SemaChecking.cpp revision 09bddcf8c0ce4cc2f2a18e050e971539e8a396f8
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 if (FDecl) 503 for (specific_attr_iterator<FormatAttr> 504 I = FDecl->specific_attr_begin<FormatAttr>(), 505 E = FDecl->specific_attr_end<FormatAttr>(); I != E ; ++I) 506 if (CheckFormatArguments(*I, Args, IsMemberFunction, CallType, Loc, 507 Range)) 508 HandledFormatString = true; 509 510 // Refuse POD arguments that weren't caught by the format string 511 // checks above. 512 if (!HandledFormatString && CallType != VariadicDoesNotApply) 513 for (unsigned ArgIdx = NumProtoArgs; ArgIdx < Args.size(); ++ArgIdx) { 514 // Args[ArgIdx] can be null in malformed code. 515 if (const Expr *Arg = Args[ArgIdx]) 516 variadicArgumentPODCheck(Arg, CallType); 517 } 518 519 if (FDecl) { 520 for (specific_attr_iterator<NonNullAttr> 521 I = FDecl->specific_attr_begin<NonNullAttr>(), 522 E = FDecl->specific_attr_end<NonNullAttr>(); I != E; ++I) 523 CheckNonNullArguments(*I, Args.data(), Loc); 524 525 // Type safety checking. 526 for (specific_attr_iterator<ArgumentWithTypeTagAttr> 527 i = FDecl->specific_attr_begin<ArgumentWithTypeTagAttr>(), 528 e = FDecl->specific_attr_end<ArgumentWithTypeTagAttr>(); 529 i != e; ++i) { 530 CheckArgumentWithTypeTag(*i, Args.data()); 531 } 532 } 533} 534 535/// CheckConstructorCall - Check a constructor call for correctness and safety 536/// properties not enforced by the C type system. 537void Sema::CheckConstructorCall(FunctionDecl *FDecl, 538 ArrayRef<const Expr *> Args, 539 const FunctionProtoType *Proto, 540 SourceLocation Loc) { 541 VariadicCallType CallType = 542 Proto->isVariadic() ? VariadicConstructor : VariadicDoesNotApply; 543 checkCall(FDecl, Args, Proto->getNumArgs(), 544 /*IsMemberFunction=*/true, Loc, SourceRange(), CallType); 545} 546 547/// CheckFunctionCall - Check a direct function call for various correctness 548/// and safety properties not strictly enforced by the C type system. 549bool Sema::CheckFunctionCall(FunctionDecl *FDecl, CallExpr *TheCall, 550 const FunctionProtoType *Proto) { 551 bool IsMemberOperatorCall = isa<CXXOperatorCallExpr>(TheCall) && 552 isa<CXXMethodDecl>(FDecl); 553 bool IsMemberFunction = isa<CXXMemberCallExpr>(TheCall) || 554 IsMemberOperatorCall; 555 VariadicCallType CallType = getVariadicCallType(FDecl, Proto, 556 TheCall->getCallee()); 557 unsigned NumProtoArgs = Proto ? Proto->getNumArgs() : 0; 558 Expr** Args = TheCall->getArgs(); 559 unsigned NumArgs = TheCall->getNumArgs(); 560 if (IsMemberOperatorCall) { 561 // If this is a call to a member operator, hide the first argument 562 // from checkCall. 563 // FIXME: Our choice of AST representation here is less than ideal. 564 ++Args; 565 --NumArgs; 566 } 567 checkCall(FDecl, llvm::makeArrayRef<const Expr *>(Args, NumArgs), 568 NumProtoArgs, 569 IsMemberFunction, TheCall->getRParenLoc(), 570 TheCall->getCallee()->getSourceRange(), CallType); 571 572 IdentifierInfo *FnInfo = FDecl->getIdentifier(); 573 // None of the checks below are needed for functions that don't have 574 // simple names (e.g., C++ conversion functions). 575 if (!FnInfo) 576 return false; 577 578 unsigned CMId = FDecl->getMemoryFunctionKind(); 579 if (CMId == 0) 580 return false; 581 582 // Handle memory setting and copying functions. 583 if (CMId == Builtin::BIstrlcpy || CMId == Builtin::BIstrlcat) 584 CheckStrlcpycatArguments(TheCall, FnInfo); 585 else if (CMId == Builtin::BIstrncat) 586 CheckStrncatArguments(TheCall, FnInfo); 587 else 588 CheckMemaccessArguments(TheCall, CMId, FnInfo); 589 590 return false; 591} 592 593bool Sema::CheckObjCMethodCall(ObjCMethodDecl *Method, SourceLocation lbrac, 594 ArrayRef<const Expr *> Args) { 595 VariadicCallType CallType = 596 Method->isVariadic() ? VariadicMethod : VariadicDoesNotApply; 597 598 checkCall(Method, Args, Method->param_size(), 599 /*IsMemberFunction=*/false, 600 lbrac, Method->getSourceRange(), CallType); 601 602 return false; 603} 604 605bool Sema::CheckPointerCall(NamedDecl *NDecl, CallExpr *TheCall, 606 const FunctionProtoType *Proto) { 607 const VarDecl *V = dyn_cast<VarDecl>(NDecl); 608 if (!V) 609 return false; 610 611 QualType Ty = V->getType(); 612 if (!Ty->isBlockPointerType() && !Ty->isFunctionPointerType()) 613 return false; 614 615 VariadicCallType CallType; 616 if (!Proto || !Proto->isVariadic()) { 617 CallType = VariadicDoesNotApply; 618 } else if (Ty->isBlockPointerType()) { 619 CallType = VariadicBlock; 620 } else { // Ty->isFunctionPointerType() 621 CallType = VariadicFunction; 622 } 623 unsigned NumProtoArgs = Proto ? Proto->getNumArgs() : 0; 624 625 checkCall(NDecl, 626 llvm::makeArrayRef<const Expr *>(TheCall->getArgs(), 627 TheCall->getNumArgs()), 628 NumProtoArgs, /*IsMemberFunction=*/false, 629 TheCall->getRParenLoc(), 630 TheCall->getCallee()->getSourceRange(), CallType); 631 632 return false; 633} 634 635/// Checks function calls when a FunctionDecl or a NamedDecl is not available, 636/// such as function pointers returned from functions. 637bool Sema::CheckOtherCall(CallExpr *TheCall, const FunctionProtoType *Proto) { 638 VariadicCallType CallType = getVariadicCallType(/*FDecl=*/0, Proto, 639 TheCall->getCallee()); 640 unsigned NumProtoArgs = Proto ? Proto->getNumArgs() : 0; 641 642 checkCall(/*FDecl=*/0, 643 llvm::makeArrayRef<const Expr *>(TheCall->getArgs(), 644 TheCall->getNumArgs()), 645 NumProtoArgs, /*IsMemberFunction=*/false, 646 TheCall->getRParenLoc(), 647 TheCall->getCallee()->getSourceRange(), CallType); 648 649 return false; 650} 651 652ExprResult Sema::SemaAtomicOpsOverloaded(ExprResult TheCallResult, 653 AtomicExpr::AtomicOp Op) { 654 CallExpr *TheCall = cast<CallExpr>(TheCallResult.get()); 655 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 656 657 // All these operations take one of the following forms: 658 enum { 659 // C __c11_atomic_init(A *, C) 660 Init, 661 // C __c11_atomic_load(A *, int) 662 Load, 663 // void __atomic_load(A *, CP, int) 664 Copy, 665 // C __c11_atomic_add(A *, M, int) 666 Arithmetic, 667 // C __atomic_exchange_n(A *, CP, int) 668 Xchg, 669 // void __atomic_exchange(A *, C *, CP, int) 670 GNUXchg, 671 // bool __c11_atomic_compare_exchange_strong(A *, C *, CP, int, int) 672 C11CmpXchg, 673 // bool __atomic_compare_exchange(A *, C *, CP, bool, int, int) 674 GNUCmpXchg 675 } Form = Init; 676 const unsigned NumArgs[] = { 2, 2, 3, 3, 3, 4, 5, 6 }; 677 const unsigned NumVals[] = { 1, 0, 1, 1, 1, 2, 2, 3 }; 678 // where: 679 // C is an appropriate type, 680 // A is volatile _Atomic(C) for __c11 builtins and is C for GNU builtins, 681 // CP is C for __c11 builtins and GNU _n builtins and is C * otherwise, 682 // M is C if C is an integer, and ptrdiff_t if C is a pointer, and 683 // the int parameters are for orderings. 