075 diag::warn_format_non_standard_conversion_spec); 3076 3077 if (!FS.hasStandardConversionSpecifier(S.getLangOpts())) 3078 HandleNonStandardConversionSpecifier(CS, startSpecifier, specifierLen); 3079 3080 // The remaining checks depend on the data arguments. 3081 if (HasVAListArg) 3082 return true; 3083 3084 if (!CheckNumArgs(FS, CS, startSpecifier, specifierLen, argIndex)) 3085 return false; 3086 3087 // Check that the argument type matches the format specifier. 3088 const Expr *Ex = getDataArg(argIndex); 3089 if (!Ex) 3090 return true; 3091 3092 const analyze_format_string::ArgType &AT = FS.getArgType(S.Context); 3093 if (AT.isValid() && !AT.matchesType(S.Context, Ex->getType())) { 3094 ScanfSpecifier fixedFS = FS; 3095 bool success = fixedFS.fixType(Ex->getType(), S.getLangOpts(), 3096 S.Context); 3097 3098 if (success) { 3099 // Get the fix string from the fixed format specifier. 3100 SmallString<128> buf; 3101 llvm::raw_svector_ostream os(buf); 3102 fixedFS.toString(os); 3103 3104 EmitFormatDiagnostic( 3105 S.PDiag(diag::warn_printf_conversion_argument_type_mismatch) 3106 << AT.getRepresentativeTypeName(S.Context) << Ex->getType() 3107 << Ex->getSourceRange(), 3108 Ex->getLocStart(), 3109 /*IsStringLocation*/false, 3110 getSpecifierRange(startSpecifier, specifierLen), 3111 FixItHint::CreateReplacement( 3112 getSpecifierRange(startSpecifier, specifierLen), 3113 os.str())); 3114 } else { 3115 EmitFormatDiagnostic( 3116 S.PDiag(diag::warn_printf_conversion_argument_type_mismatch) 3117 << AT.getRepresentativeTypeName(S.Context) << Ex->getType() 3118 << Ex->getSourceRange(), 3119 Ex->getLocStart(), 3120 /*IsStringLocation*/false, 3121 getSpecifierRange(startSpecifier, specifierLen)); 3122 } 3123 } 3124 3125 return true; 3126} 3127 3128void Sema::CheckFormatString(const StringLiteral *FExpr, 3129 const Expr *OrigFormatExpr, 3130 ArrayRef<const Expr *> Args, 3131 bool HasVAListArg, unsigned format_idx, 3132 unsigned firstDataArg, FormatStringType Type, 3133 bool inFunctionCall, VariadicCallType CallType) { 3134 3135 // CHECK: is the format string a wide literal? 3136 if (!FExpr->isAscii() && !FExpr->isUTF8()) { 3137 CheckFormatHandler::EmitFormatDiagnostic( 3138 *this, inFunctionCall, Args[format_idx], 3139 PDiag(diag::warn_format_string_is_wide_literal), FExpr->getLocStart(), 3140 /*IsStringLocation*/true, OrigFormatExpr->getSourceRange()); 3141 return; 3142 } 3143 3144 // Str - The format string. NOTE: this is NOT null-terminated! 3145 StringRef StrRef = FExpr->getString(); 3146 const char *Str = StrRef.data(); 3147 unsigned StrLen = StrRef.size(); 3148 const unsigned numDataArgs = Args.size() - firstDataArg; 3149 3150 // CHECK: empty format string? 3151 if (StrLen == 0 && numDataArgs > 0) { 3152 CheckFormatHandler::EmitFormatDiagnostic( 3153 *this, inFunctionCall, Args[format_idx], 3154 PDiag(diag::warn_empty_format_string), FExpr->getLocStart(), 3155 /*IsStringLocation*/true, OrigFormatExpr->getSourceRange()); 3156 return; 3157 } 3158 3159 if (Type == FST_Printf || Type == FST_NSString) { 3160 CheckPrintfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg, 3161 numDataArgs, (Type == FST_NSString), 3162 Str, HasVAListArg, Args, format_idx, 3163 inFunctionCall, CallType); 3164 3165 if (!analyze_format_string::ParsePrintfString(H, Str, Str + StrLen, 3166 getLangOpts(), 3167 Context.getTargetInfo())) 3168 H.DoneProcessing(); 3169 } else if (Type == FST_Scanf) { 3170 CheckScanfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg, numDataArgs, 3171 Str, HasVAListArg, Args, format_idx, 3172 inFunctionCall, CallType); 3173 3174 if (!analyze_format_string::ParseScanfString(H, Str, Str + StrLen, 3175 getLangOpts(), 3176 Context.getTargetInfo())) 3177 H.DoneProcessing(); 3178 } // TODO: handle other formats 3179} 3180 3181//===--- CHECK: Standard memory functions ---------------------------------===// 3182 3183/// \brief Determine whether the given type is a dynamic class type (e.g., 3184/// whether it has a vtable). 3185static bool isDynamicClassType(QualType T) { 3186 if (CXXRecordDecl *Record = T->getAsCXXRecordDecl()) 3187 if (CXXRecordDecl *Definition = Record->getDefinition()) 3188 if (Definition->isDynamicClass()) 3189 return true; 3190 3191 return false; 3192} 3193 3194/// \brief If E is a sizeof expression, returns its argument expression, 3195/// otherwise returns NULL. 3196static const Expr *getSizeOfExprArg(const Expr* E) { 3197 if (const UnaryExprOrTypeTraitExpr *SizeOf = 3198 dyn_cast<UnaryExprOrTypeTraitExpr>(E)) 3199 if (SizeOf->getKind() == clang::UETT_SizeOf && !SizeOf->isArgumentType()) 3200 return SizeOf->getArgumentExpr()->IgnoreParenImpCasts(); 3201 3202 return 0; 3203} 3204 3205/// \brief If E is a sizeof expression, returns its argument type. 3206static QualType getSizeOfArgType(const Expr* E) { 3207 if (const UnaryExprOrTypeTraitExpr *SizeOf = 3208 dyn_cast<UnaryExprOrTypeTraitExpr>(E)) 3209 if (SizeOf->getKind() == clang::UETT_SizeOf) 3210 return SizeOf->getTypeOfArgument(); 3211 3212 return QualType(); 3213} 3214 3215/// \brief Check for dangerous or invalid arguments to memset(). 3216/// 3217/// This issues warnings on known problematic, dangerous or unspecified 3218/// arguments to the standard 'memset', 'memcpy', 'memmove', and 'memcmp' 3219/// function calls. 3220/// 3221/// \param Call The call expression to diagnose. 3222void Sema::CheckMemaccessArguments(const CallExpr *Call, 3223 unsigned BId, 3224 IdentifierInfo *FnName) { 3225 assert(BId != 0); 3226 3227 // It is possible to have a non-standard definition of memset. Validate 3228 // we have enough arguments, and if not, abort further checking. 3229 unsigned ExpectedNumArgs = (BId == Builtin::BIstrndup ? 2 : 3); 3230 if (Call->getNumArgs() < ExpectedNumArgs) 3231 return; 3232 3233 unsigned LastArg = (BId == Builtin::BImemset || 3234 BId == Builtin::BIstrndup ? 1 : 2); 3235 unsigned LenArg = (BId == Builtin::BIstrndup ? 1 : 2); 3236 const Expr *LenExpr = Call->getArg(LenArg)->IgnoreParenImpCasts(); 3237 3238 // We have special checking when the length is a sizeof expression. 3239 QualType SizeOfArgTy = getSizeOfArgType(LenExpr); 3240 const Expr *SizeOfArg = getSizeOfExprArg(LenExpr); 3241 llvm::FoldingSetNodeID SizeOfArgID; 3242 3243 for (unsigned ArgIdx = 0; ArgIdx != LastArg; ++ArgIdx) { 3244 const Expr *Dest = Call->getArg(ArgIdx)->IgnoreParenImpCasts(); 3245 SourceRange ArgRange = Call->getArg(ArgIdx)->getSourceRange(); 3246 3247 QualType DestTy = Dest->getType(); 3248 if (const PointerType *DestPtrTy = DestTy->getAs<PointerType>()) { 3249 QualType PointeeTy = DestPtrTy->getPointeeType(); 3250 3251 // Never warn about void type pointers. This can be used to suppress 3252 // false positives. 3253 if (PointeeTy->isVoidType()) 3254 continue; 3255 3256 // Catch "memset(p, 0, sizeof(p))" -- needs to be sizeof(*p). Do this by 3257 // actually comparing the expressions for equality. Because computing the 3258 // expression IDs can be expensive, we only do this if the diagnostic is 3259 // enabled. 3260 if (SizeOfArg && 3261 Diags.getDiagnosticLevel(diag::warn_sizeof_pointer_expr_memaccess, 3262 SizeOfArg->getExprLoc())) { 3263 // We only compute IDs for expressions if the warning is enabled, and 3264 // cache the sizeof arg's ID. 3265 if (SizeOfArgID == llvm::FoldingSetNodeID()) 3266 SizeOfArg->Profile(SizeOfArgID, Context, true); 3267 llvm::FoldingSetNodeID DestID; 3268 Dest->Profile(DestID, Context, true); 3269 if (DestID == SizeOfArgID) { 3270 // TODO: For strncpy() and friends, this could suggest sizeof(dst) 3271 // over sizeof(src) as well. 3272 unsigned ActionIdx = 0; // Default is to suggest dereferencing. 3273 StringRef ReadableName = FnName->getName(); 3274 3275 if (const UnaryOperator *UnaryOp = dyn_cast<UnaryOperator>(Dest)) 3276 if (UnaryOp->getOpcode() == UO_AddrOf) 3277 ActionIdx = 1; // If its an address-of operator, just remove it. 3278 if (!PointeeTy->isIncompleteType() && 3279 (Context.getTypeSize(PointeeTy) == Context.getCharWidth())) 3280 ActionIdx = 2; // If the pointee's size is sizeof(char), 3281 // suggest an explicit length. 3282 3283 // If the function is defined as a builtin macro, do not show macro 3284 // expansion. 3285 SourceLocation SL = SizeOfArg->getExprLoc(); 3286 SourceRange DSR = Dest->getSourceRange(); 3287 SourceRange SSR = SizeOfArg->getSourceRange(); 3288 SourceManager &SM = PP.getSourceManager(); 3289 3290 if (SM.isMacroArgExpansion(SL)) { 3291 ReadableName = Lexer::getImmediateMacroName(SL, SM, LangOpts); 3292 SL = SM.getSpellingLoc(SL); 3293 DSR = SourceRange(SM.getSpellingLoc(DSR.getBegin()), 3294 SM.getSpellingLoc(DSR.getEnd())); 3295 SSR = SourceRange(SM.getSpellingLoc(SSR.getBegin()), 3296 SM.getSpellingLoc(SSR.getEnd())); 3297 } 3298 3299 DiagRuntimeBehavior(SL, SizeOfArg, 3300 PDiag(diag::warn_sizeof_pointer_expr_memaccess) 3301 << ReadableName 3302 << PointeeTy 3303 << DestTy 3304 << DSR 3305 << SSR); 3306 DiagRuntimeBehavior(SL, SizeOfArg, 3307 PDiag(diag::warn_sizeof_pointer_expr_memaccess_note) 3308 << ActionIdx 3309 << SSR); 3310 3311 break; 3312 } 3313 } 3314 3315 // Also check for cases where the sizeof argument is the exact same 3316 // type as the memory argument, and where it points to a user-defined 3317 // record type. 3318 if (SizeOfArgTy != QualType()) { 3319 if (PointeeTy->isRecordType() && 3320 Context.typesAreCompatible(SizeOfArgTy, DestTy)) { 3321 DiagRuntimeBehavior(LenExpr->getExprLoc(), Dest, 3322 PDiag(diag::warn_sizeof_pointer_type_memaccess) 3323 << FnName << SizeOfArgTy << ArgIdx 3324 << PointeeTy << Dest->getSourceRange() 3325 << LenExpr->getSourceRange()); 3326 break; 3327 } 3328 } 3329 3330 // Always complain about dynamic classes. 3331 if (isDynamicClassType(PointeeTy)) { 3332 3333 unsigned OperationType = 0; 3334 // "overwritten" if we're warning about the destination for any call 3335 // but memcmp; otherwise a verb appropriate to the call. 3336 if (ArgIdx != 0 || BId == Builtin::BImemcmp) { 3337 if (BId == Builtin::BImemcpy) 3338 OperationType = 1; 3339 else if(BId == Builtin::BImemmove) 3340 OperationType = 2; 3341 else if (BId == Builtin::BImemcmp) 3342 OperationType = 3; 3343 } 3344 3345 DiagRuntimeBehavior( 3346 Dest->getExprLoc(), Dest, 3347 PDiag(diag::warn_dyn_class_memaccess) 3348 << (BId == Builtin::BImemcmp ? ArgIdx + 2 : ArgIdx) 3349 << FnName << PointeeTy 3350 << OperationType 3351 << Call->getCallee()->getSourceRange()); 3352 } else if (PointeeTy.hasNonTrivialObjCLifetime() && 3353 BId != Builtin::BImemset) 3354 DiagRuntimeBehavior( 3355 Dest->getExprLoc(), Dest, 3356 PDiag(diag::warn_arc_object_memaccess) 3357 << ArgIdx << FnName << PointeeTy 3358 << Call->getCallee()->getSourceRange()); 3359 else 3360 continue; 3361 3362 DiagRuntimeBehavior( 3363 Dest->getExprLoc(), Dest, 3364 PDiag(diag::note_bad_memaccess_silence) 3365 << FixItHint::CreateInsertion(ArgRange.getBegin(), "(void*)")); 3366 break; 3367 } 3368 } 3369} 3370 3371// A little helper routine: ignore addition and subtraction of integer literals. 3372// This intentionally does not ignore all integer constant expressions because 3373// we don't want to remove sizeof(). 3374static const Expr *ignoreLiteralAdditions(const Expr *Ex, ASTContext &Ctx) { 3375 Ex = Ex->IgnoreParenCasts(); 3376 3377 for (;;) { 3378 const BinaryOperator * BO = dyn_cast<BinaryOperator>(Ex); 3379 if (!BO || !BO->isAdditiveOp()) 3380 break; 3381 3382 const Expr *RHS = BO->getRHS()->IgnoreParenCasts(); 3383 const Expr *LHS = BO->getLHS()->IgnoreParenCasts(); 3384 3385 if (isa<IntegerLiteral>(RHS)) 3386 Ex = LHS; 3387 else if (isa<IntegerLiteral>(LHS)) 3388 Ex = RHS; 3389 else 3390 break; 3391 } 3392 3393 return Ex; 3394} 3395 3396static bool isConstantSizeArrayWithMoreThanOneElement(QualType Ty, 3397 ASTContext &Context) { 3398 // Only handle constant-sized or VLAs, but not flexible members. 3399 if (const ConstantArrayType *CAT = Context.getAsConstantArrayType(Ty)) { 3400 // Only issue the FIXIT for arrays of size > 1. 3401 if (CAT->getSize().getSExtValue() <= 1) 3402 return false; 3403 } else if (!Ty->isVariableArrayType()) { 3404 return false; 3405 } 3406 return true; 3407} 3408 3409// Warn if the user has made the 'size' argument to strlcpy or strlcat 3410// be the size of the source, instead of the destination. 3411void Sema::CheckStrlcpycatArguments(const CallExpr *Call, 3412 IdentifierInfo *FnName) { 3413 3414 // Don't crash if the user has the wrong number of arguments 3415 if (Call->getNumArgs() != 3) 3416 return; 3417 3418 const Expr *SrcArg = ignoreLiteralAdditions(Call->getArg(1), Context); 3419 const Expr *SizeArg = ignoreLiteralAdditions(Call->getArg(2), Context); 3420 const Expr *CompareWithSrc = NULL; 3421 3422 // Look for 'strlcpy(dst, x, sizeof(x))' 3423 if (const Expr *Ex = getSizeOfExprArg(SizeArg)) 3424 CompareWithSrc = Ex; 3425 else { 3426 // Look for 'strlcpy(dst, x, strlen(x))' 3427 if (const CallExpr *SizeCall = dyn_cast<CallExpr>(SizeArg)) { 3428 if (SizeCall->isBuiltinCall() == Builtin::BIstrlen 3429 && SizeCall->getNumArgs() == 1) 3430 CompareWithSrc = ignoreLiteralAdditions(SizeCall->getArg(0), Context); 3431 } 3432 } 3433 3434 if (!CompareWithSrc) 3435 return; 3436 3437 // Determine if the argument to sizeof/strlen is equal to the source 3438 // argument. In principle there's all kinds of things you could do 3439 // here, for instance creating an == expression and evaluating it with 3440 // EvaluateAsBooleanCondition, but this uses a more direct technique: 3441 const DeclRefExpr *SrcArgDRE = dyn_cast<DeclRefExpr>(SrcArg); 3442 if (!SrcArgDRE) 3443 return; 3444 3445 const DeclRefExpr *CompareWithSrcDRE = dyn_cast<DeclRefExpr>(CompareWithSrc); 3446 if (!CompareWithSrcDRE || 3447 SrcArgDRE->getDecl() != CompareWithSrcDRE->getDecl()) 3448 return; 3449 3450 const Expr *OriginalSizeArg = Call->getArg(2); 3451 Diag(CompareWithSrcDRE->getLocStart(), diag::warn_strlcpycat_wrong_size) 3452 << OriginalSizeArg->getSourceRange() << FnName; 3453 3454 // Output a FIXIT hint if the destination is an array (rather than a 3455 // pointer to an array). This could be enhanced to handle some 3456 // pointers if we know the actual size, like if DstArg is 'array+2' 3457 // we could say 'sizeof(array)-2'. 3458 const Expr *DstArg = Call->getArg(0)->IgnoreParenImpCasts(); 3459 if (!isConstantSizeArrayWithMoreThanOneElement(DstArg->getType(), Context)) 3460 return; 3461 3462 SmallString<128> sizeString; 3463 llvm::raw_svector_ostream OS(sizeString); 3464 OS << "sizeof("; 3465 DstArg->printPretty(OS, 0, getPrintingPolicy()); 3466 OS << ")"; 3467 3468 Diag(OriginalSizeArg->getLocStart(), diag::note_strlcpycat_wrong_size) 3469 << FixItHint::CreateReplacement(OriginalSizeArg->getSourceRange(), 3470 OS.str()); 3471} 3472 3473/// Check if two expressions refer to the same declaration. 3474static bool referToTheSameDecl(const Expr *E1, const Expr *E2) { 3475 if (const DeclRefExpr *D1 = dyn_cast_or_null<DeclRefExpr>(E1)) 3476 if (const DeclRefExpr *D2 = dyn_cast_or_null<DeclRefExpr>(E2)) 3477 return D1->getDecl() == D2->getDecl(); 3478 return false; 3479} 3480 3481static const Expr *getStrlenExprArg(const Expr *E) { 3482 if (const CallExpr *CE = dyn_cast<CallExpr>(E)) { 3483 const FunctionDecl *FD = CE->getDirectCallee(); 3484 if (!FD || FD->getMemoryFunctionKind() != Builtin::BIstrlen) 3485 return 0; 3486 return CE->getArg(0)->IgnoreParenCasts(); 3487 } 3488 return 0; 3489} 3490 3491// Warn on anti-patterns as the 'size' argument to strncat. 3492// The correct size argument should look like following: 3493// strncat(dst, src, sizeof(dst) - strlen(dest) - 1); 3494void Sema::CheckStrncatArguments(const CallExpr *CE, 3495 IdentifierInfo *FnName) { 3496 // Don't crash if the user has the wrong number of arguments. 3497 if (CE->getNumArgs() < 3) 3498 return; 3499 const Expr *DstArg = CE->getArg(0)->IgnoreParenCasts(); 3500 const Expr *SrcArg = CE->getArg(1)->IgnoreParenCasts(); 3501 const Expr *LenArg = CE->getArg(2)->IgnoreParenCasts(); 3502 3503 // Identify common expressions, which are wrongly used as the size argument 3504 // to strncat and may lead to buffer overflows. 3505 unsigned PatternType = 0; 3506 if (const Expr *SizeOfArg = getSizeOfExprArg(LenArg)) { 3507 // - sizeof(dst) 3508 if (referToTheSameDecl(SizeOfArg, DstArg)) 3509 PatternType = 1; 3510 // - sizeof(src) 3511 else if (referToTheSameDecl(SizeOfArg, SrcArg)) 3512 PatternType = 2; 3513 } else if (const BinaryOperator *BE = dyn_cast<BinaryOperator>(LenArg)) { 3514 if (BE->getOpcode() == BO_Sub) { 3515 const Expr *L = BE->getLHS()->IgnoreParenCasts(); 3516 const Expr *R = BE->getRHS()->IgnoreParenCasts(); 3517 // - sizeof(dst) - strlen(dst) 3518 if (referToTheSameDecl(DstArg, getSizeOfExprArg(L)) && 3519 referToTheSameDecl(DstArg, getStrlenExprArg(R))) 3520 PatternType = 1; 3521 // - sizeof(src) - (anything) 3522 else if (referToTheSameDecl(SrcArg, getSizeOfExprArg(L))) 3523 PatternType = 2; 3524 } 3525 } 3526 3527 if (PatternType == 0) 3528 return; 3529 3530 // Generate the diagnostic. 3531 SourceLocation SL = LenArg->getLocStart(); 3532 SourceRange SR = LenArg->getSourceRange(); 3533 SourceManager &SM = PP.getSourceManager(); 3534 3535 // If the function is defined as a builtin macro, do not show macro expansion. 3536 if (SM.isMacroArgExpansion(SL)) { 3537 SL = SM.getSpellingLoc(SL); 3538 SR = SourceRange(SM.getSpellingLoc(SR.getBegin()), 3539 SM.getSpellingLoc(SR.getEnd())); 3540 } 3541 3542 // Check if the destination is an array (rather than a pointer to an array). 3543 QualType DstTy = DstArg->getType(); 3544 bool isKnownSizeArray = isConstantSizeArrayWithMoreThanOneElement(DstTy, 3545 Context); 3546 if (!isKnownSizeArray) { 3547 if (PatternType == 1) 3548 Diag(SL, diag::warn_strncat_wrong_size) << SR; 3549 else 3550 Diag(SL, diag::warn_strncat_src_size) << SR; 3551 return; 3552 } 3553 3554 if (PatternType == 1) 3555 Diag(SL, diag::warn_strncat_large_size) << SR; 3556 else 3557 Diag(SL, diag::warn_strncat_src_size) << SR; 3558 3559 SmallString<128> sizeString; 3560 llvm::raw_svector_ostream OS(sizeString); 3561 OS << "sizeof("; 3562 DstArg->printPretty(OS, 0, getPrintingPolicy()); 3563 OS << ") - "; 3564 OS << "strlen("; 3565 DstArg->printPretty(OS, 0, getPrintingPolicy()); 3566 OS << ") - 1"; 3567 3568 Diag(SL, diag::note_strncat_wrong_size) 3569 << FixItHint::CreateReplacement(SR, OS.str()); 3570} 3571 3572//===--- CHECK: Return Address of Stack Variable --------------------------===// 3573 3574static Expr *EvalVal(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars, 3575 Decl *ParentDecl); 3576static Expr *EvalAddr(Expr* E, SmallVectorImpl<DeclRefExpr *> &refVars, 3577 Decl *ParentDecl); 3578 3579/// CheckReturnStackAddr - Check if a return statement returns the address 3580/// of a stack variable. 3581void 3582Sema::CheckReturnStackAddr(Expr *RetValExp, QualType lhsType, 3583 SourceLocation ReturnLoc) { 3584 3585 Expr *stackE = 0; 3586 SmallVector<DeclRefExpr *, 8> refVars; 3587 3588 // Perform checking for returned stack addresses, local blocks, 3589 // label addresses or references to temporaries. 3590 if (lhsType->isPointerType() || 3591 (!getLangOpts().ObjCAutoRefCount && lhsType->isBlockPointerType())) { 3592 stackE = EvalAddr(RetValExp, refVars, /*ParentDecl=*/0); 3593 } else if (lhsType->isReferenceType()) { 3594 stackE = EvalVal(RetValExp, refVars, /*ParentDecl=*/0); 3595 } 3596 3597 if (stackE == 0) 3598 return; // Nothing suspicious was found. 3599 3600 SourceLocation diagLoc; 3601 SourceRange diagRange; 3602 if (refVars.empty()) { 3603 diagLoc = stackE->getLocStart(); 3604 diagRange = stackE->getSourceRange(); 3605 } else { 3606 // We followed through a reference variable. 'stackE' contains the 3607 // problematic expression but we will warn at the return statement pointing 3608 // at the reference variable. We will later display the "trail" of 3609 // reference variables using notes. 3610 diagLoc = refVars[0]->getLocStart(); 3611 diagRange = refVars[0]->getSourceRange(); 3612 } 3613 3614 if (DeclRefExpr *DR = dyn_cast<DeclRefExpr>(stackE)) { //address of local var. 3615 Diag(diagLoc, lhsType->isReferenceType() ? diag::warn_ret_stack_ref 3616 : diag::warn_ret_stack_addr) 3617 << DR->getDecl()->getDeclName() << diagRange; 3618 } else if (isa<BlockExpr>(stackE)) { // local block. 3619 Diag(diagLoc, diag::err_ret_local_block) << diagRange; 3620 } else if (isa<AddrLabelExpr>(stackE)) { // address of label. 3621 Diag(diagLoc, diag::warn_ret_addr_label) << diagRange; 3622 } else { // local temporary. 3623 Diag(diagLoc, lhsType->isReferenceType() ? diag::warn_ret_local_temp_ref 3624 : diag::warn_ret_local_temp_addr) 3625 << diagRange; 3626 } 3627 3628 // Display the "trail" of reference variables that we followed until we 3629 // found the problematic expression using notes. 3630 for (unsigned i = 0, e = refVars.size(); i != e; ++i) { 3631 VarDecl *VD = cast<VarDecl>(refVars[i]->getDecl()); 3632 // If this var binds to another reference var, show the range of the next 3633 // var, otherwise the var binds to the problematic expression, in which case 3634 // show the range of the expression. 3635 SourceRange range = (i < e-1) ? refVars[i+1]->getSourceRange() 3636 : stackE->getSourceRange(); 3637 Diag(VD->getLocation(), diag::note_ref_var_local_bind) 3638 << VD->getDeclName() << range; 3639 } 3640} 3641 3642/// EvalAddr - EvalAddr and EvalVal are mutually recursive functions that 3643/// check if the expression in a return statement evaluates to an address 3644/// to a location on the stack, a local block, an address of a label, or a 3645/// reference to local temporary. The recursion is used to traverse the 3646/// AST of the return expression, with recursion backtracking when we 3647/// encounter a subexpression that (1) clearly does not lead to one of the 3648/// above problematic expressions (2) is something we cannot determine leads to 3649/// a problematic expression based on such local checking. 3650/// 3651/// Both EvalAddr and EvalVal follow through reference variables to evaluate 3652/// the expression that they point to. Such variables are added to the 3653/// 'refVars' vector so that we know what the reference variable "trail" was. 3654/// 3655/// EvalAddr processes expressions that are pointers that are used as 3656/// references (and not L-values). EvalVal handles all other values. 3657/// At the base case of the recursion is a check for the above problematic 3658/// expressions. 3659/// 3660/// This implementation handles: 3661/// 3662/// * pointer-to-pointer casts 3663/// * implicit conversions from array references to pointers 3664/// * taking the address of fields 3665/// * arbitrary interplay between "&" and "*" operators 3666/// * pointer arithmetic from an address of a stack variable 3667/// * taking the address of an array element where the array is on the stack 3668static Expr *EvalAddr(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars, 3669 Decl *ParentDecl) { 3670 if (E->isTypeDependent()) 3671 return NULL; 3672 3673 // We should only be called for evaluating pointer expressions. 3674 assert((E->getType()->isAnyPointerType() || 3675 E->getType()->isBlockPointerType() || 3676 E->getType()->isObjCQualifiedIdType()) && 3677 "EvalAddr only works on pointers"); 3678 3679 E = E->IgnoreParens(); 3680 3681 // Our "symbolic interpreter" is just a dispatch off the currently 3682 // viewed AST node. We then recursively traverse the AST by calling 3683 // EvalAddr and EvalVal appropriately. 3684 switch (E->getStmtClass()) { 3685 case Stmt::DeclRefExprClass: { 3686 DeclRefExpr *DR = cast<DeclRefExpr>(E); 3687 3688 if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl())) 3689 // If this is a reference variable, follow through to the expression that 3690 // it points to. 3691 if (V->hasLocalStorage() && 3692 V->getType()->isReferenceType() && V->hasInit()) { 3693 // Add the reference variable to the "trail". 3694 refVars.push_back(DR); 3695 return EvalAddr(V->getInit(), refVars, ParentDecl); 3696 } 3697 3698 return NULL; 3699 } 3700 3701 case Stmt::UnaryOperatorClass: { 3702 // The only unary operator that make sense to handle here 3703 // is AddrOf. All others don't make sense as pointers. 3704 UnaryOperator *U = cast<UnaryOperator>(E); 3705 3706 if (U->getOpcode() == UO_AddrOf) 3707 return EvalVal(U->getSubExpr(), refVars, ParentDecl); 3708 else 3709 return NULL; 3710 } 3711 3712 case Stmt::BinaryOperatorClass: { 3713 // Handle pointer arithmetic. All other binary operators are not valid 3714 // in this context. 3715 BinaryOperator *B = cast<BinaryOperator>(E); 3716 BinaryOperatorKind op = B->getOpcode(); 3717 3718 if (op != BO_Add && op != BO_Sub) 3719 return NULL; 3720 3721 Expr *Base = B->getLHS(); 3722 3723 // Determine which argument is the real pointer base. It could be 3724 // the RHS argument instead of the LHS. 3725 if (!Base->getType()->isPointerType()) Base = B->getRHS(); 3726 3727 assert (Base->getType()->isPointerType()); 3728 return EvalAddr(Base, refVars, ParentDecl); 3729 } 3730 3731 // For conditional operators we need to see if either the LHS or RHS are 3732 // valid DeclRefExpr*s. If one of them is valid, we return it. 3733 case Stmt::ConditionalOperatorClass: { 3734 ConditionalOperator *C = cast<ConditionalOperator>(E); 3735 3736 // Handle the GNU extension for missing LHS. 3737 if (Expr *lhsExpr = C->getLHS()) { 3738 // In C++, we can have a throw-expression, which has 'void' type. 3739 if (!lhsExpr->getType()->isVoidType()) 3740 if (Expr* LHS = EvalAddr(lhsExpr, refVars, ParentDecl)) 3741 return LHS; 3742 } 3743 3744 // In C++, we can have a throw-expression, which has 'void' type. 3745 if (C->getRHS()->getType()->isVoidType()) 3746 return NULL; 3747 3748 return EvalAddr(C->getRHS(), refVars, ParentDecl); 3749 } 3750 3751 case Stmt::BlockExprClass: 3752 if (cast<BlockExpr>(E)->getBlockDecl()->hasCaptures()) 3753 return E; // local block. 3754 return NULL; 3755 3756 case Stmt::AddrLabelExprClass: 3757 return E; // address of label. 3758 3759 case Stmt::ExprWithCleanupsClass: 3760 return EvalAddr(cast<ExprWithCleanups>(E)->getSubExpr(), refVars, 3761 ParentDecl); 3762 3763 // For casts, we need to handle conversions from arrays to 3764 // pointer values, and pointer-to-pointer conversions. 3765 case Stmt::ImplicitCastExprClass: 3766 case Stmt::CStyleCastExprClass: 3767 case Stmt::CXXFunctionalCastExprClass: 3768 case Stmt::ObjCBridgedCastExprClass: 3769 case Stmt::CXXStaticCastExprClass: 3770 case Stmt::CXXDynamicCastExprClass: 3771 case Stmt::CXXConstCastExprClass: 3772 case Stmt::CXXReinterpretCastExprClass: { 3773 Expr* SubExpr = cast<CastExpr>(E)->getSubExpr(); 3774 switch (cast<CastExpr>(E)->getCastKind()) { 3775 case CK_BitCast: 3776 case CK_LValueToRValue: 3777 case CK_NoOp: 3778 case CK_BaseToDerived: 3779 case CK_DerivedToBase: 3780 case CK_UncheckedDerivedToBase: 3781 case CK_Dynamic: 3782 case CK_CPointerToObjCPointerCast: 3783 case CK_BlockPointerToObjCPointerCast: 3784 case CK_AnyPointerToBlockPointerCast: 3785 return EvalAddr(SubExpr, refVars, ParentDecl); 3786 3787 case CK_ArrayToPointerDecay: 3788 return EvalVal(SubExpr, refVars, ParentDecl); 3789 3790 default: 3791 return 0; 3792 } 3793 } 3794 3795 case Stmt::MaterializeTemporaryExprClass: 3796 if (Expr *Result = EvalAddr( 3797 cast<MaterializeTemporaryExpr>(E)->GetTemporaryExpr(), 3798 refVars, ParentDecl)) 3799 return Result; 3800 3801 return E; 3802 3803 // Everything else: we simply don't reason about them. 3804 default: 3805 return NULL; 3806 } 3807} 3808 3809 3810/// EvalVal - This function is complements EvalAddr in the mutual recursion. 3811/// See the comments for EvalAddr for more details. 3812static Expr *EvalVal(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars, 3813 Decl *ParentDecl) { 3814do { 3815 // We should only be called for evaluating non-pointer expressions, or 3816 // expressions with a pointer type that are not used as references but instead 3817 // are l-values (e.g., DeclRefExpr with a pointer type). 3818 3819 // Our "symbolic interpreter" is just a dispatch off the currently 3820 // viewed AST node. We then recursively traverse the AST by calling 3821 // EvalAddr and EvalVal appropriately. 3822 3823 E = E->IgnoreParens(); 3824 switch (E->getStmtClass()) { 3825 case Stmt::ImplicitCastExprClass: { 3826 ImplicitCastExpr *IE = cast<ImplicitCastExpr>(E); 3827 if (IE->getValueKind() == VK_LValue) { 3828 E = IE->getSubExpr(); 3829 continue; 3830 } 3831 return NULL; 3832 } 3833 3834 case Stmt::ExprWithCleanupsClass: 3835 return EvalVal(cast<ExprWithCleanups>(E)->getSubExpr(), refVars,ParentDecl); 3836 3837 case Stmt::DeclRefExprClass: { 3838 // When we hit a DeclRefExpr we are looking at code that refers to a 3839 // variable's name. If it's not a reference variable we check if it has 3840 // local storage within the function, and if so, return the expression. 3841 DeclRefExpr *DR = cast<DeclRefExpr>(E); 3842 3843 if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl())) { 3844 // Check if it refers to itself, e.g. "int& i = i;". 3845 if (V == ParentDecl) 3846 return DR; 3847 3848 if (V->hasLocalStorage()) { 3849 if (!V->getType()->isReferenceType()) 3850 return DR; 3851 3852 // Reference variable, follow through to the expression that 3853 // it points to. 3854 if (V->hasInit()) { 3855 // Add the reference variable to the "trail". 3856 refVars.push_back(DR); 3857 return EvalVal(V->getInit(), refVars, V); 3858 } 3859 } 3860 } 3861 3862 return NULL; 3863 } 3864 3865 case Stmt::UnaryOperatorClass: { 3866 // The only unary operator that make sense to handle here 3867 // is Deref. All others don't resolve to a "name." This includes 3868 // handling all sorts of rvalues passed to a unary operator. 3869 UnaryOperator *U = cast<UnaryOperator>(E); 3870 3871 if (U->getOpcode() == UO_Deref) 3872 return EvalAddr(U->getSubExpr(), refVars, ParentDecl); 3873 3874 return NULL; 3875 } 3876 3877 case Stmt::ArraySubscriptExprClass: { 3878 // Array subscripts are potential references to data on the stack. We 3879 // retrieve the DeclRefExpr* for the array variable if it indeed 3880 // has local storage. 3881 return EvalAddr(cast<ArraySubscriptExpr>(E)->getBase(), refVars,ParentDecl); 3882 } 3883 3884 case Stmt::ConditionalOperatorClass: { 3885 // For conditional operators we need to see if either the LHS or RHS are 3886 // non-NULL Expr's. If one is non-NULL, we return it. 3887 ConditionalOperator *C = cast<ConditionalOperator>(E); 3888 3889 // Handle the GNU extension for missing LHS. 3890 if (Expr *lhsExpr = C->getLHS()) 3891 if (Expr *LHS = EvalVal(lhsExpr, refVars, ParentDecl)) 3892 return LHS; 3893 3894 return EvalVal(C->getRHS(), refVars, ParentDecl); 3895 } 3896 3897 // Accesses to members are potential references to data on the stack. 3898 case Stmt::MemberExprClass: { 3899 MemberExpr *M = cast<MemberExpr>(E); 3900 3901 // Check for indirect access. We only want direct field accesses. 3902 if (M->isArrow()) 3903 return NULL; 3904 3905 // Check whether the member type is itself a reference, in which case 3906 // we're not going to refer to the member, but to what the member refers to. 3907 if (M->getMemberDecl()->getType()->isReferenceType()) 3908 return NULL; 3909 3910 return EvalVal(M->getBase(), refVars, ParentDecl); 3911 } 3912 3913 case Stmt::MaterializeTemporaryExprClass: 3914 if (Expr *Result = EvalVal( 3915 cast<MaterializeTemporaryExpr>(E)->GetTemporaryExpr(), 3916 refVars, ParentDecl)) 3917 return Result; 3918 3919 return E; 3920 3921 default: 3922 // Check that we don't return or take the address of a reference to a 3923 // temporary. This is only useful in C++. 3924 if (!E->isTypeDependent() && E->isRValue()) 3925 return E; 3926 3927 // Everything else: we simply don't reason about them. 3928 return NULL; 3929 } 3930} while (true); 3931} 3932 3933//===--- CHECK: Floating-Point comparisons (-Wfloat-equal) ---------------===// 3934 3935/// Check for comparisons of floating point operands using != and ==. 3936/// Issue a warning if these are no self-comparisons, as they are not likely 3937/// to do what the programmer intended. 3938void Sema::CheckFloatComparison(SourceLocation Loc, Expr* LHS, Expr *RHS) { 3939 Expr* LeftExprSansParen = LHS->IgnoreParenImpCasts(); 3940 Expr* RightExprSansParen = RHS->IgnoreParenImpCasts(); 3941 3942 // Special case: check for x == x (which is OK). 3943 // Do not emit warnings for such cases. 3944 if (DeclRefExpr* DRL = dyn_cast<DeclRefExpr>(LeftExprSansParen)) 3945 if (DeclRefExpr* DRR = dyn_cast<DeclRefExpr>(RightExprSansParen)) 3946 if (DRL->getDecl() == DRR->getDecl()) 3947 return; 3948 3949 3950 // Special case: check for comparisons against literals that can be exactly 3951 // represented by APFloat. In such cases, do not emit a warning. This 3952 // is a heuristic: often comparison against such literals are used to 3953 // detect if a value in a variable has not changed. This clearly can 3954 // lead to false negatives. 3955 if (FloatingLiteral* FLL = dyn_cast<FloatingLiteral>(LeftExprSansParen)) { 3956 if (FLL->isExact()) 3957 return; 3958 } else 3959 if (FloatingLiteral* FLR = dyn_cast<FloatingLiteral>(RightExprSansParen)) 3960 if (FLR->isExact()) 3961 return; 3962 3963 // Check for comparisons with builtin types. 3964 if (CallExpr* CL = dyn_cast<CallExpr>(LeftExprSansParen)) 3965 if (CL->isBuiltinCall()) 3966 return; 3967 3968 if (CallExpr* CR = dyn_cast<CallExpr>(RightExprSansParen)) 3969 if (CR->isBuiltinCall()) 3970 return; 3971 3972 // Emit the diagnostic. 3973 Diag(Loc, diag::warn_floatingpoint_eq) 3974 << LHS->getSourceRange() << RHS->getSourceRange(); 3975} 3976 3977//===--- CHECK: Integer mixed-sign comparisons (-Wsign-compare) --------===// 3978//===--- CHECK: Lossy implicit conversions (-Wconversion) --------------===// 3979 3980namespace { 3981 3982/// Structure recording the 'active' range of an integer-valued 3983/// expression. 3984struct IntRange { 3985 /// The number of bits active in the int. 3986 unsigned Width; 3987 3988 /// True if the int is known not to have negative values. 