684 685 assert(AtomicExpr::AO__c11_atomic_init == 0 && 686 AtomicExpr::AO__c11_atomic_fetch_xor + 1 == AtomicExpr::AO__atomic_load 687 && "need to update code for modified C11 atomics"); 688 bool IsC11 = Op >= AtomicExpr::AO__c11_atomic_init && 689 Op <= AtomicExpr::AO__c11_atomic_fetch_xor; 690 bool IsN = Op == AtomicExpr::AO__atomic_load_n || 691 Op == AtomicExpr::AO__atomic_store_n || 692 Op == AtomicExpr::AO__atomic_exchange_n || 693 Op == AtomicExpr::AO__atomic_compare_exchange_n; 694 bool IsAddSub = false; 695 696 switch (Op) { 697 case AtomicExpr::AO__c11_atomic_init: 698 Form = Init; 699 break; 700 701 case AtomicExpr::AO__c11_atomic_load: 702 case AtomicExpr::AO__atomic_load_n: 703 Form = Load; 704 break; 705 706 case AtomicExpr::AO__c11_atomic_store: 707 case AtomicExpr::AO__atomic_load: 708 case AtomicExpr::AO__atomic_store: 709 case AtomicExpr::AO__atomic_store_n: 710 Form = Copy; 711 break; 712 713 case AtomicExpr::AO__c11_atomic_fetch_add: 714 case AtomicExpr::AO__c11_atomic_fetch_sub: 715 case AtomicExpr::AO__atomic_fetch_add: 716 case AtomicExpr::AO__atomic_fetch_sub: 717 case AtomicExpr::AO__atomic_add_fetch: 718 case AtomicExpr::AO__atomic_sub_fetch: 719 IsAddSub = true; 720 // Fall through. 721 case AtomicExpr::AO__c11_atomic_fetch_and: 722 case AtomicExpr::AO__c11_atomic_fetch_or: 723 case AtomicExpr::AO__c11_atomic_fetch_xor: 724 case AtomicExpr::AO__atomic_fetch_and: 725 case AtomicExpr::AO__atomic_fetch_or: 726 case AtomicExpr::AO__atomic_fetch_xor: 727 case AtomicExpr::AO__atomic_fetch_nand: 728 case AtomicExpr::AO__atomic_and_fetch: 729 case AtomicExpr::AO__atomic_or_fetch: 730 case AtomicExpr::AO__atomic_xor_fetch: 731 case AtomicExpr::AO__atomic_nand_fetch: 732 Form = Arithmetic; 733 break; 734 735 case AtomicExpr::AO__c11_atomic_exchange: 736 case AtomicExpr::AO__atomic_exchange_n: 737 Form = Xchg; 738 break; 739 740 case AtomicExpr::AO__atomic_exchange: 741 Form = GNUXchg; 742 break; 743 744 case AtomicExpr::AO__c11_atomic_compare_exchange_strong: 745 case AtomicExpr::AO__c11_atomic_compare_exchange_weak: 746 Form = C11CmpXchg; 747 break; 748 749 case AtomicExpr::AO__atomic_compare_exchange: 750 case AtomicExpr::AO__atomic_compare_exchange_n: 751 Form = GNUCmpXchg; 752 break; 753 } 754 755 // Check we have the right number of arguments. 756 if (TheCall->getNumArgs() < NumArgs[Form]) { 757 Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 758 << 0 << NumArgs[Form] << TheCall->getNumArgs() 759 << TheCall->getCallee()->getSourceRange(); 760 return ExprError(); 761 } else if (TheCall->getNumArgs() > NumArgs[Form]) { 762 Diag(TheCall->getArg(NumArgs[Form])->getLocStart(), 763 diag::err_typecheck_call_too_many_args) 764 << 0 << NumArgs[Form] << TheCall->getNumArgs() 765 << TheCall->getCallee()->getSourceRange(); 766 return ExprError(); 767 } 768 769 // Inspect the first argument of the atomic operation. 770 Expr *Ptr = TheCall->getArg(0); 771 Ptr = DefaultFunctionArrayLvalueConversion(Ptr).get(); 772 const PointerType *pointerType = Ptr->getType()->getAs<PointerType>(); 773 if (!pointerType) { 774 Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer) 775 << Ptr->getType() << Ptr->getSourceRange(); 776 return ExprError(); 777 } 778 779 // For a __c11 builtin, this should be a pointer to an _Atomic type. 780 QualType AtomTy = pointerType->getPointeeType(); // 'A' 781 QualType ValType = AtomTy; // 'C' 782 if (IsC11) { 783 if (!AtomTy->isAtomicType()) { 784 Diag(DRE->getLocStart(), diag::err_atomic_op_needs_atomic) 785 << Ptr->getType() << Ptr->getSourceRange(); 786 return ExprError(); 787 } 788 if (AtomTy.isConstQualified()) { 789 Diag(DRE->getLocStart(), diag::err_atomic_op_needs_non_const_atomic) 790 << Ptr->getType() << Ptr->getSourceRange(); 791 return ExprError(); 792 } 793 ValType = AtomTy->getAs<AtomicType>()->getValueType(); 794 } 795 796 // For an arithmetic operation, the implied arithmetic must be well-formed. 797 if (Form == Arithmetic) { 798 // gcc does not enforce these rules for GNU atomics, but we do so for sanity. 799 if (IsAddSub && !ValType->isIntegerType() && !ValType->isPointerType()) { 800 Diag(DRE->getLocStart(), diag::err_atomic_op_needs_atomic_int_or_ptr) 801 << IsC11 << Ptr->getType() << Ptr->getSourceRange(); 802 return ExprError(); 803 } 804 if (!IsAddSub && !ValType->isIntegerType()) { 805 Diag(DRE->getLocStart(), diag::err_atomic_op_bitwise_needs_atomic_int) 806 << IsC11 << Ptr->getType() << Ptr->getSourceRange(); 807 return ExprError(); 808 } 809 } else if (IsN && !ValType->isIntegerType() && !ValType->isPointerType()) { 810 // For __atomic_*_n operations, the value type must be a scalar integral or 811 // pointer type which is 1, 2, 4, 8 or 16 bytes in length. 812 Diag(DRE->getLocStart(), diag::err_atomic_op_needs_atomic_int_or_ptr) 813 << IsC11 << Ptr->getType() << Ptr->getSourceRange(); 814 return ExprError(); 815 } 816 817 if (!IsC11 && !AtomTy.isTriviallyCopyableType(Context)) { 818 // For GNU atomics, require a trivially-copyable type. This is not part of 819 // the GNU atomics specification, but we enforce it for sanity. 820 Diag(DRE->getLocStart(), diag::err_atomic_op_needs_trivial_copy) 821 << Ptr->getType() << Ptr->getSourceRange(); 822 return ExprError(); 823 } 824 825 // FIXME: For any builtin other than a load, the ValType must not be 826 // const-qualified. 827 828 switch (ValType.getObjCLifetime()) { 829 case Qualifiers::OCL_None: 830 case Qualifiers::OCL_ExplicitNone: 831 // okay 832 break; 833 834 case Qualifiers::OCL_Weak: 835 case Qualifiers::OCL_Strong: 836 case Qualifiers::OCL_Autoreleasing: 837 // FIXME: Can this happen? By this point, ValType should be known 838 // to be trivially copyable. 839 Diag(DRE->getLocStart(), diag::err_arc_atomic_ownership) 840 << ValType << Ptr->getSourceRange(); 841 return ExprError(); 842 } 843 844 QualType ResultType = ValType; 845 if (Form == Copy || Form == GNUXchg || Form == Init) 846 ResultType = Context.VoidTy; 847 else if (Form == C11CmpXchg || Form == GNUCmpXchg) 848 ResultType = Context.BoolTy; 849 850 // The type of a parameter passed 'by value'. In the GNU atomics, such 851 // arguments are actually passed as pointers. 852 QualType ByValType = ValType; // 'CP' 853 if (!IsC11 && !IsN) 854 ByValType = Ptr->getType(); 855 856 // The first argument --- the pointer --- has a fixed type; we 857 // deduce the types of the rest of the arguments accordingly. Walk 858 // the remaining arguments, converting them to the deduced value type. 859 for (unsigned i = 1; i != NumArgs[Form]; ++i) { 860 QualType Ty; 861 if (i < NumVals[Form] + 1) { 862 switch (i) { 863 case 1: 864 // The second argument is the non-atomic operand. For arithmetic, this 865 // is always passed by value, and for a compare_exchange it is always 866 // passed by address. For the rest, GNU uses by-address and C11 uses 867 // by-value. 