3989 bool NonNegative; 3990 3991 IntRange(unsigned Width, bool NonNegative) 3992 : Width(Width), NonNegative(NonNegative) 3993 {} 3994 3995 /// Returns the range of the bool type. 3996 static IntRange forBoolType() { 3997 return IntRange(1, true); 3998 } 3999 4000 /// Returns the range of an opaque value of the given integral type. 4001 static IntRange forValueOfType(ASTContext &C, QualType T) { 4002 return forValueOfCanonicalType(C, 4003 T->getCanonicalTypeInternal().getTypePtr()); 4004 } 4005 4006 /// Returns the range of an opaque value of a canonical integral type. 4007 static IntRange forValueOfCanonicalType(ASTContext &C, const Type *T) { 4008 assert(T->isCanonicalUnqualified()); 4009 4010 if (const VectorType *VT = dyn_cast<VectorType>(T)) 4011 T = VT->getElementType().getTypePtr(); 4012 if (const ComplexType *CT = dyn_cast<ComplexType>(T)) 4013 T = CT->getElementType().getTypePtr(); 4014 4015 // For enum types, use the known bit width of the enumerators. 4016 if (const EnumType *ET = dyn_cast<EnumType>(T)) { 4017 EnumDecl *Enum = ET->getDecl(); 4018 if (!Enum->isCompleteDefinition()) 4019 return IntRange(C.getIntWidth(QualType(T, 0)), false); 4020 4021 unsigned NumPositive = Enum->getNumPositiveBits(); 4022 unsigned NumNegative = Enum->getNumNegativeBits(); 4023 4024 if (NumNegative == 0) 4025 return IntRange(NumPositive, true/*NonNegative*/); 4026 else 4027 return IntRange(std::max(NumPositive + 1, NumNegative), 4028 false/*NonNegative*/); 4029 } 4030 4031 const BuiltinType *BT = cast<BuiltinType>(T); 4032 assert(BT->isInteger()); 4033 4034 return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger()); 4035 } 4036 4037 /// Returns the "target" range of a canonical integral type, i.e. 4038 /// the range of values expressible in the type. 4039 /// 4040 /// This matches forValueOfCanonicalType except that enums have the 4041 /// full range of their type, not the range of their enumerators. 4042 static IntRange forTargetOfCanonicalType(ASTContext &C, const Type *T) { 4043 assert(T->isCanonicalUnqualified()); 4044 4045 if (const VectorType *VT = dyn_cast<VectorType>(T)) 4046 T = VT->getElementType().getTypePtr(); 4047 if (const ComplexType *CT = dyn_cast<ComplexType>(T)) 4048 T = CT->getElementType().getTypePtr(); 4049 if (const EnumType *ET = dyn_cast<EnumType>(T)) 4050 T = C.getCanonicalType(ET->getDecl()->getIntegerType()).getTypePtr(); 4051 4052 const BuiltinType *BT = cast<BuiltinType>(T); 4053 assert(BT->isInteger()); 4054 4055 return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger()); 4056 } 4057 4058 /// Returns the supremum of two ranges: i.e. their conservative merge. 4059 static IntRange join(IntRange L, IntRange R) { 4060 return IntRange(std::max(L.Width, R.Width), 4061 L.NonNegative && R.NonNegative); 4062 } 4063 4064 /// Returns the infinum of two ranges: i.e. their aggressive merge. 4065 static IntRange meet(IntRange L, IntRange R) { 4066 return IntRange(std::min(L.Width, R.Width), 4067 L.NonNegative || R.NonNegative); 4068 } 4069}; 4070 4071static IntRange GetValueRange(ASTContext &C, llvm::APSInt &value, 4072 unsigned MaxWidth) { 4073 if (value.isSigned() && value.isNegative()) 4074 return IntRange(value.getMinSignedBits(), false); 4075 4076 if (value.getBitWidth() > MaxWidth) 4077 value = value.trunc(MaxWidth); 4078 4079 // isNonNegative() just checks the sign bit without considering 4080 // signedness. 4081 return IntRange(value.getActiveBits(), true); 4082} 4083 4084static IntRange GetValueRange(ASTContext &C, APValue &result, QualType Ty, 4085 unsigned MaxWidth) { 4086 if (result.isInt()) 4087 return GetValueRange(C, result.getInt(), MaxWidth); 4088 4089 if (result.isVector()) { 4090 IntRange R = GetValueRange(C, result.getVectorElt(0), Ty, MaxWidth); 4091 for (unsigned i = 1, e = result.getVectorLength(); i != e; ++i) { 4092 IntRange El = GetValueRange(C, result.getVectorElt(i), Ty, MaxWidth); 4093 R = IntRange::join(R, El); 4094 } 4095 return R; 4096 } 4097 4098 if (result.isComplexInt()) { 4099 IntRange R = GetValueRange(C, result.getComplexIntReal(), MaxWidth); 4100 IntRange I = GetValueRange(C, result.getComplexIntImag(), MaxWidth); 4101 return IntRange::join(R, I); 4102 } 4103 4104 // This can happen with lossless casts to intptr_t of "based" lvalues. 4105 // Assume it might use arbitrary bits. 4106 // FIXME: The only reason we need to pass the type in here is to get 4107 // the sign right on this one case. It would be nice if APValue 4108 // preserved this. 4109 assert(result.isLValue() || result.isAddrLabelDiff()); 4110 return IntRange(MaxWidth, Ty->isUnsignedIntegerOrEnumerationType()); 4111} 4112 4113/// Pseudo-evaluate the given integer expression, estimating the 4114/// range of values it might take. 4115/// 4116/// \param MaxWidth - the width to which the value will be truncated 4117static IntRange GetExprRange(ASTContext &C, Expr *E, unsigned MaxWidth) { 4118 E = E->IgnoreParens(); 4119 4120 // Try a full evaluation first. 4121 Expr::EvalResult result; 4122 if (E->EvaluateAsRValue(result, C)) 4123 return GetValueRange(C, result.Val, E->getType(), MaxWidth); 4124 4125 // I think we only want to look through implicit casts here; if the 4126 // user has an explicit widening cast, we should treat the value as 4127 // being of the new, wider type. 4128 if (ImplicitCastExpr *CE = dyn_cast<ImplicitCastExpr>(E)) { 4129 if (CE->getCastKind() == CK_NoOp || CE->getCastKind() == CK_LValueToRValue) 4130 return GetExprRange(C, CE->getSubExpr(), MaxWidth); 4131 4132 IntRange OutputTypeRange = IntRange::forValueOfType(C, CE->getType()); 4133 4134 bool isIntegerCast = (CE->getCastKind() == CK_IntegralCast); 4135 4136 // Assume that non-integer casts can span the full range of the type. 4137 if (!isIntegerCast) 4138 return OutputTypeRange; 4139 4140 IntRange SubRange 4141 = GetExprRange(C, CE->getSubExpr(), 4142 std::min(MaxWidth, OutputTypeRange.Width)); 4143 4144 // Bail out if the subexpr's range is as wide as the cast type. 4145 if (SubRange.Width >= OutputTypeRange.Width) 4146 return OutputTypeRange; 4147 4148 // Otherwise, we take the smaller width, and we're non-negative if 4149 // either the output type or the subexpr is. 4150 return IntRange(SubRange.Width, 4151 SubRange.NonNegative || OutputTypeRange.NonNegative); 4152 } 4153 4154 if (ConditionalOperator *CO = dyn_cast<ConditionalOperator>(E)) { 4155 // If we can fold the condition, just take that operand. 4156 bool CondResult; 4157 if (CO->getCond()->EvaluateAsBooleanCondition(CondResult, C)) 4158 return GetExprRange(C, CondResult ? CO->getTrueExpr() 4159 : CO->getFalseExpr(), 4160 MaxWidth); 4161 4162 // Otherwise, conservatively merge. 4163 IntRange L = GetExprRange(C, CO->getTrueExpr(), MaxWidth); 4164 IntRange R = GetExprRange(C, CO->getFalseExpr(), MaxWidth); 4165 return IntRange::join(L, R); 4166 } 4167 4168 if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) { 4169 switch (BO->getOpcode()) { 4170 4171 // Boolean-valued operations are single-bit and positive. 4172 case BO_LAnd: 4173 case BO_LOr: 4174 case BO_LT: 4175 case BO_GT: 4176 case BO_LE: 4177 case BO_GE: 4178 case BO_EQ: 4179 case BO_NE: 4180 return IntRange::forBoolType(); 4181 4182 // The type of the assignments is the type of the LHS, so the RHS 4183 // is not necessarily the same type. 4184 case BO_MulAssign: 4185 case BO_DivAssign: 4186 case BO_RemAssign: 4187 case BO_AddAssign: 4188 case BO_SubAssign: 4189 case BO_XorAssign: 4190 case BO_OrAssign: 4191 // TODO: bitfields? 4192 return IntRange::forValueOfType(C, E->getType()); 4193 4194 // Simple assignments just pass through the RHS, which will have 4195 // been coerced to the LHS type. 4196 case BO_Assign: 4197 // TODO: bitfields? 4198 return GetExprRange(C, BO->getRHS(), MaxWidth); 4199 4200 // Operations with opaque sources are black-listed. 4201 case BO_PtrMemD: 4202 case BO_PtrMemI: 4203 return IntRange::forValueOfType(C, E->getType()); 4204 4205 // Bitwise-and uses the *infinum* of the two source ranges. 4206 case BO_And: 4207 case BO_AndAssign: 4208 return IntRange::meet(GetExprRange(C, BO->getLHS(), MaxWidth), 4209 GetExprRange(C, BO->getRHS(), MaxWidth)); 4210 4211 // Left shift gets black-listed based on a judgement call. 4212 case BO_Shl: 4213 // ...except that we want to treat '1 << (blah)' as logically 4214 // positive. It's an important idiom. 4215 if (IntegerLiteral *I 4216 = dyn_cast<IntegerLiteral>(BO->getLHS()->IgnoreParenCasts())) { 4217 if (I->getValue() == 1) { 4218 IntRange R = IntRange::forValueOfType(C, E->getType()); 4219 return IntRange(R.Width, /*NonNegative*/ true); 4220 } 4221 } 4222 // fallthrough 4223 4224 case BO_ShlAssign: 4225 return IntRange::forValueOfType(C, E->getType()); 4226 4227 // Right shift by a constant can narrow its left argument. 4228 case BO_Shr: 4229 case BO_ShrAssign: { 4230 IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth); 4231 4232 // If the shift amount is a positive constant, drop the width by 4233 // that much. 4234 llvm::APSInt shift; 4235 if (BO->getRHS()->isIntegerConstantExpr(shift, C) && 4236 shift.isNonNegative()) { 4237 unsigned zext = shift.getZExtValue(); 4238 if (zext >= L.Width) 4239 L.Width = (L.NonNegative ? 0 : 1); 4240 else 4241 L.Width -= zext; 4242 } 4243 4244 return L; 4245 } 4246 4247 // Comma acts as its right operand. 4248 case BO_Comma: 4249 return GetExprRange(C, BO->getRHS(), MaxWidth); 4250 4251 // Black-list pointer subtractions. 4252 case BO_Sub: 4253 if (BO->getLHS()->getType()->isPointerType()) 4254 return IntRange::forValueOfType(C, E->getType()); 4255 break; 4256 4257 // The width of a division result is mostly determined by the size 4258 // of the LHS. 4259 case BO_Div: { 4260 // Don't 'pre-truncate' the operands. 4261 unsigned opWidth = C.getIntWidth(E->getType()); 4262 IntRange L = GetExprRange(C, BO->getLHS(), opWidth); 4263 4264 // If the divisor is constant, use that. 4265 llvm::APSInt divisor; 4266 if (BO->getRHS()->isIntegerConstantExpr(divisor, C)) { 4267 unsigned log2 = divisor.logBase2(); // floor(log_2(divisor)) 4268 if (log2 >= L.Width) 4269 L.Width = (L.NonNegative ? 0 : 1); 4270 else 4271 L.Width = std::min(L.Width - log2, MaxWidth); 4272 return L; 4273 } 4274 4275 // Otherwise, just use the LHS's width. 4276 IntRange R = GetExprRange(C, BO->getRHS(), opWidth); 4277 return IntRange(L.Width, L.NonNegative && R.NonNegative); 4278 } 4279 4280 // The result of a remainder can't be larger than the result of 4281 // either side. 4282 case BO_Rem: { 4283 // Don't 'pre-truncate' the operands. 4284 unsigned opWidth = C.getIntWidth(E->getType()); 4285 IntRange L = GetExprRange(C, BO->getLHS(), opWidth); 4286 IntRange R = GetExprRange(C, BO->getRHS(), opWidth); 4287 4288 IntRange meet = IntRange::meet(L, R); 4289 meet.Width = std::min(meet.Width, MaxWidth); 4290 return meet; 4291 } 4292 4293 // The default behavior is okay for these. 4294 case BO_Mul: 4295 case BO_Add: 4296 case BO_Xor: 4297 case BO_Or: 4298 break; 4299 } 4300 4301 // The default case is to treat the operation as if it were closed 4302 // on the narrowest type that encompasses both operands. 4303 IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth); 4304 IntRange R = GetExprRange(C, BO->getRHS(), MaxWidth); 4305 return IntRange::join(L, R); 4306 } 4307 4308 if (UnaryOperator *UO = dyn_cast<UnaryOperator>(E)) { 4309 switch (UO->getOpcode()) { 4310 // Boolean-valued operations are white-listed. 4311 case UO_LNot: 4312 return IntRange::forBoolType(); 4313 4314 // Operations with opaque sources are black-listed. 4315 case UO_Deref: 4316 case UO_AddrOf: // should be impossible 4317 return IntRange::forValueOfType(C, E->getType()); 4318 4319 default: 4320 return GetExprRange(C, UO->getSubExpr(), MaxWidth); 4321 } 4322 } 4323 4324 if (dyn_cast<OffsetOfExpr>(E)) { 4325 IntRange::forValueOfType(C, E->getType()); 4326 } 4327 4328 if (FieldDecl *BitField = E->getSourceBitField()) 4329 return IntRange(BitField->getBitWidthValue(C), 4330 BitField->getType()->isUnsignedIntegerOrEnumerationType()); 4331 4332 return IntRange::forValueOfType(C, E->getType()); 4333} 4334 4335static IntRange GetExprRange(ASTContext &C, Expr *E) { 4336 return GetExprRange(C, E, C.getIntWidth(E->getType())); 4337} 4338 4339/// Checks whether the given value, which currently has the given 4340/// source semantics, has the same value when coerced through the 4341/// target semantics. 4342static bool IsSameFloatAfterCast(const llvm::APFloat &value, 4343 const llvm::fltSemantics &Src, 4344 const llvm::fltSemantics &Tgt) { 4345 llvm::APFloat truncated = value; 4346 4347 bool ignored; 4348 truncated.convert(Src, llvm::APFloat::rmNearestTiesToEven, &ignored); 4349 truncated.convert(Tgt, llvm::APFloat::rmNearestTiesToEven, &ignored); 4350 4351 return truncated.bitwiseIsEqual(value); 4352} 4353 4354/// Checks whether the given value, which currently has the given 4355/// source semantics, has the same value when coerced through the 4356/// target semantics. 4357/// 4358/// The value might be a vector of floats (or a complex number). 4359static bool IsSameFloatAfterCast(const APValue &value, 4360 const llvm::fltSemantics &Src, 4361 const llvm::fltSemantics &Tgt) { 4362 if (value.isFloat()) 4363 return IsSameFloatAfterCast(value.getFloat(), Src, Tgt); 4364 4365 if (value.isVector()) { 4366 for (unsigned i = 0, e = value.getVectorLength(); i != e; ++i) 4367 if (!IsSameFloatAfterCast(value.getVectorElt(i), Src, Tgt)) 4368 return false; 4369 return true; 4370 } 4371 4372 assert(value.isComplexFloat()); 4373 return (IsSameFloatAfterCast(value.getComplexFloatReal(), Src, Tgt) && 4374 IsSameFloatAfterCast(value.getComplexFloatImag(), Src, Tgt)); 4375} 4376 4377static void AnalyzeImplicitConversions(Sema &S, Expr *E, SourceLocation CC); 4378 4379static bool IsZero(Sema &S, Expr *E) { 4380 // Suppress cases where we are comparing against an enum constant. 4381 if (const DeclRefExpr *DR = 4382 dyn_cast<DeclRefExpr>(E->IgnoreParenImpCasts())) 4383 if (isa<EnumConstantDecl>(DR->getDecl())) 4384 return false; 4385 4386 // Suppress cases where the '0' value is expanded from a macro. 4387 if (E->getLocStart().isMacroID()) 4388 return false; 4389 4390 llvm::APSInt Value; 4391 return E->isIntegerConstantExpr(Value, S.Context) && Value == 0; 4392} 4393 4394static bool HasEnumType(Expr *E) { 4395 // Strip off implicit integral promotions. 4396 while (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(E)) { 4397 if (ICE->getCastKind() != CK_IntegralCast && 4398 ICE->getCastKind() != CK_NoOp) 4399 break; 4400 E = ICE->getSubExpr(); 4401 } 4402 4403 return E->getType()->isEnumeralType(); 4404} 4405 4406static void CheckTrivialUnsignedComparison(Sema &S, BinaryOperator *E) { 4407 BinaryOperatorKind op = E->getOpcode(); 4408 if (E->isValueDependent()) 4409 return; 4410 4411 if (op == BO_LT && IsZero(S, E->getRHS())) { 4412 S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison) 4413 << "< 0" << "false" << HasEnumType(E->getLHS()) 4414 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 4415 } else if (op == BO_GE && IsZero(S, E->getRHS())) { 4416 S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison) 4417 << ">= 0" << "true" << HasEnumType(E->getLHS()) 4418 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 4419 } else if (op == BO_GT && IsZero(S, E->getLHS())) { 4420 S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison) 4421 << "0 >" << "false" << HasEnumType(E->getRHS()) 4422 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 4423 } else if (op == BO_LE && IsZero(S, E->getLHS())) { 4424 S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison) 4425 << "0 <=" << "true" << HasEnumType(E->getRHS()) 4426 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 4427 } 4428} 4429 4430static void DiagnoseOutOfRangeComparison(Sema &S, BinaryOperator *E, 4431 Expr *Constant, Expr *Other, 4432 llvm::APSInt Value, 4433 bool RhsConstant) { 4434 // 0 values are handled later by CheckTrivialUnsignedComparison(). 4435 if (Value == 0) 4436 return; 4437 4438 BinaryOperatorKind op = E->getOpcode(); 4439 QualType OtherT = Other->getType(); 4440 QualType ConstantT = Constant->getType(); 4441 QualType CommonT = E->getLHS()->getType(); 4442 if (S.Context.hasSameUnqualifiedType(OtherT, ConstantT)) 4443 return; 4444 assert((OtherT->isIntegerType() && ConstantT->isIntegerType()) 4445 && "comparison with non-integer type"); 4446 4447 bool ConstantSigned = ConstantT->isSignedIntegerType(); 4448 bool CommonSigned = CommonT->isSignedIntegerType(); 4449 4450 bool EqualityOnly = false; 4451 4452 // TODO: Investigate using GetExprRange() to get tighter bounds on 4453 // on the bit ranges. 4454 IntRange OtherRange = IntRange::forValueOfType(S.Context, OtherT); 4455 unsigned OtherWidth = OtherRange.Width; 4456 4457 if (CommonSigned) { 4458 // The common type is signed, therefore no signed to unsigned conversion. 4459 if (!OtherRange.NonNegative) { 4460 // Check that the constant is representable in type OtherT. 4461 if (ConstantSigned) { 4462 if (OtherWidth >= Value.getMinSignedBits()) 4463 return; 4464 } else { // !ConstantSigned 4465 if (OtherWidth >= Value.getActiveBits() + 1) 4466 return; 4467 } 4468 } else { // !OtherSigned 4469 // Check that the constant is representable in type OtherT. 4470 // Negative values are out of range. 4471 if (ConstantSigned) { 4472 if (Value.isNonNegative() && OtherWidth >= Value.getActiveBits()) 4473 return; 4474 } else { // !ConstantSigned 4475 if (OtherWidth >= Value.getActiveBits()) 4476 return; 4477 } 4478 } 4479 } else { // !CommonSigned 4480 if (OtherRange.NonNegative) { 4481 if (OtherWidth >= Value.getActiveBits()) 4482 return; 4483 } else if (!OtherRange.NonNegative && !ConstantSigned) { 4484 // Check to see if the constant is representable in OtherT. 