868 assert(Form != Load); 869 if (Form == Init || (Form == Arithmetic && ValType->isIntegerType())) 870 Ty = ValType; 871 else if (Form == Copy || Form == Xchg) 872 Ty = ByValType; 873 else if (Form == Arithmetic) 874 Ty = Context.getPointerDiffType(); 875 else 876 Ty = Context.getPointerType(ValType.getUnqualifiedType()); 877 break; 878 case 2: 879 // The third argument to compare_exchange / GNU exchange is a 880 // (pointer to a) desired value. 881 Ty = ByValType; 882 break; 883 case 3: 884 // The fourth argument to GNU compare_exchange is a 'weak' flag. 885 Ty = Context.BoolTy; 886 break; 887 } 888 } else { 889 // The order(s) are always converted to int. 890 Ty = Context.IntTy; 891 } 892 893 InitializedEntity Entity = 894 InitializedEntity::InitializeParameter(Context, Ty, false); 895 ExprResult Arg = TheCall->getArg(i); 896 Arg = PerformCopyInitialization(Entity, SourceLocation(), Arg); 897 if (Arg.isInvalid()) 898 return true; 899 TheCall->setArg(i, Arg.get()); 900 } 901 902 // Permute the arguments into a 'consistent' order. 903 SmallVector<Expr*, 5> SubExprs; 904 SubExprs.push_back(Ptr); 905 switch (Form) { 906 case Init: 907 // Note, AtomicExpr::getVal1() has a special case for this atomic. 908 SubExprs.push_back(TheCall->getArg(1)); // Val1 909 break; 910 case Load: 911 SubExprs.push_back(TheCall->getArg(1)); // Order 912 break; 913 case Copy: 914 case Arithmetic: 915 case Xchg: 916 SubExprs.push_back(TheCall->getArg(2)); // Order 917 SubExprs.push_back(TheCall->getArg(1)); // Val1 918 break; 919 case GNUXchg: 920 // Note, AtomicExpr::getVal2() has a special case for this atomic. 921 SubExprs.push_back(TheCall->getArg(3)); // Order 922 SubExprs.push_back(TheCall->getArg(1)); // Val1 923 SubExprs.push_back(TheCall->getArg(2)); // Val2 924 break; 925 case C11CmpXchg: 926 SubExprs.push_back(TheCall->getArg(3)); // Order 927 SubExprs.push_back(TheCall->getArg(1)); // Val1 928 SubExprs.push_back(TheCall->getArg(4)); // OrderFail 929 SubExprs.push_back(TheCall->getArg(2)); // Val2 930 break; 931 case GNUCmpXchg: 932 SubExprs.push_back(TheCall->getArg(4)); // Order 933 SubExprs.push_back(TheCall->getArg(1)); // Val1 934 SubExprs.push_back(TheCall->getArg(5)); // OrderFail 935 SubExprs.push_back(TheCall->getArg(2)); // Val2 936 SubExprs.push_back(TheCall->getArg(3)); // Weak 937 break; 938 } 939 940 AtomicExpr *AE = new (Context) AtomicExpr(TheCall->getCallee()->getLocStart(), 941 SubExprs, ResultType, Op, 942 TheCall->getRParenLoc()); 943 944 if ((Op == AtomicExpr::AO__c11_atomic_load || 945 (Op == AtomicExpr::AO__c11_atomic_store)) && 946 Context.AtomicUsesUnsupportedLibcall(AE)) 947 Diag(AE->getLocStart(), diag::err_atomic_load_store_uses_lib) << 948 ((Op == AtomicExpr::AO__c11_atomic_load) ? 0 : 1); 949 950 return Owned(AE); 951} 952 953 954/// checkBuiltinArgument - Given a call to a builtin function, perform 955/// normal type-checking on the given argument, updating the call in 956/// place. This is useful when a builtin function requires custom 957/// type-checking for some of its arguments but not necessarily all of 958/// them. 959/// 960/// Returns true on error. 961static bool checkBuiltinArgument(Sema &S, CallExpr *E, unsigned ArgIndex) { 962 FunctionDecl *Fn = E->getDirectCallee(); 963 assert(Fn && "builtin call without direct callee!"); 964 965 ParmVarDecl *Param = Fn->getParamDecl(ArgIndex); 966 InitializedEntity Entity = 967 InitializedEntity::InitializeParameter(S.Context, Param); 968 969 ExprResult Arg = E->getArg(0); 970 Arg = S.PerformCopyInitialization(Entity, SourceLocation(), Arg); 971 if (Arg.isInvalid()) 972 return true; 973 974 E->setArg(ArgIndex, Arg.take()); 975 return false; 976} 977 978/// SemaBuiltinAtomicOverloaded - We have a call to a function like 979/// __sync_fetch_and_add, which is an overloaded function based on the pointer 980/// type of its first argument. The main ActOnCallExpr routines have already 981/// promoted the types of arguments because all of these calls are prototyped as 982/// void(...). 983/// 984/// This function goes through and does final semantic checking for these 985/// builtins, 986ExprResult 987Sema::SemaBuiltinAtomicOverloaded(ExprResult TheCallResult) { 988 CallExpr *TheCall = (CallExpr *)TheCallResult.get(); 989 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 990 FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl()); 991 992 // Ensure that we have at least one argument to do type inference from. 993 if (TheCall->getNumArgs() < 1) { 994 Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args_at_least) 995 << 0 << 1 << TheCall->getNumArgs() 996 << TheCall->getCallee()->getSourceRange(); 997 return ExprError(); 998 } 999 1000 // Inspect the first argument of the atomic builtin. This should always be 1001 // a pointer type, whose element is an integral scalar or pointer type. 1002 // Because it is a pointer type, we don't have to worry about any implicit 1003 // casts here. 1004 // FIXME: We don't allow floating point scalars as input. 1005 Expr *FirstArg = TheCall->getArg(0); 1006 ExprResult FirstArgResult = DefaultFunctionArrayLvalueConversion(FirstArg); 1007 if (FirstArgResult.isInvalid()) 1008 return ExprError(); 1009 FirstArg = FirstArgResult.take(); 1010 TheCall->setArg(0, FirstArg); 1011 1012 const PointerType *pointerType = FirstArg->getType()->getAs<PointerType>(); 1013 if (!pointerType) { 1014 Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer) 1015 << FirstArg->getType() << FirstArg->getSourceRange(); 1016 return ExprError(); 1017 } 1018 1019 QualType ValType = pointerType->getPointeeType(); 1020 if (!ValType->isIntegerType() && !ValType->isAnyPointerType() && 1021 !ValType->isBlockPointerType()) { 1022 Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer_intptr) 1023 << FirstArg->getType() << FirstArg->getSourceRange(); 1024 return ExprError(); 1025 } 1026 1027 switch (ValType.getObjCLifetime()) { 1028 case Qualifiers::OCL_None: 1029 case Qualifiers::OCL_ExplicitNone: 1030 // okay 1031 break; 1032 1033 case Qualifiers::OCL_Weak: 1034 case Qualifiers::OCL_Strong: 1035 case Qualifiers::OCL_Autoreleasing: 1036 Diag(DRE->getLocStart(), diag::err_arc_atomic_ownership) 1037 << ValType << FirstArg->getSourceRange(); 1038 return ExprError(); 1039 } 1040 1041 // Strip any qualifiers off ValType. 1042 ValType = ValType.getUnqualifiedType(); 1043 1044 // The majority of builtins return a value, but a few have special return 1045 // types, so allow them to override appropriately below. 1046 QualType ResultType = ValType; 1047 1048 // We need to figure out which concrete builtin this maps onto. For example, 1049 // __sync_fetch_and_add with a 2 byte object turns into 1050 // __sync_fetch_and_add_2. 