4485 if (OtherWidth > Value.getActiveBits()) 4486 return; 4487 // Check to see if the constant is equivalent to a negative value 4488 // cast to CommonT. 4489 if (S.Context.getIntWidth(ConstantT) == S.Context.getIntWidth(CommonT) && 4490 Value.isNegative() && Value.getMinSignedBits() <= OtherWidth) 4491 return; 4492 // The constant value rests between values that OtherT can represent after 4493 // conversion. Relational comparison still works, but equality 4494 // comparisons will be tautological. 4495 EqualityOnly = true; 4496 } else { // OtherSigned && ConstantSigned 4497 assert(0 && "Two signed types converted to unsigned types."); 4498 } 4499 } 4500 4501 bool PositiveConstant = !ConstantSigned || Value.isNonNegative(); 4502 4503 bool IsTrue = true; 4504 if (op == BO_EQ || op == BO_NE) { 4505 IsTrue = op == BO_NE; 4506 } else if (EqualityOnly) { 4507 return; 4508 } else if (RhsConstant) { 4509 if (op == BO_GT || op == BO_GE) 4510 IsTrue = !PositiveConstant; 4511 else // op == BO_LT || op == BO_LE 4512 IsTrue = PositiveConstant; 4513 } else { 4514 if (op == BO_LT || op == BO_LE) 4515 IsTrue = !PositiveConstant; 4516 else // op == BO_GT || op == BO_GE 4517 IsTrue = PositiveConstant; 4518 } 4519 4520 // If this is a comparison to an enum constant, include that 4521 // constant in the diagnostic. 4522 const EnumConstantDecl *ED = 0; 4523 if (const DeclRefExpr *DR = dyn_cast<DeclRefExpr>(Constant)) 4524 ED = dyn_cast<EnumConstantDecl>(DR->getDecl()); 4525 4526 SmallString<64> PrettySourceValue; 4527 llvm::raw_svector_ostream OS(PrettySourceValue); 4528 if (ED) 4529 OS << '\'' << *ED << "' (" << Value << ")"; 4530 else 4531 OS << Value; 4532 4533 S.Diag(E->getOperatorLoc(), diag::warn_out_of_range_compare) 4534 << OS.str() << OtherT << IsTrue 4535 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 4536} 4537 4538/// Analyze the operands of the given comparison. Implements the 4539/// fallback case from AnalyzeComparison. 4540static void AnalyzeImpConvsInComparison(Sema &S, BinaryOperator *E) { 4541 AnalyzeImplicitConversions(S, E->getLHS(), E->getOperatorLoc()); 4542 AnalyzeImplicitConversions(S, E->getRHS(), E->getOperatorLoc()); 4543} 4544 4545/// \brief Implements -Wsign-compare. 4546/// 4547/// \param E the binary operator to check for warnings 4548static void AnalyzeComparison(Sema &S, BinaryOperator *E) { 4549 // The type the comparison is being performed in. 4550 QualType T = E->getLHS()->getType(); 4551 assert(S.Context.hasSameUnqualifiedType(T, E->getRHS()->getType()) 4552 && "comparison with mismatched types"); 4553 if (E->isValueDependent()) 4554 return AnalyzeImpConvsInComparison(S, E); 4555 4556 Expr *LHS = E->getLHS()->IgnoreParenImpCasts(); 4557 Expr *RHS = E->getRHS()->IgnoreParenImpCasts(); 4558 4559 bool IsComparisonConstant = false; 4560 4561 // Check whether an integer constant comparison results in a value 4562 // of 'true' or 'false'. 4563 if (T->isIntegralType(S.Context)) { 4564 llvm::APSInt RHSValue; 4565 bool IsRHSIntegralLiteral = 4566 RHS->isIntegerConstantExpr(RHSValue, S.Context); 4567 llvm::APSInt LHSValue; 4568 bool IsLHSIntegralLiteral = 4569 LHS->isIntegerConstantExpr(LHSValue, S.Context); 4570 if (IsRHSIntegralLiteral && !IsLHSIntegralLiteral) 4571 DiagnoseOutOfRangeComparison(S, E, RHS, LHS, RHSValue, true); 4572 else if (!IsRHSIntegralLiteral && IsLHSIntegralLiteral) 4573 DiagnoseOutOfRangeComparison(S, E, LHS, RHS, LHSValue, false); 4574 else 4575 IsComparisonConstant = 4576 (IsRHSIntegralLiteral && IsLHSIntegralLiteral); 4577 } else if (!T->hasUnsignedIntegerRepresentation()) 4578 IsComparisonConstant = E->isIntegerConstantExpr(S.Context); 4579 4580 // We don't do anything special if this isn't an unsigned integral 4581 // comparison: we're only interested in integral comparisons, and 4582 // signed comparisons only happen in cases we don't care to warn about. 4583 // 4584 // We also don't care about value-dependent expressions or expressions 4585 // whose result is a constant. 4586 if (!T->hasUnsignedIntegerRepresentation() || IsComparisonConstant) 4587 return AnalyzeImpConvsInComparison(S, E); 4588 4589 // Check to see if one of the (unmodified) operands is of different 4590 // signedness. 4591 Expr *signedOperand, *unsignedOperand; 4592 if (LHS->getType()->hasSignedIntegerRepresentation()) { 4593 assert(!RHS->getType()->hasSignedIntegerRepresentation() && 4594 "unsigned comparison between two signed integer expressions?"); 4595 signedOperand = LHS; 4596 unsignedOperand = RHS; 4597 } else if (RHS->getType()->hasSignedIntegerRepresentation()) { 4598 signedOperand = RHS; 4599 unsignedOperand = LHS; 4600 } else { 4601 CheckTrivialUnsignedComparison(S, E); 4602 return AnalyzeImpConvsInComparison(S, E); 4603 } 4604 4605 // Otherwise, calculate the effective range of the signed operand. 4606 IntRange signedRange = GetExprRange(S.Context, signedOperand); 4607 4608 // Go ahead and analyze implicit conversions in the operands. Note 4609 // that we skip the implicit conversions on both sides. 4610 AnalyzeImplicitConversions(S, LHS, E->getOperatorLoc()); 4611 AnalyzeImplicitConversions(S, RHS, E->getOperatorLoc()); 4612 4613 // If the signed range is non-negative, -Wsign-compare won't fire, 4614 // but we should still check for comparisons which are always true 4615 // or false. 4616 if (signedRange.NonNegative) 4617 return CheckTrivialUnsignedComparison(S, E); 4618 4619 // For (in)equality comparisons, if the unsigned operand is a 4620 // constant which cannot collide with a overflowed signed operand, 4621 // then reinterpreting the signed operand as unsigned will not 4622 // change the result of the comparison. 4623 if (E->isEqualityOp()) { 4624 unsigned comparisonWidth = S.Context.getIntWidth(T); 4625 IntRange unsignedRange = GetExprRange(S.Context, unsignedOperand); 4626 4627 // We should never be unable to prove that the unsigned operand is 4628 // non-negative. 4629 assert(unsignedRange.NonNegative && "unsigned range includes negative?"); 4630 4631 if (unsignedRange.Width < comparisonWidth) 4632 return; 4633 } 4634 4635 S.DiagRuntimeBehavior(E->getOperatorLoc(), E, 4636 S.PDiag(diag::warn_mixed_sign_comparison) 4637 << LHS->getType() << RHS->getType() 4638 << LHS->getSourceRange() << RHS->getSourceRange()); 4639} 4640 4641/// Analyzes an attempt to assign the given value to a bitfield. 4642/// 4643/// Returns true if there was something fishy about the attempt. 4644static bool AnalyzeBitFieldAssignment(Sema &S, FieldDecl *Bitfield, Expr *Init, 4645 SourceLocation InitLoc) { 4646 assert(Bitfield->isBitField()); 4647 if (Bitfield->isInvalidDecl()) 4648 return false; 4649 4650 // White-list bool bitfields. 4651 if (Bitfield->getType()->isBooleanType()) 4652 return false; 4653 4654 // Ignore value- or type-dependent expressions. 4655 if (Bitfield->getBitWidth()->isValueDependent() || 4656 Bitfield->getBitWidth()->isTypeDependent() || 4657 Init->isValueDependent() || 4658 Init->isTypeDependent()) 4659 return false; 4660 4661 Expr *OriginalInit = Init->IgnoreParenImpCasts(); 4662 4663 llvm::APSInt Value; 4664 if (!OriginalInit->EvaluateAsInt(Value, S.Context, Expr::SE_AllowSideEffects)) 4665 return false; 4666 4667 unsigned OriginalWidth = Value.getBitWidth(); 4668 unsigned FieldWidth = Bitfield->getBitWidthValue(S.Context); 4669 4670 if (OriginalWidth <= FieldWidth) 4671 return false; 4672 4673 // Compute the value which the bitfield will contain. 4674 llvm::APSInt TruncatedValue = Value.trunc(FieldWidth); 4675 TruncatedValue.setIsSigned(Bitfield->getType()->isSignedIntegerType()); 4676 4677 // Check whether the stored value is equal to the original value. 4678 TruncatedValue = TruncatedValue.extend(OriginalWidth); 4679 if (llvm::APSInt::isSameValue(Value, TruncatedValue)) 4680 return false; 4681 4682 // Special-case bitfields of width 1: booleans are naturally 0/1, and 4683 // therefore don't strictly fit into a signed bitfield of width 1. 4684 if (FieldWidth == 1 && Value == 1) 4685 return false; 4686 4687 std::string PrettyValue = Value.toString(10); 4688 std::string PrettyTrunc = TruncatedValue.toString(10); 4689 4690 S.Diag(InitLoc, diag::warn_impcast_bitfield_precision_constant) 4691 << PrettyValue << PrettyTrunc << OriginalInit->getType() 4692 << Init->getSourceRange(); 4693 4694 return true; 4695} 4696 4697/// Analyze the given simple or compound assignment for warning-worthy 4698/// operations. 4699static void AnalyzeAssignment(Sema &S, BinaryOperator *E) { 4700 // Just recurse on the LHS. 4701 AnalyzeImplicitConversions(S, E->getLHS(), E->getOperatorLoc()); 4702 4703 // We want to recurse on the RHS as normal unless we're assigning to 4704 // a bitfield. 4705 if (FieldDecl *Bitfield = E->getLHS()->getSourceBitField()) { 4706 if (AnalyzeBitFieldAssignment(S, Bitfield, E->getRHS(), 4707 E->getOperatorLoc())) { 4708 // Recurse, ignoring any implicit conversions on the RHS. 4709 return AnalyzeImplicitConversions(S, E->getRHS()->IgnoreParenImpCasts(), 4710 E->getOperatorLoc()); 4711 } 4712 } 4713 4714 AnalyzeImplicitConversions(S, E->getRHS(), E->getOperatorLoc()); 4715} 4716 4717/// Diagnose an implicit cast; purely a helper for CheckImplicitConversion. 4718static void DiagnoseImpCast(Sema &S, Expr *E, QualType SourceType, QualType T, 4719 SourceLocation CContext, unsigned diag, 4720 bool pruneControlFlow = false) { 4721 if (pruneControlFlow) { 4722 S.DiagRuntimeBehavior(E->getExprLoc(), E, 4723 S.PDiag(diag) 4724 << SourceType << T << E->getSourceRange() 4725 << SourceRange(CContext)); 4726 return; 4727 } 4728 S.Diag(E->getExprLoc(), diag) 4729 << SourceType << T << E->getSourceRange() << SourceRange(CContext); 4730} 4731 4732/// Diagnose an implicit cast; purely a helper for CheckImplicitConversion. 4733static void DiagnoseImpCast(Sema &S, Expr *E, QualType T, 4734 SourceLocation CContext, unsigned diag, 4735 bool pruneControlFlow = false) { 4736 DiagnoseImpCast(S, E, E->getType(), T, CContext, diag, pruneControlFlow); 4737} 4738 4739/// Diagnose an implicit cast from a literal expression. Does not warn when the 4740/// cast wouldn't lose information. 4741void DiagnoseFloatingLiteralImpCast(Sema &S, FloatingLiteral *FL, QualType T, 4742 SourceLocation CContext) { 4743 // Try to convert the literal exactly to an integer. If we can, don't warn. 4744 bool isExact = false; 4745 const llvm::APFloat &Value = FL->getValue(); 4746 llvm::APSInt IntegerValue(S.Context.getIntWidth(T), 4747 T->hasUnsignedIntegerRepresentation()); 4748 if (Value.convertToInteger(IntegerValue, 4749 llvm::APFloat::rmTowardZero, &isExact) 4750 == llvm::APFloat::opOK && isExact) 4751 return; 4752 4753 SmallString<16> PrettySourceValue; 4754 Value.toString(PrettySourceValue); 4755 SmallString<16> PrettyTargetValue; 4756 if (T->isSpecificBuiltinType(BuiltinType::Bool)) 4757 PrettyTargetValue = IntegerValue == 0 ? "false" : "true"; 4758 else 4759 IntegerValue.toString(PrettyTargetValue); 4760 4761 S.Diag(FL->getExprLoc(), diag::warn_impcast_literal_float_to_integer) 4762 << FL->getType() << T.getUnqualifiedType() << PrettySourceValue 4763 << PrettyTargetValue << FL->getSourceRange() << SourceRange(CContext); 4764} 4765 4766std::string PrettyPrintInRange(const llvm::APSInt &Value, IntRange Range) { 4767 if (!Range.Width) return "0"; 4768 4769 llvm::APSInt ValueInRange = Value; 4770 ValueInRange.setIsSigned(!Range.NonNegative); 4771 ValueInRange = ValueInRange.trunc(Range.Width); 4772 return ValueInRange.toString(10); 4773} 4774 4775static bool IsImplicitBoolFloatConversion(Sema &S, Expr *Ex, bool ToBool) { 4776 if (!isa<ImplicitCastExpr>(Ex)) 4777 return false; 4778 4779 Expr *InnerE = Ex->IgnoreParenImpCasts(); 4780 const Type *Target = S.Context.getCanonicalType(Ex->getType()).getTypePtr(); 4781 const Type *Source = 4782 S.Context.getCanonicalType(InnerE->getType()).getTypePtr(); 4783 if (Target->isDependentType()) 4784 return false; 4785 4786 const BuiltinType *FloatCandidateBT = 4787 dyn_cast<BuiltinType>(ToBool ? Source : Target); 4788 const Type *BoolCandidateType = ToBool ? Target : Source; 4789 4790 return (BoolCandidateType->isSpecificBuiltinType(BuiltinType::Bool) && 4791 FloatCandidateBT && (FloatCandidateBT->isFloatingPoint())); 4792} 4793 4794void CheckImplicitArgumentConversions(Sema &S, CallExpr *TheCall, 4795 SourceLocation CC) { 4796 unsigned NumArgs = TheCall->getNumArgs(); 4797 for (unsigned i = 0; i < NumArgs; ++i) { 4798 Expr *CurrA = TheCall->getArg(i); 4799 if (!IsImplicitBoolFloatConversion(S, CurrA, true)) 4800 continue; 4801 4802 bool IsSwapped = ((i > 0) && 4803 IsImplicitBoolFloatConversion(S, TheCall->getArg(i - 1), false)); 4804 IsSwapped |= ((i < (NumArgs - 1)) && 4805 IsImplicitBoolFloatConversion(S, TheCall->getArg(i + 1), false)); 4806 if (IsSwapped) { 4807 // Warn on this floating-point to bool conversion. 4808 DiagnoseImpCast(S, CurrA->IgnoreParenImpCasts(), 4809 CurrA->getType(), CC, 4810 diag::warn_impcast_floating_point_to_bool); 4811 } 4812 } 4813} 4814 4815void CheckImplicitConversion(Sema &S, Expr *E, QualType T, 4816 SourceLocation CC, bool *ICContext = 0) { 4817 if (E->isTypeDependent() || E->isValueDependent()) return; 4818 4819 const Type *Source = S.Context.getCanonicalType(E->getType()).getTypePtr(); 4820 const Type *Target = S.Context.getCanonicalType(T).getTypePtr(); 4821 if (Source == Target) return; 4822 if (Target->isDependentType()) return; 4823 4824 // If the conversion context location is invalid don't complain. We also 4825 // don't want to emit a warning if the issue occurs from the expansion of 4826 // a system macro. The problem is that 'getSpellingLoc()' is slow, so we 4827 // delay this check as long as possible. Once we detect we are in that 4828 // scenario, we just return. 4829 if (CC.isInvalid()) 4830 return; 4831 4832 // Diagnose implicit casts to bool. 4833 if (Target->isSpecificBuiltinType(BuiltinType::Bool)) { 4834 if (isa<StringLiteral>(E)) 4835 // Warn on string literal to bool. Checks for string literals in logical 4836 // expressions, for instances, assert(0 && "error here"), is prevented 4837 // by a check in AnalyzeImplicitConversions(). 4838 return DiagnoseImpCast(S, E, T, CC, 4839 diag::warn_impcast_string_literal_to_bool); 4840 if (Source->isFunctionType()) { 4841 // Warn on function to bool. Checks free functions and static member 4842 // functions. Weakly imported functions are excluded from the check, 4843 // since it's common to test their value to check whether the linker 4844 // found a definition for them. 4845 ValueDecl *D = 0; 4846 if (DeclRefExpr* R = dyn_cast<DeclRefExpr>(E)) { 4847 D = R->getDecl(); 4848 } else if (MemberExpr *M = dyn_cast<MemberExpr>(E)) { 4849 D = M->getMemberDecl(); 4850 } 4851 4852 if (D && !D->isWeak()) { 4853 if (FunctionDecl* F = dyn_cast<FunctionDecl>(D)) { 4854 S.Diag(E->getExprLoc(), diag::warn_impcast_function_to_bool) 4855 << F << E->getSourceRange() << SourceRange(CC); 4856 S.Diag(E->getExprLoc(), diag::note_function_to_bool_silence) 4857 << FixItHint::CreateInsertion(E->getExprLoc(), "&"); 4858 QualType ReturnType; 4859 UnresolvedSet<4> NonTemplateOverloads; 4860 S.isExprCallable(*E, ReturnType, NonTemplateOverloads); 4861 if (!ReturnType.isNull() 4862 && ReturnType->isSpecificBuiltinType(BuiltinType::Bool)) 4863 S.Diag(E->getExprLoc(), diag::note_function_to_bool_call) 4864 << FixItHint::CreateInsertion( 4865 S.getPreprocessor().getLocForEndOfToken(E->getLocEnd()), "()"); 4866 return; 4867 } 4868 } 4869 } 4870 } 4871 4872 // Strip vector types. 4873 if (isa<VectorType>(Source)) { 4874 if (!isa<VectorType>(Target)) { 4875 if (S.SourceMgr.isInSystemMacro(CC)) 4876 return; 4877 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_vector_scalar); 4878 } 4879 4880 // If the vector cast is cast between two vectors of the same size, it is 4881 // a bitcast, not a conversion. 4882 if (S.Context.getTypeSize(Source) == S.Context.getTypeSize(Target)) 4883 return; 4884 4885 Source = cast<VectorType>(Source)->getElementType().getTypePtr(); 4886 Target = cast<VectorType>(Target)->getElementType().getTypePtr(); 4887 } 4888 4889 // Strip complex types. 4890 if (isa<ComplexType>(Source)) { 4891 if (!isa<ComplexType>(Target)) { 4892 if (S.SourceMgr.isInSystemMacro(CC)) 4893 return; 4894 4895 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_complex_scalar); 4896 } 4897 4898 Source = cast<ComplexType>(Source)->getElementType().getTypePtr(); 4899 Target = cast<ComplexType>(Target)->getElementType().getTypePtr(); 4900 } 4901 4902 const BuiltinType *SourceBT = dyn_cast<BuiltinType>(Source); 4903 const BuiltinType *TargetBT = dyn_cast<BuiltinType>(Target); 4904 4905 // If the source is floating point... 4906 if (SourceBT && SourceBT->isFloatingPoint()) { 4907 // ...and the target is floating point... 4908 if (TargetBT && TargetBT->isFloatingPoint()) { 4909 // ...then warn if we're dropping FP rank. 4910 4911 // Builtin FP kinds are ordered by increasing FP rank. 4912 if (SourceBT->getKind() > TargetBT->getKind()) { 4913 // Don't warn about float constants that are precisely 4914 // representable in the target type. 4915 Expr::EvalResult result; 4916 if (E->EvaluateAsRValue(result, S.Context)) { 4917 // Value might be a float, a float vector, or a float complex. 