1051#define BUILTIN_ROW(x) \ 1052 { Builtin::BI##x##_1, Builtin::BI##x##_2, Builtin::BI##x##_4, \ 1053 Builtin::BI##x##_8, Builtin::BI##x##_16 } 1054 1055 static const unsigned BuiltinIndices[][5] = { 1056 BUILTIN_ROW(__sync_fetch_and_add), 1057 BUILTIN_ROW(__sync_fetch_and_sub), 1058 BUILTIN_ROW(__sync_fetch_and_or), 1059 BUILTIN_ROW(__sync_fetch_and_and), 1060 BUILTIN_ROW(__sync_fetch_and_xor), 1061 1062 BUILTIN_ROW(__sync_add_and_fetch), 1063 BUILTIN_ROW(__sync_sub_and_fetch), 1064 BUILTIN_ROW(__sync_and_and_fetch), 1065 BUILTIN_ROW(__sync_or_and_fetch), 1066 BUILTIN_ROW(__sync_xor_and_fetch), 1067 1068 BUILTIN_ROW(__sync_val_compare_and_swap), 1069 BUILTIN_ROW(__sync_bool_compare_and_swap), 1070 BUILTIN_ROW(__sync_lock_test_and_set), 1071 BUILTIN_ROW(__sync_lock_release), 1072 BUILTIN_ROW(__sync_swap) 1073 }; 1074#undef BUILTIN_ROW 1075 1076 // Determine the index of the size. 1077 unsigned SizeIndex; 1078 switch (Context.getTypeSizeInChars(ValType).getQuantity()) { 1079 case 1: SizeIndex = 0; break; 1080 case 2: SizeIndex = 1; break; 1081 case 4: SizeIndex = 2; break; 1082 case 8: SizeIndex = 3; break; 1083 case 16: SizeIndex = 4; break; 1084 default: 1085 Diag(DRE->getLocStart(), diag::err_atomic_builtin_pointer_size) 1086 << FirstArg->getType() << FirstArg->getSourceRange(); 1087 return ExprError(); 1088 } 1089 1090 // Each of these builtins has one pointer argument, followed by some number of 1091 // values (0, 1 or 2) followed by a potentially empty varags list of stuff 1092 // that we ignore. Find out which row of BuiltinIndices to read from as well 1093 // as the number of fixed args. 1094 unsigned BuiltinID = FDecl->getBuiltinID(); 1095 unsigned BuiltinIndex, NumFixed = 1; 1096 switch (BuiltinID) { 1097 default: llvm_unreachable("Unknown overloaded atomic builtin!"); 1098 case Builtin::BI__sync_fetch_and_add: 1099 case Builtin::BI__sync_fetch_and_add_1: 1100 case Builtin::BI__sync_fetch_and_add_2: 1101 case Builtin::BI__sync_fetch_and_add_4: 1102 case Builtin::BI__sync_fetch_and_add_8: 1103 case Builtin::BI__sync_fetch_and_add_16: 1104 BuiltinIndex = 0; 1105 break; 1106 1107 case Builtin::BI__sync_fetch_and_sub: 1108 case Builtin::BI__sync_fetch_and_sub_1: 1109 case Builtin::BI__sync_fetch_and_sub_2: 1110 case Builtin::BI__sync_fetch_and_sub_4: 1111 case Builtin::BI__sync_fetch_and_sub_8: 1112 case Builtin::BI__sync_fetch_and_sub_16: 1113 BuiltinIndex = 1; 1114 break; 1115 1116 case Builtin::BI__sync_fetch_and_or: 1117 case Builtin::BI__sync_fetch_and_or_1: 1118 case Builtin::BI__sync_fetch_and_or_2: 1119 case Builtin::BI__sync_fetch_and_or_4: 1120 case Builtin::BI__sync_fetch_and_or_8: 1121 case Builtin::BI__sync_fetch_and_or_16: 1122 BuiltinIndex = 2; 1123 break; 1124 1125 case Builtin::BI__sync_fetch_and_and: 1126 case Builtin::BI__sync_fetch_and_and_1: 1127 case Builtin::BI__sync_fetch_and_and_2: 1128 case Builtin::BI__sync_fetch_and_and_4: 1129 case Builtin::BI__sync_fetch_and_and_8: 1130 case Builtin::BI__sync_fetch_and_and_16: 1131 BuiltinIndex = 3; 1132 break; 1133 1134 case Builtin::BI__sync_fetch_and_xor: 1135 case Builtin::BI__sync_fetch_and_xor_1: 1136 case Builtin::BI__sync_fetch_and_xor_2: 1137 case Builtin::BI__sync_fetch_and_xor_4: 1138 case Builtin::BI__sync_fetch_and_xor_8: 1139 case Builtin::BI__sync_fetch_and_xor_16: 1140 BuiltinIndex = 4; 1141 break; 1142 1143 case Builtin::BI__sync_add_and_fetch: 1144 case Builtin::BI__sync_add_and_fetch_1: 1145 case Builtin::BI__sync_add_and_fetch_2: 1146 case Builtin::BI__sync_add_and_fetch_4: 1147 case Builtin::BI__sync_add_and_fetch_8: 1148 case Builtin::BI__sync_add_and_fetch_16: 1149 BuiltinIndex = 5; 1150 break; 1151 1152 case Builtin::BI__sync_sub_and_fetch: 1153 case Builtin::BI__sync_sub_and_fetch_1: 1154 case Builtin::BI__sync_sub_and_fetch_2: 1155 case Builtin::BI__sync_sub_and_fetch_4: 1156 case Builtin::BI__sync_sub_and_fetch_8: 1157 case Builtin::BI__sync_sub_and_fetch_16: 1158 BuiltinIndex = 6; 1159 break; 1160 1161 case Builtin::BI__sync_and_and_fetch: 1162 case Builtin::BI__sync_and_and_fetch_1: 1163 case Builtin::BI__sync_and_and_fetch_2: 1164 case Builtin::BI__sync_and_and_fetch_4: 1165 case Builtin::BI__sync_and_and_fetch_8: 1166 case Builtin::BI__sync_and_and_fetch_16: 1167 BuiltinIndex = 7; 1168 break; 1169 1170 case Builtin::BI__sync_or_and_fetch: 1171 case Builtin::BI__sync_or_and_fetch_1: 1172 case Builtin::BI__sync_or_and_fetch_2: 1173 case Builtin::BI__sync_or_and_fetch_4: 1174 case Builtin::BI__sync_or_and_fetch_8: 1175 case Builtin::BI__sync_or_and_fetch_16: 1176 BuiltinIndex = 8; 1177 break; 1178 1179 case Builtin::BI__sync_xor_and_fetch: 1180 case Builtin::BI__sync_xor_and_fetch_1: 1181 case Builtin::BI__sync_xor_and_fetch_2: 1182 case Builtin::BI__sync_xor_and_fetch_4: 1183 case Builtin::BI__sync_xor_and_fetch_8: 1184 case Builtin::BI__sync_xor_and_fetch_16: 1185 BuiltinIndex = 9; 1186 break; 1187 1188 case Builtin::BI__sync_val_compare_and_swap: 1189 case Builtin::BI__sync_val_compare_and_swap_1: 1190 case Builtin::BI__sync_val_compare_and_swap_2: 1191 case Builtin::BI__sync_val_compare_and_swap_4: 1192 case Builtin::BI__sync_val_compare_and_swap_8: 1193 case Builtin::BI__sync_val_compare_and_swap_16: 1194 BuiltinIndex = 10; 1195 NumFixed = 2; 1196 break; 1197 1198 case Builtin::BI__sync_bool_compare_and_swap: 1199 case Builtin::BI__sync_bool_compare_and_swap_1: 1200 case Builtin::BI__sync_bool_compare_and_swap_2: 1201 case Builtin::BI__sync_bool_compare_and_swap_4: 1202 case Builtin::BI__sync_bool_compare_and_swap_8: 1203 case Builtin::BI__sync_bool_compare_and_swap_16: 1204 BuiltinIndex = 11; 1205 NumFixed = 2; 1206 ResultType = Context.BoolTy; 1207 break; 1208 1209 case Builtin::BI__sync_lock_test_and_set: 1210 case Builtin::BI__sync_lock_test_and_set_1: 1211 case Builtin::BI__sync_lock_test_and_set_2: 1212 case Builtin::BI__sync_lock_test_and_set_4: 1213 case Builtin::BI__sync_lock_test_and_set_8: 1214 case Builtin::BI__sync_lock_test_and_set_16: 1215 BuiltinIndex = 12; 1216 break; 1217 1218 case Builtin::BI__sync_lock_release: 1219 case Builtin::BI__sync_lock_release_1: 1220 case Builtin::BI__sync_lock_release_2: 1221 case Builtin::BI__sync_lock_release_4: 1222 case Builtin::BI__sync_lock_release_8: 1223 case Builtin::BI__sync_lock_release_16: 1224 BuiltinIndex = 13; 1225 NumFixed = 0; 1226 ResultType = Context.VoidTy; 1227 break; 1228 1229 case Builtin::BI__sync_swap: 1230 case Builtin::BI__sync_swap_1: 1231 case Builtin::BI__sync_swap_2: 1232 case Builtin::BI__sync_swap_4: 1233 case Builtin::BI__sync_swap_8: 1234 case Builtin::BI__sync_swap_16: 1235 BuiltinIndex = 14; 1236 break; 1237 } 1238 1239 // Now that we know how many fixed arguments we expect, first check that we 1240 // have at least that many. 1241 if (TheCall->getNumArgs() < 1+NumFixed) { 1242 Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args_at_least) 1243 << 0 << 1+NumFixed << TheCall->getNumArgs() 1244 << TheCall->getCallee()->getSourceRange(); 1245 return ExprError(); 1246 } 1247 1248 // Get the decl for the concrete builtin from this, we can tell what the 1249 // concrete integer type we should convert to is. 1250 unsigned NewBuiltinID = BuiltinIndices[BuiltinIndex][SizeIndex]; 1251 const char *NewBuiltinName = Context.BuiltinInfo.GetName(NewBuiltinID); 1252 FunctionDecl *NewBuiltinDecl; 1253 if (NewBuiltinID == BuiltinID) 1254 NewBuiltinDecl = FDecl; 1255 else { 1256 // Perform builtin lookup to avoid redeclaring it. 1257 DeclarationName DN(&Context.Idents.get(NewBuiltinName)); 1258 LookupResult Res(*this, DN, DRE->getLocStart(), LookupOrdinaryName); 1259 LookupName(Res, TUScope, /*AllowBuiltinCreation=*/true); 1260 assert(Res.getFoundDecl()); 1261 NewBuiltinDecl = dyn_cast<FunctionDecl>(Res.getFoundDecl()); 1262 if (NewBuiltinDecl == 0) 1263 return ExprError(); 1264 } 1265 1266 // The first argument --- the pointer --- has a fixed type; we 1267 // deduce the types of the rest of the arguments accordingly. Walk 1268 // the remaining arguments, converting them to the deduced value type. 1269 for (unsigned i = 0; i != NumFixed; ++i) { 1270 ExprResult Arg = TheCall->getArg(i+1); 1271 1272 // GCC does an implicit conversion to the pointer or integer ValType. This 1273 // can fail in some cases (1i -> int**), check for this error case now. 1274 // Initialize the argument. 1275 InitializedEntity Entity = InitializedEntity::InitializeParameter(Context, 1276 ValType, /*consume*/ false); 1277 Arg = PerformCopyInitialization(Entity, SourceLocation(), Arg); 1278 if (Arg.isInvalid()) 1279 return ExprError(); 1280 1281 // Okay, we have something that *can* be converted to the right type. Check 1282 // to see if there is a potentially weird extension going on here. This can 1283 // happen when you do an atomic operation on something like an char* and 1284 // pass in 42. The 42 gets converted to char. This is even more strange 1285 // for things like 45.123 -> char, etc. 1286 // FIXME: Do this check. 1287 TheCall->setArg(i+1, Arg.take()); 1288 } 1289 1290 ASTContext& Context = this->getASTContext(); 1291 1292 // Create a new DeclRefExpr to refer to the new decl. 1293 DeclRefExpr* NewDRE = DeclRefExpr::Create( 1294 Context, 1295 DRE->getQualifierLoc(), 1296 SourceLocation(), 1297 NewBuiltinDecl, 1298 /*enclosing*/ false, 1299 DRE->getLocation(), 1300 Context.BuiltinFnTy, 1301 DRE->getValueKind()); 1302 1303 // Set the callee in the CallExpr. 1304 // FIXME: This loses syntactic information. 1305 QualType CalleePtrTy = Context.getPointerType(NewBuiltinDecl->getType()); 1306 ExprResult PromotedCall = ImpCastExprToType(NewDRE, CalleePtrTy, 1307 CK_BuiltinFnToFnPtr); 1308 TheCall->setCallee(PromotedCall.take()); 1309 1310 // Change the result type of the call to match the original value type. This 1311 // is arbitrary, but the codegen for these builtins ins design to handle it 1312 // gracefully. 1313 TheCall->setType(ResultType); 1314 1315 return TheCallResult; 1316} 1317 1318/// CheckObjCString - Checks that the argument to the builtin 1319/// CFString constructor is correct 1320/// Note: It might also make sense to do the UTF-16 conversion here (would 1321/// simplify the backend). 1322bool Sema::CheckObjCString(Expr *Arg) { 1323 Arg = Arg->IgnoreParenCasts(); 1324 StringLiteral *Literal = dyn_cast<StringLiteral>(Arg); 1325 1326 if (!Literal || !Literal->isAscii()) { 1327 Diag(Arg->getLocStart(), diag::err_cfstring_literal_not_string_constant) 1328 << Arg->getSourceRange(); 1329 return true; 1330 } 1331 1332 if (Literal->containsNonAsciiOrNull()) { 1333 StringRef String = Literal->getString(); 1334 unsigned NumBytes = String.size(); 1335 SmallVector<UTF16, 128> ToBuf(NumBytes); 1336 const UTF8 *FromPtr = (const UTF8 *)String.data(); 1337 UTF16 *ToPtr = &ToBuf[0]; 1338 1339 ConversionResult Result = ConvertUTF8toUTF16(&FromPtr, FromPtr + NumBytes, 1340 &ToPtr, ToPtr + NumBytes, 1341 strictConversion); 1342 // Check for conversion failure. 1343 if (Result != conversionOK) 1344 Diag(Arg->getLocStart(), 1345 diag::warn_cfstring_truncated) << Arg->getSourceRange(); 1346 } 1347 return false; 1348} 1349 1350/// SemaBuiltinVAStart - Check the arguments to __builtin_va_start for validity. 1351/// Emit an error and return true on failure, return false on success. 1352bool Sema::SemaBuiltinVAStart(CallExpr *TheCall) { 1353 Expr *Fn = TheCall->getCallee(); 1354 if (TheCall->getNumArgs() > 2) { 1355 Diag(TheCall->getArg(2)->getLocStart(), 1356 diag::err_typecheck_call_too_many_args) 1357 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 1358 << Fn->getSourceRange() 1359 << SourceRange(TheCall->getArg(2)->getLocStart(), 1360 (*(TheCall->arg_end()-1))->getLocEnd()); 1361 return true; 1362 } 1363 1364 if (TheCall->getNumArgs() < 2) { 1365 return Diag(TheCall->getLocEnd(), 1366 diag::err_typecheck_call_too_few_args_at_least) 1367 << 0 /*function call*/ << 2 << TheCall->getNumArgs(); 1368 } 1369 1370 // Type-check the first argument normally. 1371 if (checkBuiltinArgument(*this, TheCall, 0)) 1372 return true; 1373 1374 // Determine whether the current function is variadic or not. 1375 BlockScopeInfo *CurBlock = getCurBlock(); 1376 bool isVariadic; 1377 if (CurBlock) 1378 isVariadic = CurBlock->TheDecl->isVariadic(); 1379 else if (FunctionDecl *FD = getCurFunctionDecl()) 1380 isVariadic = FD->isVariadic(); 1381 else 1382 isVariadic = getCurMethodDecl()->isVariadic(); 1383 1384 if (!isVariadic) { 1385 Diag(Fn->getLocStart(), diag::err_va_start_used_in_non_variadic_function); 1386 return true; 1387 } 1388 1389 // Verify that the second argument to the builtin is the last argument of the 1390 // current function or method. 1391 bool SecondArgIsLastNamedArgument = false; 1392 const Expr *Arg = TheCall->getArg(1)->IgnoreParenCasts(); 1393 1394 // These are valid if SecondArgIsLastNamedArgument is false after the next 1395 // block. 1396 QualType Type; 1397 SourceLocation ParamLoc; 1398 1399 if (const DeclRefExpr *DR = dyn_cast<DeclRefExpr>(Arg)) { 1400 if (const ParmVarDecl *PV = dyn_cast<ParmVarDecl>(DR->getDecl())) { 1401 // FIXME: This isn't correct for methods (results in bogus warning). 1402 // Get the last formal in the current function. 1403 const ParmVarDecl *LastArg; 1404 if (CurBlock) 1405 LastArg = *(CurBlock->TheDecl->param_end()-1); 1406 else if (FunctionDecl *FD = getCurFunctionDecl()) 1407 LastArg = *(FD->param_end()-1); 1408 else 1409 LastArg = *(getCurMethodDecl()->param_end()-1); 1410 SecondArgIsLastNamedArgument = PV == LastArg; 1411 1412 Type = PV->getType(); 1413 ParamLoc = PV->getLocation(); 1414 } 1415 } 1416 1417 if (!SecondArgIsLastNamedArgument) 1418 Diag(TheCall->getArg(1)->getLocStart(), 1419 diag::warn_second_parameter_of_va_start_not_last_named_argument); 1420 else if (Type->isReferenceType()) { 1421 Diag(Arg->getLocStart(), 1422 diag::warn_va_start_of_reference_type_is_undefined); 1423 Diag(ParamLoc, diag::note_parameter_type) << Type; 1424 } 1425 1426 return false; 1427} 1428 1429/// SemaBuiltinUnorderedCompare - Handle functions like __builtin_isgreater and 1430/// friends. This is declared to take (...), so we have to check everything. 