4918 if (IsSameFloatAfterCast(result.Val, 4919 S.Context.getFloatTypeSemantics(QualType(TargetBT, 0)), 4920 S.Context.getFloatTypeSemantics(QualType(SourceBT, 0)))) 4921 return; 4922 } 4923 4924 if (S.SourceMgr.isInSystemMacro(CC)) 4925 return; 4926 4927 DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_float_precision); 4928 } 4929 return; 4930 } 4931 4932 // If the target is integral, always warn. 4933 if (TargetBT && TargetBT->isInteger()) { 4934 if (S.SourceMgr.isInSystemMacro(CC)) 4935 return; 4936 4937 Expr *InnerE = E->IgnoreParenImpCasts(); 4938 // We also want to warn on, e.g., "int i = -1.234" 4939 if (UnaryOperator *UOp = dyn_cast<UnaryOperator>(InnerE)) 4940 if (UOp->getOpcode() == UO_Minus || UOp->getOpcode() == UO_Plus) 4941 InnerE = UOp->getSubExpr()->IgnoreParenImpCasts(); 4942 4943 if (FloatingLiteral *FL = dyn_cast<FloatingLiteral>(InnerE)) { 4944 DiagnoseFloatingLiteralImpCast(S, FL, T, CC); 4945 } else { 4946 DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_float_integer); 4947 } 4948 } 4949 4950 // If the target is bool, warn if expr is a function or method call. 4951 if (Target->isSpecificBuiltinType(BuiltinType::Bool) && 4952 isa<CallExpr>(E)) { 4953 // Check last argument of function call to see if it is an 4954 // implicit cast from a type matching the type the result 4955 // is being cast to. 4956 CallExpr *CEx = cast<CallExpr>(E); 4957 unsigned NumArgs = CEx->getNumArgs(); 4958 if (NumArgs > 0) { 4959 Expr *LastA = CEx->getArg(NumArgs - 1); 4960 Expr *InnerE = LastA->IgnoreParenImpCasts(); 4961 const Type *InnerType = 4962 S.Context.getCanonicalType(InnerE->getType()).getTypePtr(); 4963 if (isa<ImplicitCastExpr>(LastA) && (InnerType == Target)) { 4964 // Warn on this floating-point to bool conversion 4965 DiagnoseImpCast(S, E, T, CC, 4966 diag::warn_impcast_floating_point_to_bool); 4967 } 4968 } 4969 } 4970 return; 4971 } 4972 4973 if ((E->isNullPointerConstant(S.Context, Expr::NPC_ValueDependentIsNotNull) 4974 == Expr::NPCK_GNUNull) && !Target->isAnyPointerType() 4975 && !Target->isBlockPointerType() && !Target->isMemberPointerType() 4976 && Target->isScalarType() && !Target->isNullPtrType()) { 4977 SourceLocation Loc = E->getSourceRange().getBegin(); 4978 if (Loc.isMacroID()) 4979 Loc = S.SourceMgr.getImmediateExpansionRange(Loc).first; 4980 if (!Loc.isMacroID() || CC.isMacroID()) 4981 S.Diag(Loc, diag::warn_impcast_null_pointer_to_integer) 4982 << T << clang::SourceRange(CC) 4983 << FixItHint::CreateReplacement(Loc, S.getFixItZeroLiteralForType(T)); 4984 } 4985 4986 if (!Source->isIntegerType() || !Target->isIntegerType()) 4987 return; 4988 4989 // TODO: remove this early return once the false positives for constant->bool 4990 // in templates, macros, etc, are reduced or removed. 4991 if (Target->isSpecificBuiltinType(BuiltinType::Bool)) 4992 return; 4993 4994 IntRange SourceRange = GetExprRange(S.Context, E); 4995 IntRange TargetRange = IntRange::forTargetOfCanonicalType(S.Context, Target); 4996 4997 if (SourceRange.Width > TargetRange.Width) { 4998 // If the source is a constant, use a default-on diagnostic. 4999 // TODO: this should happen for bitfield stores, too. 5000 llvm::APSInt Value(32); 5001 if (E->isIntegerConstantExpr(Value, S.Context)) { 5002 if (S.SourceMgr.isInSystemMacro(CC)) 5003 return; 5004 5005 std::string PrettySourceValue = Value.toString(10); 5006 std::string PrettyTargetValue = PrettyPrintInRange(Value, TargetRange); 5007 5008 S.DiagRuntimeBehavior(E->getExprLoc(), E, 5009 S.PDiag(diag::warn_impcast_integer_precision_constant) 5010 << PrettySourceValue << PrettyTargetValue 5011 << E->getType() << T << E->getSourceRange() 5012 << clang::SourceRange(CC)); 5013 return; 5014 } 5015 5016 // People want to build with -Wshorten-64-to-32 and not -Wconversion. 5017 if (S.SourceMgr.isInSystemMacro(CC)) 5018 return; 5019 5020 if (TargetRange.Width == 32 && S.Context.getIntWidth(E->getType()) == 64) 5021 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_integer_64_32, 5022 /* pruneControlFlow */ true); 5023 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_integer_precision); 5024 } 5025 5026 if ((TargetRange.NonNegative && !SourceRange.NonNegative) || 5027 (!TargetRange.NonNegative && SourceRange.NonNegative && 5028 SourceRange.Width == TargetRange.Width)) { 5029 5030 if (S.SourceMgr.isInSystemMacro(CC)) 5031 return; 5032 5033 unsigned DiagID = diag::warn_impcast_integer_sign; 5034 5035 // Traditionally, gcc has warned about this under -Wsign-compare. 5036 // We also want to warn about it in -Wconversion. 5037 // So if -Wconversion is off, use a completely identical diagnostic 5038 // in the sign-compare group. 5039 // The conditional-checking code will 5040 if (ICContext) { 5041 DiagID = diag::warn_impcast_integer_sign_conditional; 5042 *ICContext = true; 5043 } 5044 5045 return DiagnoseImpCast(S, E, T, CC, DiagID); 5046 } 5047 5048 // Diagnose conversions between different enumeration types. 5049 // In C, we pretend that the type of an EnumConstantDecl is its enumeration 5050 // type, to give us better diagnostics. 5051 QualType SourceType = E->getType(); 5052 if (!S.getLangOpts().CPlusPlus) { 5053 if (DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(E)) 5054 if (EnumConstantDecl *ECD = dyn_cast<EnumConstantDecl>(DRE->getDecl())) { 5055 EnumDecl *Enum = cast<EnumDecl>(ECD->getDeclContext()); 5056 SourceType = S.Context.getTypeDeclType(Enum); 5057 Source = S.Context.getCanonicalType(SourceType).getTypePtr(); 5058 } 5059 } 5060 5061 if (const EnumType *SourceEnum = Source->getAs<EnumType>()) 5062 if (const EnumType *TargetEnum = Target->getAs<EnumType>()) 5063 if (SourceEnum->getDecl()->hasNameForLinkage() && 5064 TargetEnum->getDecl()->hasNameForLinkage() && 5065 SourceEnum != TargetEnum) { 5066 if (S.SourceMgr.isInSystemMacro(CC)) 5067 return; 5068 5069 return DiagnoseImpCast(S, E, SourceType, T, CC, 5070 diag::warn_impcast_different_enum_types); 5071 } 5072 5073 return; 5074} 5075 5076void CheckConditionalOperator(Sema &S, ConditionalOperator *E, 5077 SourceLocation CC, QualType T); 5078 5079void CheckConditionalOperand(Sema &S, Expr *E, QualType T, 5080 SourceLocation CC, bool &ICContext) { 5081 E = E->IgnoreParenImpCasts(); 5082 5083 if (isa<ConditionalOperator>(E)) 5084 return CheckConditionalOperator(S, cast<ConditionalOperator>(E), CC, T); 5085 5086 AnalyzeImplicitConversions(S, E, CC); 5087 if (E->getType() != T) 5088 return CheckImplicitConversion(S, E, T, CC, &ICContext); 5089 return; 5090} 5091 5092void CheckConditionalOperator(Sema &S, ConditionalOperator *E, 5093 SourceLocation CC, QualType T) { 5094 AnalyzeImplicitConversions(S, E->getCond(), CC); 5095 5096 bool Suspicious = false; 5097 CheckConditionalOperand(S, E->getTrueExpr(), T, CC, Suspicious); 5098 CheckConditionalOperand(S, E->getFalseExpr(), T, CC, Suspicious); 5099 5100 // If -Wconversion would have warned about either of the candidates 5101 // for a signedness conversion to the context type... 5102 if (!Suspicious) return; 5103 5104 // ...but it's currently ignored... 5105 if (S.Diags.getDiagnosticLevel(diag::warn_impcast_integer_sign_conditional, 5106 CC)) 5107 return; 5108 5109 // ...then check whether it would have warned about either of the 5110 // candidates for a signedness conversion to the condition type. 5111 if (E->getType() == T) return; 5112 5113 Suspicious = false; 5114 CheckImplicitConversion(S, E->getTrueExpr()->IgnoreParenImpCasts(), 5115 E->getType(), CC, &Suspicious); 5116 if (!Suspicious) 5117 CheckImplicitConversion(S, E->getFalseExpr()->IgnoreParenImpCasts(), 5118 E->getType(), CC, &Suspicious); 5119} 5120 5121/// AnalyzeImplicitConversions - Find and report any interesting 5122/// implicit conversions in the given expression. There are a couple 5123/// of competing diagnostics here, -Wconversion and -Wsign-compare. 5124void AnalyzeImplicitConversions(Sema &S, Expr *OrigE, SourceLocation CC) { 5125 QualType T = OrigE->getType(); 5126 Expr *E = OrigE->IgnoreParenImpCasts(); 5127 5128 if (E->isTypeDependent() || E->isValueDependent()) 5129 return; 5130 5131 // For conditional operators, we analyze the arguments as if they 5132 // were being fed directly into the output. 5133 if (isa<ConditionalOperator>(E)) { 5134 ConditionalOperator *CO = cast<ConditionalOperator>(E); 5135 CheckConditionalOperator(S, CO, CC, T); 5136 return; 5137 } 5138 5139 // Check implicit argument conversions for function calls. 5140 if (CallExpr *Call = dyn_cast<CallExpr>(E)) 5141 CheckImplicitArgumentConversions(S, Call, CC); 5142 5143 // Go ahead and check any implicit conversions we might have skipped. 5144 // The non-canonical typecheck is just an optimization; 5145 // CheckImplicitConversion will filter out dead implicit conversions. 5146 if (E->getType() != T) 5147 CheckImplicitConversion(S, E, T, CC); 5148 5149 // Now continue drilling into this expression. 5150 5151 if (PseudoObjectExpr * POE = dyn_cast<PseudoObjectExpr>(E)) { 5152 if (POE->getResultExpr()) 5153 E = POE->getResultExpr(); 5154 } 5155 5156 if (const OpaqueValueExpr *OVE = dyn_cast<OpaqueValueExpr>(E)) 5157 return AnalyzeImplicitConversions(S, OVE->getSourceExpr(), CC); 5158 5159 // Skip past explicit casts. 5160 if (isa<ExplicitCastExpr>(E)) { 5161 E = cast<ExplicitCastExpr>(E)->getSubExpr()->IgnoreParenImpCasts(); 5162 return AnalyzeImplicitConversions(S, E, CC); 5163 } 5164 5165 if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) { 5166 // Do a somewhat different check with comparison operators. 5167 if (BO->isComparisonOp()) 5168 return AnalyzeComparison(S, BO); 5169 5170 // And with simple assignments. 5171 if (BO->getOpcode() == BO_Assign) 5172 return AnalyzeAssignment(S, BO); 5173 } 5174 5175 // These break the otherwise-useful invariant below. Fortunately, 5176 // we don't really need to recurse into them, because any internal 5177 // expressions should have been analyzed already when they were 5178 // built into statements. 5179 if (isa<StmtExpr>(E)) return; 5180 5181 // Don't descend into unevaluated contexts. 5182 if (isa<UnaryExprOrTypeTraitExpr>(E)) return; 5183 5184 // Now just recurse over the expression's children. 5185 CC = E->getExprLoc(); 5186 BinaryOperator *BO = dyn_cast<BinaryOperator>(E); 5187 bool IsLogicalOperator = BO && BO->isLogicalOp(); 5188 for (Stmt::child_range I = E->children(); I; ++I) { 5189 Expr *ChildExpr = dyn_cast_or_null<Expr>(*I); 5190 if (!ChildExpr) 5191 continue; 5192 5193 if (IsLogicalOperator && 5194 isa<StringLiteral>(ChildExpr->IgnoreParenImpCasts())) 5195 // Ignore checking string literals that are in logical operators. 5196 continue; 5197 AnalyzeImplicitConversions(S, ChildExpr, CC); 5198 } 5199} 5200 5201} // end anonymous namespace 5202 5203/// Diagnoses "dangerous" implicit conversions within the given 5204/// expression (which is a full expression). Implements -Wconversion 5205/// and -Wsign-compare. 5206/// 5207/// \param CC the "context" location of the implicit conversion, i.e. 5208/// the most location of the syntactic entity requiring the implicit 5209/// conversion 5210void Sema::CheckImplicitConversions(Expr *E, SourceLocation CC) { 5211 // Don't diagnose in unevaluated contexts. 5212 if (isUnevaluatedContext()) 5213 return; 5214 5215 // Don't diagnose for value- or type-dependent expressions. 5216 if (E->isTypeDependent() || E->isValueDependent()) 5217 return; 5218 5219 // Check for array bounds violations in cases where the check isn't triggered 5220 // elsewhere for other Expr types (like BinaryOperators), e.g. when an 5221 // ArraySubscriptExpr is on the RHS of a variable initialization. 5222 CheckArrayAccess(E); 5223 5224 // This is not the right CC for (e.g.) a variable initialization. 5225 AnalyzeImplicitConversions(*this, E, CC); 5226} 5227 5228/// Diagnose when expression is an integer constant expression and its evaluation 5229/// results in integer overflow 5230void Sema::CheckForIntOverflow (Expr *E) { 5231 if (isa<BinaryOperator>(E->IgnoreParens())) { 5232 llvm::SmallVector<PartialDiagnosticAt, 4> Diags; 5233 E->EvaluateForOverflow(Context, &Diags); 5234 } 5235} 5236 5237namespace { 5238/// \brief Visitor for expressions which looks for unsequenced operations on the 5239/// same object. 5240class SequenceChecker : public EvaluatedExprVisitor<SequenceChecker> { 5241 /// \brief A tree of sequenced regions within an expression. Two regions are 5242 /// unsequenced if one is an ancestor or a descendent of the other. When we 5243 /// finish processing an expression with sequencing, such as a comma 5244 /// expression, we fold its tree nodes into its parent, since they are 5245 /// unsequenced with respect to nodes we will visit later. 5246 class SequenceTree { 5247 struct Value { 5248 explicit Value(unsigned Parent) : Parent(Parent), Merged(false) {} 5249 unsigned Parent : 31; 5250 bool Merged : 1; 5251 }; 5252 llvm::SmallVector<Value, 8> Values; 5253 5254 public: 5255 /// \brief A region within an expression which may be sequenced with respect 5256 /// to some other region. 5257 class Seq { 5258 explicit Seq(unsigned N) : Index(N) {} 5259 unsigned Index; 5260 friend class SequenceTree; 5261 public: 5262 Seq() : Index(0) {} 5263 }; 5264 5265 SequenceTree() { Values.push_back(Value(0)); } 5266 Seq root() const { return Seq(0); } 5267 5268 /// \brief Create a new sequence of operations, which is an unsequenced 5269 /// subset of \p Parent. This sequence of operations is sequenced with 5270 /// respect to other children of \p Parent. 5271 Seq allocate(Seq Parent) { 5272 Values.push_back(Value(Parent.Index)); 5273 return Seq(Values.size() - 1); 5274 } 5275 5276 /// \brief Merge a sequence of operations into its parent. 5277 void merge(Seq S) { 5278 Values[S.Index].Merged = true; 5279 } 5280 5281 /// \brief Determine whether two operations are unsequenced. This operation 5282 /// is asymmetric: \p Cur should be the more recent sequence, and \p Old 5283 /// should have been merged into its parent as appropriate. 5284 bool isUnsequenced(Seq Cur, Seq Old) { 5285 unsigned C = representative(Cur.Index); 5286 unsigned Target = representative(Old.Index); 5287 while (C >= Target) { 5288 if (C == Target) 5289 return true; 5290 C = Values[C].Parent; 5291 } 5292 return false; 5293 } 5294 5295 private: 5296 /// \brief Pick a representative for a sequence. 5297 unsigned representative(unsigned K) { 5298 if (Values[K].Merged) 5299 // Perform path compression as we go. 5300 return Values[K].Parent = representative(Values[K].Parent); 5301 return K; 5302 } 5303 }; 5304 5305 /// An object for which we can track unsequenced uses. 5306 typedef NamedDecl *Object; 5307 5308 /// Different flavors of object usage which we track. We only track the 5309 /// least-sequenced usage of each kind. 5310 enum UsageKind { 5311 /// A read of an object. Multiple unsequenced reads are OK. 5312 UK_Use, 5313 /// A modification of an object which is sequenced before the value 5314 /// computation of the expression, such as ++n. 5315 UK_ModAsValue, 5316 /// A modification of an object which is not sequenced before the value 5317 /// computation of the expression, such as n++. 5318 UK_ModAsSideEffect, 5319 5320 UK_Count = UK_ModAsSideEffect + 1 5321 }; 5322 5323 struct Usage { 5324 Usage() : Use(0), Seq() {} 5325 Expr *Use; 5326 SequenceTree::Seq Seq; 5327 }; 5328 5329 struct UsageInfo { 5330 UsageInfo() : Diagnosed(false) {} 5331 Usage Uses[UK_Count]; 5332 /// Have we issued a diagnostic for this variable already? 5333 bool Diagnosed; 5334 }; 5335 typedef llvm::SmallDenseMap<Object, UsageInfo, 16> UsageInfoMap; 5336 5337 Sema &SemaRef; 5338 /// Sequenced regions within the expression. 5339 SequenceTree Tree; 5340 /// Declaration modifications and references which we have seen. 5341 UsageInfoMap UsageMap; 5342 /// The region we are currently within. 5343 SequenceTree::Seq Region; 5344 /// Filled in with declarations which were modified as a side-effect 5345 /// (that is, post-increment operations). 5346 llvm::SmallVectorImpl<std::pair<Object, Usage> > *ModAsSideEffect; 5347 /// Expressions to check later. We defer checking these to reduce 5348 /// stack usage. 5349 llvm::SmallVectorImpl<Expr*> &WorkList; 5350 5351 /// RAII object wrapping the visitation of a sequenced subexpression of an 5352 /// expression. At the end of this process, the side-effects of the evaluation 5353 /// become sequenced with respect to the value computation of the result, so 5354 /// we downgrade any UK_ModAsSideEffect within the evaluation to 5355 /// UK_ModAsValue. 