1431bool Sema::SemaBuiltinUnorderedCompare(CallExpr *TheCall) { 1432 if (TheCall->getNumArgs() < 2) 1433 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 1434 << 0 << 2 << TheCall->getNumArgs()/*function call*/; 1435 if (TheCall->getNumArgs() > 2) 1436 return Diag(TheCall->getArg(2)->getLocStart(), 1437 diag::err_typecheck_call_too_many_args) 1438 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 1439 << SourceRange(TheCall->getArg(2)->getLocStart(), 1440 (*(TheCall->arg_end()-1))->getLocEnd()); 1441 1442 ExprResult OrigArg0 = TheCall->getArg(0); 1443 ExprResult OrigArg1 = TheCall->getArg(1); 1444 1445 // Do standard promotions between the two arguments, returning their common 1446 // type. 1447 QualType Res = UsualArithmeticConversions(OrigArg0, OrigArg1, false); 1448 if (OrigArg0.isInvalid() || OrigArg1.isInvalid()) 1449 return true; 1450 1451 // Make sure any conversions are pushed back into the call; this is 1452 // type safe since unordered compare builtins are declared as "_Bool 1453 // foo(...)". 1454 TheCall->setArg(0, OrigArg0.get()); 1455 TheCall->setArg(1, OrigArg1.get()); 1456 1457 if (OrigArg0.get()->isTypeDependent() || OrigArg1.get()->isTypeDependent()) 1458 return false; 1459 1460 // If the common type isn't a real floating type, then the arguments were 1461 // invalid for this operation. 1462 if (Res.isNull() || !Res->isRealFloatingType()) 1463 return Diag(OrigArg0.get()->getLocStart(), 1464 diag::err_typecheck_call_invalid_ordered_compare) 1465 << OrigArg0.get()->getType() << OrigArg1.get()->getType() 1466 << SourceRange(OrigArg0.get()->getLocStart(), OrigArg1.get()->getLocEnd()); 1467 1468 return false; 1469} 1470 1471/// SemaBuiltinSemaBuiltinFPClassification - Handle functions like 1472/// __builtin_isnan and friends. This is declared to take (...), so we have 1473/// to check everything. We expect the last argument to be a floating point 1474/// value. 1475bool Sema::SemaBuiltinFPClassification(CallExpr *TheCall, unsigned NumArgs) { 1476 if (TheCall->getNumArgs() < NumArgs) 1477 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 1478 << 0 << NumArgs << TheCall->getNumArgs()/*function call*/; 1479 if (TheCall->getNumArgs() > NumArgs) 1480 return Diag(TheCall->getArg(NumArgs)->getLocStart(), 1481 diag::err_typecheck_call_too_many_args) 1482 << 0 /*function call*/ << NumArgs << TheCall->getNumArgs() 1483 << SourceRange(TheCall->getArg(NumArgs)->getLocStart(), 1484 (*(TheCall->arg_end()-1))->getLocEnd()); 1485 1486 Expr *OrigArg = TheCall->getArg(NumArgs-1); 1487 1488 if (OrigArg->isTypeDependent()) 1489 return false; 1490 1491 // This operation requires a non-_Complex floating-point number. 1492 if (!OrigArg->getType()->isRealFloatingType()) 1493 return Diag(OrigArg->getLocStart(), 1494 diag::err_typecheck_call_invalid_unary_fp) 1495 << OrigArg->getType() << OrigArg->getSourceRange(); 1496 1497 // If this is an implicit conversion from float -> double, remove it. 1498 if (ImplicitCastExpr *Cast = dyn_cast<ImplicitCastExpr>(OrigArg)) { 1499 Expr *CastArg = Cast->getSubExpr(); 1500 if (CastArg->getType()->isSpecificBuiltinType(BuiltinType::Float)) { 1501 assert(Cast->getType()->isSpecificBuiltinType(BuiltinType::Double) && 1502 "promotion from float to double is the only expected cast here"); 1503 Cast->setSubExpr(0); 1504 TheCall->setArg(NumArgs-1, CastArg); 1505 } 1506 } 1507 1508 return false; 1509} 1510 1511/// SemaBuiltinShuffleVector - Handle __builtin_shufflevector. 1512// This is declared to take (...), so we have to check everything. 1513ExprResult Sema::SemaBuiltinShuffleVector(CallExpr *TheCall) { 1514 if (TheCall->getNumArgs() < 2) 1515 return ExprError(Diag(TheCall->getLocEnd(), 1516 diag::err_typecheck_call_too_few_args_at_least) 1517 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 1518 << TheCall->getSourceRange()); 1519 1520 // Determine which of the following types of shufflevector we're checking: 1521 // 1) unary, vector mask: (lhs, mask) 1522 // 2) binary, vector mask: (lhs, rhs, mask) 1523 // 3) binary, scalar mask: (lhs, rhs, index, ..., index) 1524 QualType resType = TheCall->getArg(0)->getType(); 1525 unsigned numElements = 0; 1526 1527 if (!TheCall->getArg(0)->isTypeDependent() && 1528 !TheCall->getArg(1)->isTypeDependent()) { 1529 QualType LHSType = TheCall->getArg(0)->getType(); 1530 QualType RHSType = TheCall->getArg(1)->getType(); 1531 1532 if (!LHSType->isVectorType() || !RHSType->isVectorType()) { 1533 Diag(TheCall->getLocStart(), diag::err_shufflevector_non_vector) 1534 << SourceRange(TheCall->getArg(0)->getLocStart(), 1535 TheCall->getArg(1)->getLocEnd()); 1536 return ExprError(); 1537 } 1538 1539 numElements = LHSType->getAs<VectorType>()->getNumElements(); 1540 unsigned numResElements = TheCall->getNumArgs() - 2; 1541 1542 // Check to see if we have a call with 2 vector arguments, the unary shuffle 1543 // with mask. If so, verify that RHS is an integer vector type with the 1544 // same number of elts as lhs. 1545 if (TheCall->getNumArgs() == 2) { 1546 if (!RHSType->hasIntegerRepresentation() || 1547 RHSType->getAs<VectorType>()->getNumElements() != numElements) 1548 Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector) 1549 << SourceRange(TheCall->getArg(1)->getLocStart(), 1550 TheCall->getArg(1)->getLocEnd()); 1551 } 1552 else if (!Context.hasSameUnqualifiedType(LHSType, RHSType)) { 1553 Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector) 1554 << SourceRange(TheCall->getArg(0)->getLocStart(), 1555 TheCall->getArg(1)->getLocEnd()); 1556 return ExprError(); 1557 } else if (numElements != numResElements) { 1558 QualType eltType = LHSType->getAs<VectorType>()->getElementType(); 1559 resType = Context.getVectorType(eltType, numResElements, 1560 VectorType::GenericVector); 1561 } 1562 } 1563 1564 for (unsigned i = 2; i < TheCall->getNumArgs(); i++) { 1565 if (TheCall->getArg(i)->isTypeDependent() || 1566 TheCall->getArg(i)->isValueDependent()) 1567 continue; 1568 1569 llvm::APSInt Result(32); 1570 if (!TheCall->getArg(i)->isIntegerConstantExpr(Result, Context)) 1571 return ExprError(Diag(TheCall->getLocStart(), 1572 diag::err_shufflevector_nonconstant_argument) 1573 << TheCall->getArg(i)->getSourceRange()); 1574 1575 if (Result.getActiveBits() > 64 || Result.getZExtValue() >= numElements*2) 1576 return ExprError(Diag(TheCall->getLocStart(), 1577 diag::err_shufflevector_argument_too_large) 1578 << TheCall->getArg(i)->getSourceRange()); 1579 } 1580 1581 SmallVector<Expr*, 32> exprs; 1582 1583 for (unsigned i = 0, e = TheCall->getNumArgs(); i != e; i++) { 1584 exprs.push_back(TheCall->getArg(i)); 1585 TheCall->setArg(i, 0); 1586 } 1587 1588 return Owned(new (Context) ShuffleVectorExpr(Context, exprs, resType, 1589 TheCall->getCallee()->getLocStart(), 1590 TheCall->getRParenLoc())); 1591} 1592 1593/// SemaBuiltinPrefetch - Handle __builtin_prefetch. 