5356 struct SequencedSubexpression { 5357 SequencedSubexpression(SequenceChecker &Self) 5358 : Self(Self), OldModAsSideEffect(Self.ModAsSideEffect) { 5359 Self.ModAsSideEffect = &ModAsSideEffect; 5360 } 5361 ~SequencedSubexpression() { 5362 for (unsigned I = 0, E = ModAsSideEffect.size(); I != E; ++I) { 5363 UsageInfo &U = Self.UsageMap[ModAsSideEffect[I].first]; 5364 U.Uses[UK_ModAsSideEffect] = ModAsSideEffect[I].second; 5365 Self.addUsage(U, ModAsSideEffect[I].first, 5366 ModAsSideEffect[I].second.Use, UK_ModAsValue); 5367 } 5368 Self.ModAsSideEffect = OldModAsSideEffect; 5369 } 5370 5371 SequenceChecker &Self; 5372 llvm::SmallVector<std::pair<Object, Usage>, 4> ModAsSideEffect; 5373 llvm::SmallVectorImpl<std::pair<Object, Usage> > *OldModAsSideEffect; 5374 }; 5375 5376 /// \brief Find the object which is produced by the specified expression, 5377 /// if any. 5378 Object getObject(Expr *E, bool Mod) const { 5379 E = E->IgnoreParenCasts(); 5380 if (UnaryOperator *UO = dyn_cast<UnaryOperator>(E)) { 5381 if (Mod && (UO->getOpcode() == UO_PreInc || UO->getOpcode() == UO_PreDec)) 5382 return getObject(UO->getSubExpr(), Mod); 5383 } else if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) { 5384 if (BO->getOpcode() == BO_Comma) 5385 return getObject(BO->getRHS(), Mod); 5386 if (Mod && BO->isAssignmentOp()) 5387 return getObject(BO->getLHS(), Mod); 5388 } else if (MemberExpr *ME = dyn_cast<MemberExpr>(E)) { 5389 // FIXME: Check for more interesting cases, like "x.n = ++x.n". 5390 if (isa<CXXThisExpr>(ME->getBase()->IgnoreParenCasts())) 5391 return ME->getMemberDecl(); 5392 } else if (DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(E)) 5393 // FIXME: If this is a reference, map through to its value. 5394 return DRE->getDecl(); 5395 return 0; 5396 } 5397 5398 /// \brief Note that an object was modified or used by an expression. 5399 void addUsage(UsageInfo &UI, Object O, Expr *Ref, UsageKind UK) { 5400 Usage &U = UI.Uses[UK]; 5401 if (!U.Use || !Tree.isUnsequenced(Region, U.Seq)) { 5402 if (UK == UK_ModAsSideEffect && ModAsSideEffect) 5403 ModAsSideEffect->push_back(std::make_pair(O, U)); 5404 U.Use = Ref; 5405 U.Seq = Region; 5406 } 5407 } 5408 /// \brief Check whether a modification or use conflicts with a prior usage. 5409 void checkUsage(Object O, UsageInfo &UI, Expr *Ref, UsageKind OtherKind, 5410 bool IsModMod) { 5411 if (UI.Diagnosed) 5412 return; 5413 5414 const Usage &U = UI.Uses[OtherKind]; 5415 if (!U.Use || !Tree.isUnsequenced(Region, U.Seq)) 5416 return; 5417 5418 Expr *Mod = U.Use; 5419 Expr *ModOrUse = Ref; 5420 if (OtherKind == UK_Use) 5421 std::swap(Mod, ModOrUse); 5422 5423 SemaRef.Diag(Mod->getExprLoc(), 5424 IsModMod ? diag::warn_unsequenced_mod_mod 5425 : diag::warn_unsequenced_mod_use) 5426 << O << SourceRange(ModOrUse->getExprLoc()); 5427 UI.Diagnosed = true; 5428 } 5429 5430 void notePreUse(Object O, Expr *Use) { 5431 UsageInfo &U = UsageMap[O]; 5432 // Uses conflict with other modifications. 5433 checkUsage(O, U, Use, UK_ModAsValue, false); 5434 } 5435 void notePostUse(Object O, Expr *Use) { 5436 UsageInfo &U = UsageMap[O]; 5437 checkUsage(O, U, Use, UK_ModAsSideEffect, false); 5438 addUsage(U, O, Use, UK_Use); 5439 } 5440 5441 void notePreMod(Object O, Expr *Mod) { 5442 UsageInfo &U = UsageMap[O]; 5443 // Modifications conflict with other modifications and with uses. 5444 checkUsage(O, U, Mod, UK_ModAsValue, true); 5445 checkUsage(O, U, Mod, UK_Use, false); 5446 } 5447 void notePostMod(Object O, Expr *Use, UsageKind UK) { 5448 UsageInfo &U = UsageMap[O]; 5449 checkUsage(O, U, Use, UK_ModAsSideEffect, true); 5450 addUsage(U, O, Use, UK); 5451 } 5452 5453public: 5454 SequenceChecker(Sema &S, Expr *E, 5455 llvm::SmallVectorImpl<Expr*> &WorkList) 5456 : EvaluatedExprVisitor<SequenceChecker>(S.Context), SemaRef(S), 5457 Region(Tree.root()), ModAsSideEffect(0), WorkList(WorkList) { 5458 Visit(E); 5459 } 5460 5461 void VisitStmt(Stmt *S) { 5462 // Skip all statements which aren't expressions for now. 5463 } 5464 5465 void VisitExpr(Expr *E) { 5466 // By default, just recurse to evaluated subexpressions. 5467 EvaluatedExprVisitor<SequenceChecker>::VisitStmt(E); 5468 } 5469 5470 void VisitCastExpr(CastExpr *E) { 5471 Object O = Object(); 5472 if (E->getCastKind() == CK_LValueToRValue) 5473 O = getObject(E->getSubExpr(), false); 5474 5475 if (O) 5476 notePreUse(O, E); 5477 VisitExpr(E); 5478 if (O) 5479 notePostUse(O, E); 5480 } 5481 5482 void VisitBinComma(BinaryOperator *BO) { 5483 // C++11 [expr.comma]p1: 5484 // Every value computation and side effect associated with the left 5485 // expression is sequenced before every value computation and side 5486 // effect associated with the right expression. 5487 SequenceTree::Seq LHS = Tree.allocate(Region); 5488 SequenceTree::Seq RHS = Tree.allocate(Region); 5489 SequenceTree::Seq OldRegion = Region; 5490 5491 { 5492 SequencedSubexpression SeqLHS(*this); 5493 Region = LHS; 5494 Visit(BO->getLHS()); 5495 } 5496 5497 Region = RHS; 5498 Visit(BO->getRHS()); 5499 5500 Region = OldRegion; 5501 5502 // Forget that LHS and RHS are sequenced. They are both unsequenced 5503 // with respect to other stuff. 5504 Tree.merge(LHS); 5505 Tree.merge(RHS); 5506 } 5507 5508 void VisitBinAssign(BinaryOperator *BO) { 5509 // The modification is sequenced after the value computation of the LHS 5510 // and RHS, so check it before inspecting the operands and update the 5511 // map afterwards. 5512 Object O = getObject(BO->getLHS(), true); 5513 if (!O) 5514 return VisitExpr(BO); 5515 5516 notePreMod(O, BO); 5517 5518 // C++11 [expr.ass]p7: 5519 // E1 op= E2 is equivalent to E1 = E1 op E2, except that E1 is evaluated 5520 // only once. 5521 // 5522 // Therefore, for a compound assignment operator, O is considered used 5523 // everywhere except within the evaluation of E1 itself. 5524 if (isa<CompoundAssignOperator>(BO)) 5525 notePreUse(O, BO); 5526 5527 Visit(BO->getLHS()); 5528 5529 if (isa<CompoundAssignOperator>(BO)) 5530 notePostUse(O, BO); 5531 5532 Visit(BO->getRHS()); 5533 5534 notePostMod(O, BO, UK_ModAsValue); 5535 } 5536 void VisitCompoundAssignOperator(CompoundAssignOperator *CAO) { 5537 VisitBinAssign(CAO); 5538 } 5539 5540 void VisitUnaryPreInc(UnaryOperator *UO) { VisitUnaryPreIncDec(UO); } 5541 void VisitUnaryPreDec(UnaryOperator *UO) { VisitUnaryPreIncDec(UO); } 5542 void VisitUnaryPreIncDec(UnaryOperator *UO) { 5543 Object O = getObject(UO->getSubExpr(), true); 5544 if (!O) 5545 return VisitExpr(UO); 5546 5547 notePreMod(O, UO); 5548 Visit(UO->getSubExpr()); 5549 notePostMod(O, UO, UK_ModAsValue); 5550 } 5551 5552 void VisitUnaryPostInc(UnaryOperator *UO) { VisitUnaryPostIncDec(UO); } 5553 void VisitUnaryPostDec(UnaryOperator *UO) { VisitUnaryPostIncDec(UO); } 5554 void VisitUnaryPostIncDec(UnaryOperator *UO) { 5555 Object O = getObject(UO->getSubExpr(), true); 5556 if (!O) 5557 return VisitExpr(UO); 5558 5559 notePreMod(O, UO); 5560 Visit(UO->getSubExpr()); 5561 notePostMod(O, UO, UK_ModAsSideEffect); 5562 } 5563 5564 /// Don't visit the RHS of '&&' or '||' if it might not be evaluated. 5565 void VisitBinLOr(BinaryOperator *BO) { 5566 // The side-effects of the LHS of an '&&' are sequenced before the 5567 // value computation of the RHS, and hence before the value computation 5568 // of the '&&' itself, unless the LHS evaluates to zero. We treat them 5569 // as if they were unconditionally sequenced. 5570 { 5571 SequencedSubexpression Sequenced(*this); 5572 Visit(BO->getLHS()); 5573 } 5574 5575 bool Result; 5576 if (!BO->getLHS()->isValueDependent() && 5577 BO->getLHS()->EvaluateAsBooleanCondition(Result, SemaRef.Context)) { 5578 if (!Result) 5579 Visit(BO->getRHS()); 5580 } else { 5581 // Check for unsequenced operations in the RHS, treating it as an 5582 // entirely separate evaluation. 5583 // 5584 // FIXME: If there are operations in the RHS which are unsequenced 5585 // with respect to operations outside the RHS, and those operations 5586 // are unconditionally evaluated, diagnose them. 5587 WorkList.push_back(BO->getRHS()); 5588 } 5589 } 5590 void VisitBinLAnd(BinaryOperator *BO) { 5591 { 5592 SequencedSubexpression Sequenced(*this); 5593 Visit(BO->getLHS()); 5594 } 5595 5596 bool Result; 5597 if (!BO->getLHS()->isValueDependent() && 5598 BO->getLHS()->EvaluateAsBooleanCondition(Result, SemaRef.Context)) { 5599 if (Result) 5600 Visit(BO->getRHS()); 5601 } else { 5602 WorkList.push_back(BO->getRHS()); 5603 } 5604 } 5605 5606 // Only visit the condition, unless we can be sure which subexpression will 5607 // be chosen. 5608 void VisitAbstractConditionalOperator(AbstractConditionalOperator *CO) { 5609 SequencedSubexpression Sequenced(*this); 5610 Visit(CO->getCond()); 5611 5612 bool Result; 5613 if (!CO->getCond()->isValueDependent() && 5614 CO->getCond()->EvaluateAsBooleanCondition(Result, SemaRef.Context)) 5615 Visit(Result ? CO->getTrueExpr() : CO->getFalseExpr()); 5616 else { 5617 WorkList.push_back(CO->getTrueExpr()); 5618 WorkList.push_back(CO->getFalseExpr()); 5619 } 5620 } 5621 5622 void VisitCXXConstructExpr(CXXConstructExpr *CCE) { 5623 if (!CCE->isListInitialization()) 5624 return VisitExpr(CCE); 5625 5626 // In C++11, list initializations are sequenced. 5627 llvm::SmallVector<SequenceTree::Seq, 32> Elts; 5628 SequenceTree::Seq Parent = Region; 5629 for (CXXConstructExpr::arg_iterator I = CCE->arg_begin(), 5630 E = CCE->arg_end(); 5631 I != E; ++I) { 5632 Region = Tree.allocate(Parent); 5633 Elts.push_back(Region); 5634 Visit(*I); 5635 } 5636 5637 // Forget that the initializers are sequenced. 5638 Region = Parent; 5639 for (unsigned I = 0; I < Elts.size(); ++I) 5640 Tree.merge(Elts[I]); 5641 } 5642 5643 void VisitInitListExpr(InitListExpr *ILE) { 5644 if (!SemaRef.getLangOpts().CPlusPlus11) 5645 return VisitExpr(ILE); 5646 5647 // In C++11, list initializations are sequenced. 5648 llvm::SmallVector<SequenceTree::Seq, 32> Elts; 5649 SequenceTree::Seq Parent = Region; 5650 for (unsigned I = 0; I < ILE->getNumInits(); ++I) { 5651 Expr *E = ILE->getInit(I); 5652 if (!E) continue; 5653 Region = Tree.allocate(Parent); 5654 Elts.push_back(Region); 5655 Visit(E); 5656 } 5657 5658 // Forget that the initializers are sequenced. 5659 Region = Parent; 5660 for (unsigned I = 0; I < Elts.size(); ++I) 5661 Tree.merge(Elts[I]); 5662 } 5663}; 5664} 5665 5666void Sema::CheckUnsequencedOperations(Expr *E) { 5667 llvm::SmallVector<Expr*, 8> WorkList; 5668 WorkList.push_back(E); 5669 while (!WorkList.empty()) { 5670 Expr *Item = WorkList.back(); 5671 WorkList.pop_back(); 5672 SequenceChecker(*this, Item, WorkList); 5673 } 5674} 5675 5676void Sema::CheckCompletedExpr(Expr *E, SourceLocation CheckLoc, 5677 bool IsConstexpr) { 5678 CheckImplicitConversions(E, CheckLoc); 5679 CheckUnsequencedOperations(E); 5680 if (!IsConstexpr && !E->isValueDependent()) 5681 CheckForIntOverflow(E); 5682} 5683 5684void Sema::CheckBitFieldInitialization(SourceLocation InitLoc, 5685 FieldDecl *BitField, 5686 Expr *Init) { 5687 (void) AnalyzeBitFieldAssignment(*this, BitField, Init, InitLoc); 5688} 5689 5690/// CheckParmsForFunctionDef - Check that the parameters of the given 5691/// function are appropriate for the definition of a function. This 5692/// takes care of any checks that cannot be performed on the 5693/// declaration itself, e.g., that the types of each of the function 5694/// parameters are complete. 5695bool Sema::CheckParmsForFunctionDef(ParmVarDecl **P, ParmVarDecl **PEnd, 5696 bool CheckParameterNames) { 5697 bool HasInvalidParm = false; 5698 for (; P != PEnd; ++P) { 5699 ParmVarDecl *Param = *P; 5700 5701 // C99 6.7.5.3p4: the parameters in a parameter type list in a 5702 // function declarator that is part of a function definition of 5703 // that function shall not have incomplete type. 5704 // 5705 // This is also C++ [dcl.fct]p6. 5706 if (!Param->isInvalidDecl() && 5707 RequireCompleteType(Param->getLocation(), Param->getType(), 5708 diag::err_typecheck_decl_incomplete_type)) { 5709 Param->setInvalidDecl(); 5710 HasInvalidParm = true; 5711 } 5712 5713 // C99 6.9.1p5: If the declarator includes a parameter type list, the 5714 // declaration of each parameter shall include an identifier. 5715 if (CheckParameterNames && 5716 Param->getIdentifier() == 0 && 5717 !Param->isImplicit() && 5718 !getLangOpts().CPlusPlus) 5719 Diag(Param->getLocation(), diag::err_parameter_name_omitted); 5720 5721 // C99 6.7.5.3p12: 5722 // If the function declarator is not part of a definition of that 5723 // function, parameters may have incomplete type and may use the [*] 5724 // notation in their sequences of declarator specifiers to specify 5725 // variable length array types. 5726 QualType PType = Param->getOriginalType(); 5727 while (const ArrayType *AT = Context.getAsArrayType(PType)) { 5728 if (AT->getSizeModifier() == ArrayType::Star) { 5729 // FIXME: This diagnostic should point the '[*]' if source-location 5730 // information is added for it. 5731 Diag(Param->getLocation(), diag::err_array_star_in_function_definition); 5732 break; 5733 } 5734 PType= AT->getElementType(); 5735 } 5736 } 5737 5738 return HasInvalidParm; 5739} 5740 5741/// CheckCastAlign - Implements -Wcast-align, which warns when a 5742/// pointer cast increases the alignment requirements. 5743void Sema::CheckCastAlign(Expr *Op, QualType T, SourceRange TRange) { 5744 // This is actually a lot of work to potentially be doing on every 5745 // cast; don't do it if we're ignoring -Wcast_align (as is the default). 5746 if (getDiagnostics().getDiagnosticLevel(diag::warn_cast_align, 5747 TRange.getBegin()) 5748 == DiagnosticsEngine::Ignored) 5749 return; 5750 5751 // Ignore dependent types. 5752 if (T->isDependentType() || Op->getType()->isDependentType()) 5753 return; 5754 5755 // Require that the destination be a pointer type. 5756 const PointerType *DestPtr = T->getAs<PointerType>(); 5757 if (!DestPtr) return; 5758 5759 // If the destination has alignment 1, we're done. 5760 QualType DestPointee = DestPtr->getPointeeType(); 5761 if (DestPointee->isIncompleteType()) return; 5762 CharUnits DestAlign = Context.getTypeAlignInChars(DestPointee); 5763 if (DestAlign.isOne()) return; 5764 5765 // Require that the source be a pointer type. 5766 const PointerType *SrcPtr = Op->getType()->getAs<PointerType>(); 5767 if (!SrcPtr) return; 5768 QualType SrcPointee = SrcPtr->getPointeeType(); 5769 5770 // Whitelist casts from cv void*. We already implicitly 5771 // whitelisted casts to cv void*, since they have alignment 1. 5772 // Also whitelist casts involving incomplete types, which implicitly 5773 // includes 'void'. 5774 if (SrcPointee->isIncompleteType()) return; 5775 5776 CharUnits SrcAlign = Context.getTypeAlignInChars(SrcPointee); 5777 if (SrcAlign >= DestAlign) return; 5778 5779 Diag(TRange.getBegin(), diag::warn_cast_align) 5780 << Op->getType() << T 5781 << static_cast<unsigned>(SrcAlign.getQuantity()) 5782 << static_cast<unsigned>(DestAlign.getQuantity()) 5783 << TRange << Op->getSourceRange(); 5784} 5785 5786static const Type* getElementType(const Expr *BaseExpr) { 5787 const Type* EltType = BaseExpr->getType().getTypePtr(); 5788 if (EltType->isAnyPointerType()) 5789 return EltType->getPointeeType().getTypePtr(); 5790 else if (EltType->isArrayType()) 5791 return EltType->getBaseElementTypeUnsafe(); 5792 return EltType; 5793} 5794 5795/// \brief Check whether this array fits the idiom of a size-one tail padded 5796/// array member of a struct. 5797/// 5798/// We avoid emitting out-of-bounds access warnings for such arrays as they are 5799/// commonly used to emulate flexible arrays in C89 code. 5800static bool IsTailPaddedMemberArray(Sema &S, llvm::APInt Size, 5801 const NamedDecl *ND) { 5802 if (Size != 1 || !ND) return false; 5803 5804 const FieldDecl *FD = dyn_cast<FieldDecl>(ND); 5805 if (!FD) return false; 5806 5807 // Don't consider sizes resulting from macro expansions or template argument 5808 // substitution to form C89 tail-padded arrays. 5809 5810 TypeSourceInfo *TInfo = FD->getTypeSourceInfo(); 5811 while (TInfo) { 5812 TypeLoc TL = TInfo->getTypeLoc(); 5813 // Look through typedefs. 5814 if (TypedefTypeLoc TTL = TL.getAs<TypedefTypeLoc>()) { 5815 const TypedefNameDecl *TDL = TTL.getTypedefNameDecl(); 5816 TInfo = TDL->getTypeSourceInfo(); 5817 continue; 5818 } 5819 if (ConstantArrayTypeLoc CTL = TL.getAs<ConstantArrayTypeLoc>()) { 5820 const Expr *SizeExpr = dyn_cast<IntegerLiteral>(CTL.getSizeExpr()); 5821 if (!SizeExpr || SizeExpr->getExprLoc().isMacroID()) 5822 return false; 5823 } 5824 break; 5825 } 5826 5827 const RecordDecl *RD = dyn_cast<RecordDecl>(FD->getDeclContext()); 5828 if (!RD) return false; 5829 if (RD->isUnion()) return false; 5830 if (const CXXRecordDecl *CRD = dyn_cast<CXXRecordDecl>(RD)) { 5831 if (!CRD->isStandardLayout()) return false; 5832 } 5833 5834 // See if this is the last field decl in the record. 