1594// This is declared to take (const void*, ...) and can take two 1595// optional constant int args. 1596bool Sema::SemaBuiltinPrefetch(CallExpr *TheCall) { 1597 unsigned NumArgs = TheCall->getNumArgs(); 1598 1599 if (NumArgs > 3) 1600 return Diag(TheCall->getLocEnd(), 1601 diag::err_typecheck_call_too_many_args_at_most) 1602 << 0 /*function call*/ << 3 << NumArgs 1603 << TheCall->getSourceRange(); 1604 1605 // Argument 0 is checked for us and the remaining arguments must be 1606 // constant integers. 1607 for (unsigned i = 1; i != NumArgs; ++i) { 1608 Expr *Arg = TheCall->getArg(i); 1609 1610 // We can't check the value of a dependent argument. 1611 if (Arg->isTypeDependent() || Arg->isValueDependent()) 1612 continue; 1613 1614 llvm::APSInt Result; 1615 if (SemaBuiltinConstantArg(TheCall, i, Result)) 1616 return true; 1617 1618 // FIXME: gcc issues a warning and rewrites these to 0. These 1619 // seems especially odd for the third argument since the default 1620 // is 3. 1621 if (i == 1) { 1622 if (Result.getLimitedValue() > 1) 1623 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 1624 << "0" << "1" << Arg->getSourceRange(); 1625 } else { 1626 if (Result.getLimitedValue() > 3) 1627 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 1628 << "0" << "3" << Arg->getSourceRange(); 1629 } 1630 } 1631 1632 return false; 1633} 1634 1635/// SemaBuiltinConstantArg - Handle a check if argument ArgNum of CallExpr 1636/// TheCall is a constant expression. 1637bool Sema::SemaBuiltinConstantArg(CallExpr *TheCall, int ArgNum, 1638 llvm::APSInt &Result) { 1639 Expr *Arg = TheCall->getArg(ArgNum); 1640 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 1641 FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl()); 1642 1643 if (Arg->isTypeDependent() || Arg->isValueDependent()) return false; 1644 1645 if (!Arg->isIntegerConstantExpr(Result, Context)) 1646 return Diag(TheCall->getLocStart(), diag::err_constant_integer_arg_type) 1647 << FDecl->getDeclName() << Arg->getSourceRange(); 1648 1649 return false; 1650} 1651 1652/// SemaBuiltinObjectSize - Handle __builtin_object_size(void *ptr, 1653/// int type). This simply type checks that type is one of the defined 1654/// constants (0-3). 1655// For compatibility check 0-3, llvm only handles 0 and 2. 1656bool Sema::SemaBuiltinObjectSize(CallExpr *TheCall) { 1657 llvm::APSInt Result; 1658 1659 // We can't check the value of a dependent argument. 1660 if (TheCall->getArg(1)->isTypeDependent() || 1661 TheCall->getArg(1)->isValueDependent()) 1662 return false; 1663 1664 // Check constant-ness first. 1665 if (SemaBuiltinConstantArg(TheCall, 1, Result)) 1666 return true; 1667 1668 Expr *Arg = TheCall->getArg(1); 1669 if (Result.getSExtValue() < 0 || Result.getSExtValue() > 3) { 1670 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 1671 << "0" << "3" << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 1672 } 1673 1674 return false; 1675} 1676 1677/// SemaBuiltinLongjmp - Handle __builtin_longjmp(void *env[5], int val). 1678/// This checks that val is a constant 1. 1679bool Sema::SemaBuiltinLongjmp(CallExpr *TheCall) { 1680 Expr *Arg = TheCall->getArg(1); 1681 llvm::APSInt Result; 1682 1683 // TODO: This is less than ideal. Overload this to take a value. 1684 if (SemaBuiltinConstantArg(TheCall, 1, Result)) 1685 return true; 1686 1687 if (Result != 1) 1688 return Diag(TheCall->getLocStart(), diag::err_builtin_longjmp_invalid_val) 1689 << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 1690 1691 return false; 1692} 1693 1694// Determine if an expression is a string literal or constant string. 1695// If this function returns false on the arguments to a function expecting a 1696// format string, we will usually need to emit a warning. 1697// True string literals are then checked by CheckFormatString. 1698Sema::StringLiteralCheckType 1699Sema::checkFormatStringExpr(const Expr *E, ArrayRef<const Expr *> Args, 1700 bool HasVAListArg, 1701 unsigned format_idx, unsigned firstDataArg, 1702 FormatStringType Type, VariadicCallType CallType, 1703 bool inFunctionCall) { 1704 tryAgain: 1705 if (E->isTypeDependent() || E->isValueDependent()) 1706 return SLCT_NotALiteral; 1707 1708 E = E->IgnoreParenCasts(); 1709 1710 if (E->isNullPointerConstant(Context, Expr::NPC_ValueDependentIsNotNull)) 1711 // Technically -Wformat-nonliteral does not warn about this case. 1712 // The behavior of printf and friends in this case is implementation 1713 // dependent. Ideally if the format string cannot be null then 1714 // it should have a 'nonnull' attribute in the function prototype. 1715 return SLCT_CheckedLiteral; 1716 1717 switch (E->getStmtClass()) { 1718 case Stmt::BinaryConditionalOperatorClass: 1719 case Stmt::ConditionalOperatorClass: { 1720 // The expression is a literal if both sub-expressions were, and it was 1721 // completely checked only if both sub-expressions were checked. 1722 const AbstractConditionalOperator *C = 1723 cast<AbstractConditionalOperator>(E); 1724 StringLiteralCheckType Left = 1725 checkFormatStringExpr(C->getTrueExpr(), Args, 1726 HasVAListArg, format_idx, firstDataArg, 1727 Type, CallType, inFunctionCall); 1728 if (Left == SLCT_NotALiteral) 1729 return SLCT_NotALiteral; 1730 StringLiteralCheckType Right = 1731 checkFormatStringExpr(C->getFalseExpr(), Args, 1732 HasVAListArg, format_idx, firstDataArg, 1733 Type, CallType, inFunctionCall); 1734 return Left < Right ? Left : Right; 1735 } 1736 1737 case Stmt::ImplicitCastExprClass: { 1738 E = cast<ImplicitCastExpr>(E)->getSubExpr(); 1739 goto tryAgain; 1740 } 1741 1742 case Stmt::OpaqueValueExprClass: 1743 if (const Expr *src = cast<OpaqueValueExpr>(E)->getSourceExpr()) { 1744 E = src; 1745 goto tryAgain; 1746 } 1747 return SLCT_NotALiteral; 1748 1749 case Stmt::PredefinedExprClass: 1750 // While __func__, etc., are technically not string literals, they 1751 // cannot contain format specifiers and thus are not a security 1752 // liability. 1753 return SLCT_UncheckedLiteral; 1754 1755 case Stmt::DeclRefExprClass: { 1756 const DeclRefExpr *DR = cast<DeclRefExpr>(E); 1757 1758 // As an exception, do not flag errors for variables binding to 1759 // const string literals. 1760 if (const VarDecl *VD = dyn_cast<VarDecl>(DR->getDecl())) { 1761 bool isConstant = false; 1762 QualType T = DR->getType(); 1763 1764 if (const ArrayType *AT = Context.getAsArrayType(T)) { 1765 isConstant = AT->getElementType().isConstant(Context); 1766 } else if (const PointerType *PT = T->getAs<PointerType>()) { 1767 isConstant = T.isConstant(Context) && 1768 PT->getPointeeType().isConstant(Context); 1769 } else if (T->isObjCObjectPointerType()) { 1770 // In ObjC, there is usually no "const ObjectPointer" type, 1771 // so don't check if the pointee type is constant. 