5835 const Decl *D = FD; 5836 while ((D = D->getNextDeclInContext())) 5837 if (isa<FieldDecl>(D)) 5838 return false; 5839 return true; 5840} 5841 5842void Sema::CheckArrayAccess(const Expr *BaseExpr, const Expr *IndexExpr, 5843 const ArraySubscriptExpr *ASE, 5844 bool AllowOnePastEnd, bool IndexNegated) { 5845 IndexExpr = IndexExpr->IgnoreParenImpCasts(); 5846 if (IndexExpr->isValueDependent()) 5847 return; 5848 5849 const Type *EffectiveType = getElementType(BaseExpr); 5850 BaseExpr = BaseExpr->IgnoreParenCasts(); 5851 const ConstantArrayType *ArrayTy = 5852 Context.getAsConstantArrayType(BaseExpr->getType()); 5853 if (!ArrayTy) 5854 return; 5855 5856 llvm::APSInt index; 5857 if (!IndexExpr->EvaluateAsInt(index, Context)) 5858 return; 5859 if (IndexNegated) 5860 index = -index; 5861 5862 const NamedDecl *ND = NULL; 5863 if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(BaseExpr)) 5864 ND = dyn_cast<NamedDecl>(DRE->getDecl()); 5865 if (const MemberExpr *ME = dyn_cast<MemberExpr>(BaseExpr)) 5866 ND = dyn_cast<NamedDecl>(ME->getMemberDecl()); 5867 5868 if (index.isUnsigned() || !index.isNegative()) { 5869 llvm::APInt size = ArrayTy->getSize(); 5870 if (!size.isStrictlyPositive()) 5871 return; 5872 5873 const Type* BaseType = getElementType(BaseExpr); 5874 if (BaseType != EffectiveType) { 5875 // Make sure we're comparing apples to apples when comparing index to size 5876 uint64_t ptrarith_typesize = Context.getTypeSize(EffectiveType); 5877 uint64_t array_typesize = Context.getTypeSize(BaseType); 5878 // Handle ptrarith_typesize being zero, such as when casting to void* 5879 if (!ptrarith_typesize) ptrarith_typesize = 1; 5880 if (ptrarith_typesize != array_typesize) { 5881 // There's a cast to a different size type involved 5882 uint64_t ratio = array_typesize / ptrarith_typesize; 5883 // TODO: Be smarter about handling cases where array_typesize is not a 5884 // multiple of ptrarith_typesize 5885 if (ptrarith_typesize * ratio == array_typesize) 5886 size *= llvm::APInt(size.getBitWidth(), ratio); 5887 } 5888 } 5889 5890 if (size.getBitWidth() > index.getBitWidth()) 5891 index = index.zext(size.getBitWidth()); 5892 else if (size.getBitWidth() < index.getBitWidth()) 5893 size = size.zext(index.getBitWidth()); 5894 5895 // For array subscripting the index must be less than size, but for pointer 5896 // arithmetic also allow the index (offset) to be equal to size since 5897 // computing the next address after the end of the array is legal and 5898 // commonly done e.g. in C++ iterators and range-based for loops. 5899 if (AllowOnePastEnd ? index.ule(size) : index.ult(size)) 5900 return; 5901 5902 // Also don't warn for arrays of size 1 which are members of some 5903 // structure. These are often used to approximate flexible arrays in C89 5904 // code. 5905 if (IsTailPaddedMemberArray(*this, size, ND)) 5906 return; 5907 5908 // Suppress the warning if the subscript expression (as identified by the 5909 // ']' location) and the index expression are both from macro expansions 5910 // within a system header. 5911 if (ASE) { 5912 SourceLocation RBracketLoc = SourceMgr.getSpellingLoc( 5913 ASE->getRBracketLoc()); 5914 if (SourceMgr.isInSystemHeader(RBracketLoc)) { 5915 SourceLocation IndexLoc = SourceMgr.getSpellingLoc( 5916 IndexExpr->getLocStart()); 5917 if (SourceMgr.isFromSameFile(RBracketLoc, IndexLoc)) 5918 return; 5919 } 5920 } 5921 5922 unsigned DiagID = diag::warn_ptr_arith_exceeds_bounds; 5923 if (ASE) 5924 DiagID = diag::warn_array_index_exceeds_bounds; 5925 5926 DiagRuntimeBehavior(BaseExpr->getLocStart(), BaseExpr, 5927 PDiag(DiagID) << index.toString(10, true) 5928 << size.toString(10, true) 5929 << (unsigned)size.getLimitedValue(~0U) 5930 << IndexExpr->getSourceRange()); 5931 } else { 5932 unsigned DiagID = diag::warn_array_index_precedes_bounds; 5933 if (!ASE) { 5934 DiagID = diag::warn_ptr_arith_precedes_bounds; 5935 if (index.isNegative()) index = -index; 5936 } 5937 5938 DiagRuntimeBehavior(BaseExpr->getLocStart(), BaseExpr, 5939 PDiag(DiagID) << index.toString(10, true) 5940 << IndexExpr->getSourceRange()); 5941 } 5942 5943 if (!ND) { 5944 // Try harder to find a NamedDecl to point at in the note. 5945 while (const ArraySubscriptExpr *ASE = 5946 dyn_cast<ArraySubscriptExpr>(BaseExpr)) 5947 BaseExpr = ASE->getBase()->IgnoreParenCasts(); 5948 if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(BaseExpr)) 5949 ND = dyn_cast<NamedDecl>(DRE->getDecl()); 5950 if (const MemberExpr *ME = dyn_cast<MemberExpr>(BaseExpr)) 5951 ND = dyn_cast<NamedDecl>(ME->getMemberDecl()); 5952 } 5953 5954 if (ND) 5955 DiagRuntimeBehavior(ND->getLocStart(), BaseExpr, 5956 PDiag(diag::note_array_index_out_of_bounds) 5957 << ND->getDeclName()); 5958} 5959 5960void Sema::CheckArrayAccess(const Expr *expr) { 5961 int AllowOnePastEnd = 0; 5962 while (expr) { 5963 expr = expr->IgnoreParenImpCasts(); 5964 switch (expr->getStmtClass()) { 5965 case Stmt::ArraySubscriptExprClass: { 5966 const ArraySubscriptExpr *ASE = cast<ArraySubscriptExpr>(expr); 5967 CheckArrayAccess(ASE->getBase(), ASE->getIdx(), ASE, 5968 AllowOnePastEnd > 0); 5969 return; 5970 } 5971 case Stmt::UnaryOperatorClass: { 5972 // Only unwrap the * and & unary operators 5973 const UnaryOperator *UO = cast<UnaryOperator>(expr); 5974 expr = UO->getSubExpr(); 5975 switch (UO->getOpcode()) { 5976 case UO_AddrOf: 5977 AllowOnePastEnd++; 5978 break; 5979 case UO_Deref: 5980 AllowOnePastEnd--; 5981 break; 5982 default: 5983 return; 5984 } 5985 break; 5986 } 5987 case Stmt::ConditionalOperatorClass: { 5988 const ConditionalOperator *cond = cast<ConditionalOperator>(expr); 5989 if (const Expr *lhs = cond->getLHS()) 5990 CheckArrayAccess(lhs); 5991 if (const Expr *rhs = cond->getRHS()) 5992 CheckArrayAccess(rhs); 5993 return; 5994 } 5995 default: 5996 return; 5997 } 5998 } 5999} 6000 6001//===--- CHECK: Objective-C retain cycles ----------------------------------// 6002 6003namespace { 6004 struct RetainCycleOwner { 6005 RetainCycleOwner() : Variable(0), Indirect(false) {} 6006 VarDecl *Variable; 6007 SourceRange Range; 6008 SourceLocation Loc; 6009 bool Indirect; 6010 6011 void setLocsFrom(Expr *e) { 6012 Loc = e->getExprLoc(); 6013 Range = e->getSourceRange(); 6014 } 6015 }; 6016} 6017 6018/// Consider whether capturing the given variable can possibly lead to 6019/// a retain cycle. 6020static bool considerVariable(VarDecl *var, Expr *ref, RetainCycleOwner &owner) { 6021 // In ARC, it's captured strongly iff the variable has __strong 6022 // lifetime. In MRR, it's captured strongly if the variable is 6023 // __block and has an appropriate type. 6024 if (var->getType().getObjCLifetime() != Qualifiers::OCL_Strong) 6025 return false; 6026 6027 owner.Variable = var; 6028 if (ref) 6029 owner.setLocsFrom(ref); 6030 return true; 6031} 6032 6033static bool findRetainCycleOwner(Sema &S, Expr *e, RetainCycleOwner &owner) { 6034 while (true) { 6035 e = e->IgnoreParens(); 6036 if (CastExpr *cast = dyn_cast<CastExpr>(e)) { 6037 switch (cast->getCastKind()) { 6038 case CK_BitCast: 6039 case CK_LValueBitCast: 6040 case CK_LValueToRValue: 6041 case CK_ARCReclaimReturnedObject: 6042 e = cast->getSubExpr(); 6043 continue; 6044 6045 default: 6046 return false; 6047 } 6048 } 6049 6050 if (ObjCIvarRefExpr *ref = dyn_cast<ObjCIvarRefExpr>(e)) { 6051 ObjCIvarDecl *ivar = ref->getDecl(); 6052 if (ivar->getType().getObjCLifetime() != Qualifiers::OCL_Strong) 6053 return false; 6054 6055 // Try to find a retain cycle in the base. 6056 if (!findRetainCycleOwner(S, ref->getBase(), owner)) 6057 return false; 6058 6059 if (ref->isFreeIvar()) owner.setLocsFrom(ref); 6060 owner.Indirect = true; 6061 return true; 6062 } 6063 6064 if (DeclRefExpr *ref = dyn_cast<DeclRefExpr>(e)) { 6065 VarDecl *var = dyn_cast<VarDecl>(ref->getDecl()); 6066 if (!var) return false; 6067 return considerVariable(var, ref, owner); 6068 } 6069 6070 if (MemberExpr *member = dyn_cast<MemberExpr>(e)) { 6071 if (member->isArrow()) return false; 6072 6073 // Don't count this as an indirect ownership. 6074 e = member->getBase(); 6075 continue; 6076 } 6077 6078 if (PseudoObjectExpr *pseudo = dyn_cast<PseudoObjectExpr>(e)) { 6079 // Only pay attention to pseudo-objects on property references. 6080 ObjCPropertyRefExpr *pre 6081 = dyn_cast<ObjCPropertyRefExpr>(pseudo->getSyntacticForm() 6082 ->IgnoreParens()); 6083 if (!pre) return false; 6084 if (pre->isImplicitProperty()) return false; 6085 ObjCPropertyDecl *property = pre->getExplicitProperty(); 6086 if (!property->isRetaining() && 6087 !(property->getPropertyIvarDecl() && 6088 property->getPropertyIvarDecl()->getType() 6089 .getObjCLifetime() == Qualifiers::OCL_Strong)) 6090 return false; 6091 6092 owner.Indirect = true; 6093 if (pre->isSuperReceiver()) { 6094 owner.Variable = S.getCurMethodDecl()->getSelfDecl(); 6095 if (!owner.Variable) 6096 return false; 6097 owner.Loc = pre->getLocation(); 6098 owner.Range = pre->getSourceRange(); 6099 return true; 6100 } 6101 e = const_cast<Expr*>(cast<OpaqueValueExpr>(pre->getBase()) 6102 ->getSourceExpr()); 6103 continue; 6104 } 6105 6106 // Array ivars? 6107 6108 return false; 6109 } 6110} 6111 6112namespace { 6113 struct FindCaptureVisitor : EvaluatedExprVisitor<FindCaptureVisitor> { 6114 FindCaptureVisitor(ASTContext &Context, VarDecl *variable) 6115 : EvaluatedExprVisitor<FindCaptureVisitor>(Context), 6116 Variable(variable), Capturer(0) {} 6117 6118 VarDecl *Variable; 6119 Expr *Capturer; 6120 6121 void VisitDeclRefExpr(DeclRefExpr *ref) { 6122 if (ref->getDecl() == Variable && !Capturer) 6123 Capturer = ref; 6124 } 6125 6126 void VisitObjCIvarRefExpr(ObjCIvarRefExpr *ref) { 6127 if (Capturer) return; 6128 Visit(ref->getBase()); 6129 if (Capturer && ref->isFreeIvar()) 6130 Capturer = ref; 6131 } 6132 6133 void VisitBlockExpr(BlockExpr *block) { 6134 // Look inside nested blocks 6135 if (block->getBlockDecl()->capturesVariable(Variable)) 6136 Visit(block->getBlockDecl()->getBody()); 6137 } 6138 6139 void VisitOpaqueValueExpr(OpaqueValueExpr *OVE) { 6140 if (Capturer) return; 6141 if (OVE->getSourceExpr()) 6142 Visit(OVE->getSourceExpr()); 6143 } 6144 }; 6145} 6146 6147/// Check whether the given argument is a block which captures a 6148/// variable. 6149static Expr *findCapturingExpr(Sema &S, Expr *e, RetainCycleOwner &owner) { 6150 assert(owner.Variable && owner.Loc.isValid()); 6151 6152 e = e->IgnoreParenCasts(); 6153 6154 // Look through [^{...} copy] and Block_copy(^{...}). 6155 if (ObjCMessageExpr *ME = dyn_cast<ObjCMessageExpr>(e)) { 6156 Selector Cmd = ME->getSelector(); 6157 if (Cmd.isUnarySelector() && Cmd.getNameForSlot(0) == "copy") { 6158 e = ME->getInstanceReceiver(); 6159 if (!e) 6160 return 0; 6161 e = e->IgnoreParenCasts(); 6162 } 6163 } else if (CallExpr *CE = dyn_cast<CallExpr>(e)) { 6164 if (CE->getNumArgs() == 1) { 6165 FunctionDecl *Fn = dyn_cast_or_null<FunctionDecl>(CE->getCalleeDecl()); 6166 if (Fn) { 6167 const IdentifierInfo *FnI = Fn->getIdentifier(); 6168 if (FnI && FnI->isStr("_Block_copy")) { 6169 e = CE->getArg(0)->IgnoreParenCasts(); 6170 } 6171 } 6172 } 6173 } 6174 6175 BlockExpr *block = dyn_cast<BlockExpr>(e); 6176 if (!block || !block->getBlockDecl()->capturesVariable(owner.Variable)) 6177 return 0; 6178 6179 FindCaptureVisitor visitor(S.Context, owner.Variable); 6180 visitor.Visit(block->getBlockDecl()->getBody()); 6181 return visitor.Capturer; 6182} 6183 6184static void diagnoseRetainCycle(Sema &S, Expr *capturer, 6185 RetainCycleOwner &owner) { 6186 assert(capturer); 6187 assert(owner.Variable && owner.Loc.isValid()); 6188 6189 S.Diag(capturer->getExprLoc(), diag::warn_arc_retain_cycle) 6190 << owner.Variable << capturer->getSourceRange(); 6191 S.Diag(owner.Loc, diag::note_arc_retain_cycle_owner) 6192 << owner.Indirect << owner.Range; 6193} 6194 6195/// Check for a keyword selector that starts with the word 'add' or 6196/// 'set'. 6197static bool isSetterLikeSelector(Selector sel) { 6198 if (sel.isUnarySelector()) return false; 6199 6200 StringRef str = sel.getNameForSlot(0); 6201 while (!str.empty() && str.front() == '_') str = str.substr(1); 6202 if (str.startswith("set")) 6203 str = str.substr(3); 6204 else if (str.startswith("add")) { 6205 // Specially whitelist 'addOperationWithBlock:'. 6206 if (sel.getNumArgs() == 1 && str.startswith("addOperationWithBlock")) 6207 return false; 6208 str = str.substr(3); 6209 } 6210 else 6211 return false; 6212 6213 if (str.empty()) return true; 6214 return !isLowercase(str.front()); 6215} 6216 6217/// Check a message send to see if it's likely to cause a retain cycle. 6218void Sema::checkRetainCycles(ObjCMessageExpr *msg) { 6219 // Only check instance methods whose selector looks like a setter. 6220 if (!msg->isInstanceMessage() || !isSetterLikeSelector(msg->getSelector())) 6221 return; 6222 6223 // Try to find a variable that the receiver is strongly owned by. 6224 RetainCycleOwner owner; 6225 if (msg->getReceiverKind() == ObjCMessageExpr::Instance) { 6226 if (!findRetainCycleOwner(*this, msg->getInstanceReceiver(), owner)) 6227 return; 6228 } else { 6229 assert(msg->getReceiverKind() == ObjCMessageExpr::SuperInstance); 6230 owner.Variable = getCurMethodDecl()->getSelfDecl(); 6231 owner.Loc = msg->getSuperLoc(); 6232 owner.Range = msg->getSuperLoc(); 6233 } 6234 6235 // Check whether the receiver is captured by any of the arguments. 6236 for (unsigned i = 0, e = msg->getNumArgs(); i != e; ++i) 6237 if (Expr *capturer = findCapturingExpr(*this, msg->getArg(i), owner)) 6238 return diagnoseRetainCycle(*this, capturer, owner); 6239} 6240 6241/// Check a property assign to see if it's likely to cause a retain cycle. 6242void Sema::checkRetainCycles(Expr *receiver, Expr *argument) { 6243 RetainCycleOwner owner; 6244 if (!findRetainCycleOwner(*this, receiver, owner)) 6245 return; 6246 6247 if (Expr *capturer = findCapturingExpr(*this, argument, owner)) 6248 diagnoseRetainCycle(*this, capturer, owner); 6249} 6250 6251void Sema::checkRetainCycles(VarDecl *Var, Expr *Init) { 6252 RetainCycleOwner Owner; 6253 if (!considerVariable(Var, /*DeclRefExpr=*/0, Owner)) 6254 return; 6255 6256 // Because we don't have an expression for the variable, we have to set the 6257 // location explicitly here. 6258 Owner.Loc = Var->getLocation(); 6259 Owner.Range = Var->getSourceRange(); 6260 6261 if (Expr *Capturer = findCapturingExpr(*this, Init, Owner)) 6262 diagnoseRetainCycle(*this, Capturer, Owner); 6263} 6264 6265static bool checkUnsafeAssignLiteral(Sema &S, SourceLocation Loc, 6266 Expr *RHS, bool isProperty) { 6267 // Check if RHS is an Objective-C object literal, which also can get 6268 // immediately zapped in a weak reference. Note that we explicitly 6269 // allow ObjCStringLiterals, since those are designed to never really die. 6270 RHS = RHS->IgnoreParenImpCasts(); 6271 6272 // This enum needs to match with the 'select' in 6273 // warn_objc_arc_literal_assign (off-by-1). 6274 Sema::ObjCLiteralKind Kind = S.CheckLiteralKind(RHS); 6275 if (Kind == Sema::LK_String || Kind == Sema::LK_None) 6276 return false; 6277 6278 S.Diag(Loc, diag::warn_arc_literal_assign) 6279 << (unsigned) Kind 6280 << (isProperty ? 0 : 1) 6281 << RHS->getSourceRange(); 6282 6283 return true; 6284} 6285 6286static bool checkUnsafeAssignObject(Sema &S, SourceLocation Loc, 6287 Qualifiers::ObjCLifetime LT, 6288 Expr *RHS, bool isProperty) { 6289 // Strip off any implicit cast added to get to the one ARC-specific. 6290 while (ImplicitCastExpr *cast = dyn_cast<ImplicitCastExpr>(RHS)) { 6291 if (cast->getCastKind() == CK_ARCConsumeObject) { 6292 S.Diag(Loc, diag::warn_arc_retained_assign) 6293 << (LT == Qualifiers::OCL_ExplicitNone) 6294 << (isProperty ? 0 : 1) 6295 << RHS->getSourceRange(); 6296 return true; 6297 } 6298 RHS = cast->getSubExpr(); 6299 } 6300 6301 if (LT == Qualifiers::OCL_Weak && 6302 checkUnsafeAssignLiteral(S, Loc, RHS, isProperty)) 6303 return true; 6304 6305 return false; 6306} 6307 6308bool Sema::checkUnsafeAssigns(SourceLocation Loc, 6309 QualType LHS, Expr *RHS) { 6310 Qualifiers::ObjCLifetime LT = LHS.getObjCLifetime(); 6311 6312 if (LT != Qualifiers::OCL_Weak && LT != Qualifiers::OCL_ExplicitNone) 6313 return false; 6314 6315 if (checkUnsafeAssignObject(*this, Loc, LT, RHS, false)) 6316 return true; 6317 6318 return false; 6319} 6320 6321void Sema::checkUnsafeExprAssigns(SourceLocation Loc, 6322 Expr *LHS, Expr *RHS) { 6323 QualType LHSType; 6324 // PropertyRef on LHS type need be directly obtained from 6325 // its declaration as it has a PsuedoType. 6326 ObjCPropertyRefExpr *PRE 6327 = dyn_cast<ObjCPropertyRefExpr>(LHS->IgnoreParens()); 6328 if (PRE && !PRE->isImplicitProperty()) { 6329 const ObjCPropertyDecl *PD = PRE->getExplicitProperty(); 6330 if (PD) 6331 LHSType = PD->getType(); 6332 } 6333 6334 if (LHSType.isNull()) 6335 LHSType = LHS->getType(); 6336 6337 Qualifiers::ObjCLifetime LT = LHSType.getObjCLifetime(); 6338 6339 if (LT == Qualifiers::OCL_Weak) { 6340 DiagnosticsEngine::Level Level = 6341 Diags.getDiagnosticLevel(diag::warn_arc_repeated_use_of_weak, Loc); 6342 if (Level != DiagnosticsEngine::Ignored) 6343 getCurFunction()->markSafeWeakUse(LHS); 6344 } 6345 6346 if (checkUnsafeAssigns(Loc, LHSType, RHS)) 6347 return; 6348 6349 // FIXME. Check for other life times. 6350 if (LT != Qualifiers::OCL_None) 6351 return; 6352 6353 if (PRE) { 6354 if (PRE->isImplicitProperty()) 6355 return; 6356 const ObjCPropertyDecl *PD = PRE->getExplicitProperty(); 6357 if (!PD) 6358 return; 6359 6360 unsigned Attributes = PD->getPropertyAttributes(); 6361 if (Attributes & ObjCPropertyDecl::OBJC_PR_assign) { 6362 // when 'assign' attribute was not explicitly specified 6363 // by user, ignore it and rely on property type itself 6364 // for lifetime info. 6365 unsigned AsWrittenAttr = PD->getPropertyAttributesAsWritten(); 6366 if (!(AsWrittenAttr & ObjCPropertyDecl::OBJC_PR_assign) && 6367 LHSType->isObjCRetainableType()) 6368 return; 6369 6370 while (ImplicitCastExpr *cast = dyn_cast<ImplicitCastExpr>(RHS)) { 6371 if (cast->getCastKind() == CK_ARCConsumeObject) { 6372 Diag(Loc, diag::warn_arc_retained_property_assign) 6373 << RHS->getSourceRange(); 6374 return; 6375 } 6376 RHS = cast->getSubExpr(); 6377 } 6378 } 6379 else if (Attributes & ObjCPropertyDecl::OBJC_PR_weak) { 6380 if (checkUnsafeAssignObject(*this, Loc, Qualifiers::OCL_Weak, RHS, true)) 6381 return; 6382 } 6383 } 6384} 6385 6386//===--- CHECK: Empty statement body (-Wempty-body) ---------------------===// 6387 6388namespace { 6389bool ShouldDiagnoseEmptyStmtBody(const SourceManager &SourceMgr, 6390 SourceLocation StmtLoc, 6391 const NullStmt *Body) { 6392 // Do not warn if the body is a macro that expands to nothing, e.g: 6393 // 6394 // #define CALL(x) 6395 // if (condition) 6396 // CALL(0); 6397 // 6398 if (Body->hasLeadingEmptyMacro()) 6399 return false; 6400 6401 // Get line numbers of statement and body. 6402 bool StmtLineInvalid; 6403 unsigned StmtLine = SourceMgr.getSpellingLineNumber(StmtLoc, 6404 &StmtLineInvalid); 6405 if (StmtLineInvalid) 6406 return false; 6407 6408 bool BodyLineInvalid; 6409 unsigned BodyLine = SourceMgr.getSpellingLineNumber(Body->getSemiLoc(), 6410 &BodyLineInvalid); 6411 if (BodyLineInvalid) 6412 return false; 6413 6414 // Warn if null statement and body are on the same line. 6415 if (StmtLine != BodyLine) 6416 return false; 6417 6418 return true; 6419} 6420} // Unnamed namespace 6421 6422void Sema::DiagnoseEmptyStmtBody(SourceLocation StmtLoc, 6423 const Stmt *Body, 6424 unsigned DiagID) { 6425 // Since this is a syntactic check, don't emit diagnostic for template 6426 // instantiations, this just adds noise. 6427 if (CurrentInstantiationScope) 6428 return; 6429 6430 // The body should be a null statement. 6431 const NullStmt *NBody = dyn_cast<NullStmt>(Body); 6432 if (!NBody) 6433 return; 6434 6435 // Do the usual checks. 6436 if (!ShouldDiagnoseEmptyStmtBody(SourceMgr, StmtLoc, NBody)) 6437 return; 6438 6439 Diag(NBody->getSemiLoc(), DiagID); 6440 Diag(NBody->getSemiLoc(), diag::note_empty_body_on_separate_line); 6441} 6442 6443void Sema::DiagnoseEmptyLoopBody(const Stmt *S, 6444 const Stmt *PossibleBody) { 6445 assert(!CurrentInstantiationScope); // Ensured by caller 6446 6447 SourceLocation StmtLoc; 6448 const Stmt *Body; 6449 unsigned DiagID; 6450 if (const ForStmt *FS = dyn_cast<ForStmt>(S)) { 6451 StmtLoc = FS->getRParenLoc(); 6452 Body = FS->getBody(); 6453 DiagID = diag::warn_empty_for_body; 6454 } else if (const WhileStmt *WS = dyn_cast<WhileStmt>(S)) { 6455 StmtLoc = WS->getCond()->getSourceRange().getEnd(); 6456 Body = WS->getBody(); 6457 DiagID = diag::warn_empty_while_body; 6458 } else 6459 return; // Neither `for' nor `while'. 6460 6461 // The body should be a null statement. 6462 const NullStmt *NBody = dyn_cast<NullStmt>(Body); 6463 if (!NBody) 6464 return; 6465 6466 // Skip expensive checks if diagnostic is disabled. 6467 if (Diags.getDiagnosticLevel(DiagID, NBody->getSemiLoc()) == 6468 DiagnosticsEngine::Ignored) 6469 return; 6470 6471 // Do the usual checks. 6472 if (!ShouldDiagnoseEmptyStmtBody(SourceMgr, StmtLoc, NBody)) 6473 return; 6474 6475 // `for(...);' and `while(...);' are popular idioms, so in order to keep 6476 // noise level low, emit diagnostics only if for/while is followed by a 6477 // CompoundStmt, e.g.: 6478 // for (int i = 0; i < n; i++); 6479 // { 6480 // a(i); 6481 // } 6482 // or if for/while is followed by a statement with more indentation 6483 // than for/while itself: 6484 // for (int i = 0; i < n; i++); 6485 // a(i); 6486 bool ProbableTypo = isa<CompoundStmt>(PossibleBody); 6487 if (!ProbableTypo) { 6488 bool BodyColInvalid; 6489 unsigned BodyCol = SourceMgr.getPresumedColumnNumber( 6490 PossibleBody->getLocStart(), 6491 &BodyColInvalid); 6492 if (BodyColInvalid) 6493 return; 6494 6495 bool StmtColInvalid; 6496 unsigned StmtCol = SourceMgr.getPresumedColumnNumber( 6497 S->getLocStart(), 6498 &StmtColInvalid); 6499 if (StmtColInvalid) 6500 return; 6501 6502 if (BodyCol > StmtCol) 6503 ProbableTypo = true; 6504 } 6505 6506 if (ProbableTypo) { 6507 Diag(NBody->getSemiLoc(), DiagID); 6508 Diag(NBody->getSemiLoc(), diag::note_empty_body_on_separate_line); 6509 } 6510} 6511 6512//===--- Layout compatibility ----------------------------------------------// 6513 6514namespace { 6515 6516bool isLayoutCompatible(ASTContext &C, QualType T1, QualType T2); 6517 6518/// \brief Check if two enumeration types are layout-compatible. 6519bool isLayoutCompatible(ASTContext &C, EnumDecl *ED1, EnumDecl *ED2) { 6520 // C++11 [dcl.enum] p8: 6521 // Two enumeration types are layout-compatible if they have the same 6522 // underlying type. 6523 return ED1->isComplete() && ED2->isComplete() && 6524 C.hasSameType(ED1->getIntegerType(), ED2->getIntegerType()); 6525} 6526 6527/// \brief Check if two fields are layout-compatible. 6528bool isLayoutCompatible(ASTContext &C, FieldDecl *Field1, FieldDecl *Field2) { 6529 if (!isLayoutCompatible(C, Field1->getType(), Field2->getType())) 6530 return false; 6531 6532 if (Field1->isBitField() != Field2->isBitField()) 6533 return false; 6534 6535 if (Field1->isBitField()) { 6536 // Make sure that the bit-fields are the same length. 6537 unsigned Bits1 = Field1->getBitWidthValue(C); 6538 unsigned Bits2 = Field2->getBitWidthValue(C); 6539 6540 if (Bits1 != Bits2) 6541 return false; 6542 } 6543 6544 return true; 6545} 6546 6547/// \brief Check if two standard-layout structs are layout-compatible. 6548/// (C++11 [class.mem] p17) 6549bool isLayoutCompatibleStruct(ASTContext &C, 6550 RecordDecl *RD1, 6551 RecordDecl *RD2) { 6552 // If both records are C++ classes, check that base classes match. 6553 if (const CXXRecordDecl *D1CXX = dyn_cast<CXXRecordDecl>(RD1)) { 6554 // If one of records is a CXXRecordDecl we are in C++ mode, 6555 // thus the other one is a CXXRecordDecl, too. 6556 const CXXRecordDecl *D2CXX = cast<CXXRecordDecl>(RD2); 6557 // Check number of base classes. 6558 if (D1CXX->getNumBases() != D2CXX->getNumBases()) 6559 return false; 6560 6561 // Check the base classes. 6562 for (CXXRecordDecl::base_class_const_iterator 6563 Base1 = D1CXX->bases_begin(), 6564 BaseEnd1 = D1CXX->bases_end(), 6565 Base2 = D2CXX->bases_begin(); 6566 Base1 != BaseEnd1; 6567 ++Base1, ++Base2) { 6568 if (!isLayoutCompatible(C, Base1->getType(), Base2->getType())) 6569 return false; 6570 } 6571 } else if (const CXXRecordDecl *D2CXX = dyn_cast<CXXRecordDecl>(RD2)) { 6572 // If only RD2 is a C++ class, it should have zero base classes. 6573 if (D2CXX->getNumBases() > 0) 6574 return false; 6575 } 6576 6577 // Check the fields. 6578 RecordDecl::field_iterator Field2 = RD2->field_begin(), 6579 Field2End = RD2->field_end(), 6580 Field1 = RD1->field_begin(), 6581 Field1End = RD1->field_end(); 6582 for ( ; Field1 != Field1End && Field2 != Field2End; ++Field1, ++Field2) { 6583 if (!isLayoutCompatible(C, *Field1, *Field2)) 6584 return false; 6585 } 6586 if (Field1 != Field1End || Field2 != Field2End) 6587 return false; 6588 6589 return true; 6590} 6591 6592/// \brief Check if two standard-layout unions are layout-compatible. 6593/// (C++11 [class.mem] p18) 6594bool isLayoutCompatibleUnion(ASTContext &C, 6595 RecordDecl *RD1, 6596 RecordDecl *RD2) { 6597 llvm::SmallPtrSet<FieldDecl *, 8> UnmatchedFields; 6598 for (RecordDecl::field_iterator Field2 = RD2->field_begin(), 6599 Field2End = RD2->field_end(); 6600 Field2 != Field2End; ++Field2) { 6601 UnmatchedFields.insert(*Field2); 6602 } 6603 6604 for (RecordDecl::field_iterator Field1 = RD1->field_begin(), 6605 Field1End = RD1->field_end(); 6606 Field1 != Field1End; ++Field1) { 6607 llvm::SmallPtrSet<FieldDecl *, 8>::iterator 6608 I = UnmatchedFields.begin(), 6609 E = UnmatchedFields.end(); 6610 6611 for ( ; I != E; ++I) { 6612 if (isLayoutCompatible(C, *Field1, *I)) { 6613 bool Result = UnmatchedFields.erase(*I); 6614 (void) Result; 6615 assert(Result); 6616 break; 6617 } 6618 } 6619 if (I == E) 6620 return false; 6621 } 6622 6623 return UnmatchedFields.empty(); 6624} 6625 6626bool isLayoutCompatible(ASTContext &C, RecordDecl *RD1, RecordDecl *RD2) { 6627 if (RD1->isUnion() != RD2->isUnion()) 6628 return false; 6629 6630 if (RD1->isUnion()) 6631 return isLayoutCompatibleUnion(C, RD1, RD2); 6632 else 6633 return isLayoutCompatibleStruct(C, RD1, RD2); 6634} 6635 6636/// \brief Check if two types are layout-compatible in C++11 sense. 6637bool isLayoutCompatible(ASTContext &C, QualType T1, QualType T2) { 6638 if (T1.isNull() || T2.isNull()) 6639 return false; 6640 6641 // C++11 [basic.types] p11: 6642 // If two types T1 and T2 are the same type, then T1 and T2 are 6643 // layout-compatible types. 6644 if (C.hasSameType(T1, T2)) 6645 return true; 6646 6647 T1 = T1.getCanonicalType().getUnqualifiedType(); 6648 T2 = T2.getCanonicalType().getUnqualifiedType(); 6649 6650 const Type::TypeClass TC1 = T1->getTypeClass(); 6651 const Type::TypeClass TC2 = T2->getTypeClass(); 6652 6653 if (TC1 != TC2) 6654 return false; 6655 6656 if (TC1 == Type::Enum) { 6657 return isLayoutCompatible(C, 6658 cast<EnumType>(T1)->getDecl(), 6659 cast<EnumType>(T2)->getDecl()); 6660 } else if (TC1 == Type::Record) { 6661 if (!T1->isStandardLayoutType() || !T2->isStandardLayoutType()) 6662 return false; 6663 6664 return isLayoutCompatible(C, 6665 cast<RecordType>(T1)->getDecl(), 6666 cast<RecordType>(T2)->getDecl()); 6667 } 6668 6669 return false; 6670} 6671} 6672 6673//===--- CHECK: pointer_with_type_tag attribute: datatypes should match ----// 6674 6675namespace { 6676/// \brief Given a type tag expression find the type tag itself. 6677/// 6678/// \param TypeExpr Type tag expression, as it appears in user's code. 6679/// 6680/// \param VD Declaration of an identifier that appears in a type tag. 6681/// 6682/// \param MagicValue Type tag magic value. 6683bool FindTypeTagExpr(const Expr *TypeExpr, const ASTContext &Ctx, 6684 const ValueDecl **VD, uint64_t *MagicValue) { 6685 while(true) { 6686 if (!TypeExpr) 6687 return false; 6688 6689 TypeExpr = TypeExpr->IgnoreParenImpCasts()->IgnoreParenCasts(); 6690 6691 switch (TypeExpr->getStmtClass()) { 6692 case Stmt::UnaryOperatorClass: { 6693 const UnaryOperator *UO = cast<UnaryOperator>(TypeExpr); 6694 if (UO->getOpcode() == UO_AddrOf || UO->getOpcode() == UO_Deref) { 6695 TypeExpr = UO->getSubExpr(); 6696 continue; 6697 } 6698 return false; 6699 } 6700 6701 case Stmt::DeclRefExprClass: { 6702 const DeclRefExpr *DRE = cast<DeclRefExpr>(TypeExpr); 6703 *VD = DRE->getDecl(); 6704 return true; 6705 } 6706 6707 case Stmt::IntegerLiteralClass: { 6708 const IntegerLiteral *IL = cast<IntegerLiteral>(TypeExpr); 6709 llvm::APInt MagicValueAPInt = IL->getValue(); 6710 if (MagicValueAPInt.getActiveBits() <= 64) { 6711 *MagicValue = MagicValueAPInt.getZExtValue(); 6712 return true; 6713 } else 6714 return false; 6715 } 6716 6717 case Stmt::BinaryConditionalOperatorClass: 6718 case Stmt::ConditionalOperatorClass: { 6719 const AbstractConditionalOperator *ACO = 6720 cast<AbstractConditionalOperator>(TypeExpr); 6721 bool Result; 6722 if (ACO->getCond()->EvaluateAsBooleanCondition(Result, Ctx)) { 6723 if (Result) 6724 TypeExpr = ACO->getTrueExpr(); 6725 else 6726 TypeExpr = ACO->getFalseExpr(); 6727 continue; 6728 } 6729 return false; 6730 } 6731 6732 case Stmt::BinaryOperatorClass: { 6733 const BinaryOperator *BO = cast<BinaryOperator>(TypeExpr); 6734 if (BO->getOpcode() == BO_Comma) { 6735 TypeExpr = BO->getRHS(); 6736 continue; 6737 } 6738 return false; 6739 } 6740 6741 default: 6742 return false; 6743 } 6744 } 6745} 6746 6747/// \brief Retrieve the C type corresponding to type tag TypeExpr. 6748/// 6749/// \param TypeExpr Expression that specifies a type tag. 6750/// 6751/// \param MagicValues Registered magic values. 6752/// 6753/// \param FoundWrongKind Set to true if a type tag was found, but of a wrong 6754/// kind. 6755/// 6756/// \param TypeInfo Information about the corresponding C type. 6757/// 6758/// \returns true if the corresponding C type was found. 6759bool GetMatchingCType( 6760 const IdentifierInfo *ArgumentKind, 6761 const Expr *TypeExpr, const ASTContext &Ctx, 6762 const llvm::DenseMap<Sema::TypeTagMagicValue, 6763 Sema::TypeTagData> *MagicValues, 6764 bool &FoundWrongKind, 6765 Sema::TypeTagData &TypeInfo) { 6766 FoundWrongKind = false; 6767 6768 // Variable declaration that has type_tag_for_datatype attribute. 6769 const ValueDecl *VD = NULL; 6770 6771 uint64_t MagicValue; 6772 6773 if (!FindTypeTagExpr(TypeExpr, Ctx, &VD, &MagicValue)) 6774 return false; 6775 6776 if (VD) { 6777 for (specific_attr_iterator<TypeTagForDatatypeAttr> 6778 I = VD->specific_attr_begin<TypeTagForDatatypeAttr>(), 6779 E = VD->specific_attr_end<TypeTagForDatatypeAttr>(); 6780 I != E; ++I) { 6781 if (I->getArgumentKind() != ArgumentKind) { 6782 FoundWrongKind = true; 6783 return false; 6784 } 6785 TypeInfo.Type = I->getMatchingCType(); 6786 TypeInfo.LayoutCompatible = I->getLayoutCompatible(); 6787 TypeInfo.MustBeNull = I->getMustBeNull(); 6788 return true; 6789 } 6790 return false; 6791 } 6792 6793 if (!MagicValues) 6794 return false; 6795 6796 llvm::DenseMap<Sema::TypeTagMagicValue, 6797 Sema::TypeTagData>::const_iterator I = 6798 MagicValues->find(std::make_pair(ArgumentKind, MagicValue)); 6799 if (I == MagicValues->end()) 6800 return false; 6801 6802 TypeInfo = I->second; 6803 return true; 6804} 6805} // unnamed namespace 6806 6807void Sema::RegisterTypeTagForDatatype(const IdentifierInfo *ArgumentKind, 6808 uint64_t MagicValue, QualType Type, 6809 bool LayoutCompatible, 6810 bool MustBeNull) { 6811 if (!TypeTagForDatatypeMagicValues) 6812 TypeTagForDatatypeMagicValues.reset( 6813 new llvm::DenseMap<TypeTagMagicValue, TypeTagData>); 6814 6815 TypeTagMagicValue Magic(ArgumentKind, MagicValue); 6816 (*TypeTagForDatatypeMagicValues)[Magic] = 6817 TypeTagData(Type, LayoutCompatible, MustBeNull); 6818} 6819 6820namespace { 6821bool IsSameCharType(QualType T1, QualType T2) { 6822 const BuiltinType *BT1 = T1->getAs<BuiltinType>(); 6823 if (!BT1) 6824 return false; 6825 6826 const BuiltinType *BT2 = T2->getAs<BuiltinType>(); 6827 if (!BT2) 6828 return false; 6829 6830 BuiltinType::Kind T1Kind = BT1->getKind(); 6831 BuiltinType::Kind T2Kind = BT2->getKind(); 6832 6833 return (T1Kind == BuiltinType::SChar && T2Kind == BuiltinType::Char_S) || 6834 (T1Kind == BuiltinType::UChar && T2Kind == BuiltinType::Char_U) || 6835 (T1Kind == BuiltinType::Char_U && T2Kind == BuiltinType::UChar) || 6836 (T1Kind == BuiltinType::Char_S && T2Kind == BuiltinType::SChar); 6837} 6838} // unnamed namespace 6839 6840void Sema::CheckArgumentWithTypeTag(const ArgumentWithTypeTagAttr *Attr, 6841 const Expr * const *ExprArgs) { 6842 const IdentifierInfo *ArgumentKind = Attr->getArgumentKind(); 6843 bool IsPointerAttr = Attr->getIsPointer(); 6844 6845 const Expr *TypeTagExpr = ExprArgs[Attr->getTypeTagIdx()]; 6846 bool FoundWrongKind; 6847 TypeTagData TypeInfo; 6848 if (!GetMatchingCType(ArgumentKind, TypeTagExpr, Context, 6849 TypeTagForDatatypeMagicValues.get(), 6850 FoundWrongKind, TypeInfo)) { 6851 if (FoundWrongKind) 6852 Diag(TypeTagExpr->getExprLoc(), 6853 diag::warn_type_tag_for_datatype_wrong_kind) 6854 << TypeTagExpr->getSourceRange(); 6855 return; 6856 } 6857 6858 const Expr *ArgumentExpr = ExprArgs[Attr->getArgumentIdx()]; 6859 if (IsPointerAttr) { 6860 // Skip implicit cast of pointer to `void *' (as a function argument). 6861 if (const ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(ArgumentExpr)) 6862 if (ICE->getType()->isVoidPointerType() && 6863 ICE->getCastKind() == CK_BitCast) 6864 ArgumentExpr = ICE->getSubExpr(); 6865 } 6866 QualType ArgumentType = ArgumentExpr->getType(); 6867 6868 // Passing a `void*' pointer shouldn't trigger a warning. 6869 if (IsPointerAttr && ArgumentType->isVoidPointerType()) 6870 return; 6871 6872 if (TypeInfo.MustBeNull) { 6873 // Type tag with matching void type requires a null pointer. 6874 if (!ArgumentExpr->isNullPointerConstant(Context, 6875 Expr::NPC_ValueDependentIsNotNull)) { 6876 Diag(ArgumentExpr->getExprLoc(), 6877 diag::warn_type_safety_null_pointer_required) 6878 << ArgumentKind->getName() 6879 << ArgumentExpr->getSourceRange() 6880 << TypeTagExpr->getSourceRange(); 6881 } 6882 return; 6883 } 6884 6885 QualType RequiredType = TypeInfo.Type; 6886 if (IsPointerAttr) 6887 RequiredType = Context.getPointerType(RequiredType); 6888 6889 bool mismatch = false; 6890 if (!TypeInfo.LayoutCompatible) { 6891 mismatch = !Context.hasSameType(ArgumentType, RequiredType); 6892 6893 // C++11 [basic.fundamental] p1: 6894 // Plain char, signed char, and unsigned char are three distinct types. 6895 // 6896 // But we treat plain `char' as equivalent to `signed char' or `unsigned 6897 // char' depending on the current char signedness mode. 6898 if (mismatch) 6899 if ((IsPointerAttr && IsSameCharType(ArgumentType->getPointeeType(), 6900 RequiredType->getPointeeType())) || 6901 (!IsPointerAttr && IsSameCharType(ArgumentType, RequiredType))) 6902 mismatch = false; 6903 } else 6904 if (IsPointerAttr) 6905 mismatch = !isLayoutCompatible(Context, 6906 ArgumentType->getPointeeType(), 6907 RequiredType->getPointeeType()); 6908 else 6909 mismatch = !isLayoutCompatible(Context, ArgumentType, RequiredType); 6910 6911 if (mismatch) 6912 Diag(ArgumentExpr->getExprLoc(), diag::warn_type_safety_type_mismatch) 6913 << ArgumentType << ArgumentKind->getName() 6914 << TypeInfo.LayoutCompatible << RequiredType 6915 << ArgumentExpr->getSourceRange() 6916 << TypeTagExpr->getSourceRange(); 6917} 6918