1772 isConstant = T.isConstant(Context); 1773 } 1774 1775 if (isConstant) { 1776 if (const Expr *Init = VD->getAnyInitializer()) { 1777 // Look through initializers like const char c[] = { "foo" } 1778 if (const InitListExpr *InitList = dyn_cast<InitListExpr>(Init)) { 1779 if (InitList->isStringLiteralInit()) 1780 Init = InitList->getInit(0)->IgnoreParenImpCasts(); 1781 } 1782 return checkFormatStringExpr(Init, Args, 1783 HasVAListArg, format_idx, 1784 firstDataArg, Type, CallType, 1785 /*inFunctionCall*/false); 1786 } 1787 } 1788 1789 // For vprintf* functions (i.e., HasVAListArg==true), we add a 1790 // special check to see if the format string is a function parameter 1791 // of the function calling the printf function. If the function 1792 // has an attribute indicating it is a printf-like function, then we 1793 // should suppress warnings concerning non-literals being used in a call 1794 // to a vprintf function. For example: 1795 // 1796 // void 1797 // logmessage(char const *fmt __attribute__ (format (printf, 1, 2)), ...){ 1798 // va_list ap; 1799 // va_start(ap, fmt); 1800 // vprintf(fmt, ap); // Do NOT emit a warning about "fmt". 1801 // ... 1802 // 1803 if (HasVAListArg) { 1804 if (const ParmVarDecl *PV = dyn_cast<ParmVarDecl>(VD)) { 1805 if (const NamedDecl *ND = dyn_cast<NamedDecl>(PV->getDeclContext())) { 1806 int PVIndex = PV->getFunctionScopeIndex() + 1; 1807 for (specific_attr_iterator<FormatAttr> 1808 i = ND->specific_attr_begin<FormatAttr>(), 1809 e = ND->specific_attr_end<FormatAttr>(); i != e ; ++i) { 1810 FormatAttr *PVFormat = *i; 1811 // adjust for implicit parameter 1812 if (const CXXMethodDecl *MD = dyn_cast<CXXMethodDecl>(ND)) 1813 if (MD->isInstance()) 1814 ++PVIndex; 1815 // We also check if the formats are compatible. 1816 // We can't pass a 'scanf' string to a 'printf' function. 1817 if (PVIndex == PVFormat->getFormatIdx() && 1818 Type == GetFormatStringType(PVFormat)) 1819 return SLCT_UncheckedLiteral; 1820 } 1821 } 1822 } 1823 } 1824 } 1825 1826 return SLCT_NotALiteral; 1827 } 1828 1829 case Stmt::CallExprClass: 1830 case Stmt::CXXMemberCallExprClass: { 1831 const CallExpr *CE = cast<CallExpr>(E); 1832 if (const NamedDecl *ND = dyn_cast_or_null<NamedDecl>(CE->getCalleeDecl())) { 1833 if (const FormatArgAttr *FA = ND->getAttr<FormatArgAttr>()) { 1834 unsigned ArgIndex = FA->getFormatIdx(); 1835 if (const CXXMethodDecl *MD = dyn_cast<CXXMethodDecl>(ND)) 1836 if (MD->isInstance()) 1837 --ArgIndex; 1838 const Expr *Arg = CE->getArg(ArgIndex - 1); 1839 1840 return checkFormatStringExpr(Arg, Args, 1841 HasVAListArg, format_idx, firstDataArg, 1842 Type, CallType, inFunctionCall); 1843 } else if (const FunctionDecl *FD = dyn_cast<FunctionDecl>(ND)) { 1844 unsigned BuiltinID = FD->getBuiltinID(); 1845 if (BuiltinID == Builtin::BI__builtin___CFStringMakeConstantString || 1846 BuiltinID == Builtin::BI__builtin___NSStringMakeConstantString) { 1847 const Expr *Arg = CE->getArg(0); 1848 return checkFormatStringExpr(Arg, Args, 1849 HasVAListArg, format_idx, 1850 firstDataArg, Type, CallType, 1851 inFunctionCall); 1852 } 1853 } 1854 } 1855 1856 return SLCT_NotALiteral; 1857 } 1858 case Stmt::ObjCStringLiteralClass: 1859 case Stmt::StringLiteralClass: { 1860 const StringLiteral *StrE = NULL; 1861 1862 if (const ObjCStringLiteral *ObjCFExpr = dyn_cast<ObjCStringLiteral>(E)) 1863 StrE = ObjCFExpr->getString(); 1864 else 1865 StrE = cast<StringLiteral>(E); 1866 1867 if (StrE) { 1868 CheckFormatString(StrE, E, Args, HasVAListArg, format_idx, 1869 firstDataArg, Type, inFunctionCall, CallType); 1870 return SLCT_CheckedLiteral; 1871 } 1872 1873 return SLCT_NotALiteral; 1874 } 1875 1876 default: 1877 return SLCT_NotALiteral; 1878 } 1879} 1880 1881void 1882Sema::CheckNonNullArguments(const NonNullAttr *NonNull, 1883 const Expr * const *ExprArgs, 1884 SourceLocation CallSiteLoc) { 1885 for (NonNullAttr::args_iterator i = NonNull->args_begin(), 1886 e = NonNull->args_end(); 1887 i != e; ++i) { 1888 const Expr *ArgExpr = ExprArgs[*i]; 1889 1890 // As a special case, transparent unions initialized with zero are 1891 // considered null for the purposes of the nonnull attribute. 1892 if (const RecordType *UT = ArgExpr->getType()->getAsUnionType()) { 1893 if (UT->getDecl()->hasAttr<TransparentUnionAttr>()) 1894 if (const CompoundLiteralExpr *CLE = 1895 dyn_cast<CompoundLiteralExpr>(ArgExpr)) 1896 if (const InitListExpr *ILE = 1897 dyn_cast<InitListExpr>(CLE->getInitializer())) 1898 ArgExpr = ILE->getInit(0); 1899 } 1900 1901 bool Result; 1902 if (ArgExpr->EvaluateAsBooleanCondition(Result, Context) && !Result) 1903 Diag(CallSiteLoc, diag::warn_null_arg) << ArgExpr->getSourceRange(); 1904 } 1905} 1906 1907Sema::FormatStringType Sema::GetFormatStringType(const FormatAttr *Format) { 1908 return llvm::StringSwitch<FormatStringType>(Format->getType()) 1909 .Case("scanf", FST_Scanf) 1910 .Cases("printf", "printf0", FST_Printf) 1911 .Cases("NSString", "CFString", FST_NSString) 1912 .Case("strftime", FST_Strftime) 1913 .Case("strfmon", FST_Strfmon) 1914 .Cases("kprintf", "cmn_err", "vcmn_err", "zcmn_err", FST_Kprintf) 1915 .Default(FST_Unknown); 1916} 1917 1918/// CheckFormatArguments - Check calls to printf and scanf (and similar 1919/// functions) for correct use of format strings. 1920/// Returns true if a format string has been fully checked. 1921bool Sema::CheckFormatArguments(const FormatAttr *Format, 1922 ArrayRef<const Expr *> Args, 1923 bool IsCXXMember, 1924 VariadicCallType CallType, 1925 SourceLocation Loc, SourceRange Range) { 1926 FormatStringInfo FSI; 1927 if (getFormatStringInfo(Format, IsCXXMember, &FSI)) 1928 return CheckFormatArguments(Args, FSI.HasVAListArg, FSI.FormatIdx, 1929 FSI.FirstDataArg, GetFormatStringType(Format), 1930 CallType, Loc, Range); 1931 return false; 1932} 1933 1934bool Sema::CheckFormatArguments(ArrayRef<const Expr *> Args, 1935 bool HasVAListArg, unsigned format_idx, 1936 unsigned firstDataArg, FormatStringType Type, 1937 VariadicCallType CallType, 1938 SourceLocation Loc, SourceRange Range) { 1939 // CHECK: printf/scanf-like function is called with no format string. 1940 if (format_idx >= Args.size()) { 1941 Diag(Loc, diag::warn_missing_format_string) << Range; 1942 return false; 1943 } 1944 1945 const Expr *OrigFormatExpr = Args[format_idx]->IgnoreParenCasts(); 1946 1947 // CHECK: format string is not a string literal. 1948 // 1949 // Dynamically generated format strings are difficult to 1950 // automatically vet at compile time. Requiring that format strings 1951 // are string literals: (1) permits the checking of format strings by 1952 // the compiler and thereby (2) can practically remove the source of 1953 // many format string exploits. 1954 1955 // Format string can be either ObjC string (e.g. @"%d") or 1956 // C string (e.g. "%d") 1957 // ObjC string uses the same format specifiers as C string, so we can use 1958 // the same format string checking logic for both ObjC and C strings. 1959 StringLiteralCheckType CT = 1960 checkFormatStringExpr(OrigFormatExpr, Args, HasVAListArg, 1961 format_idx, firstDataArg, Type, CallType); 1962 if (CT != SLCT_NotALiteral) 1963 // Literal format string found, check done! 1964 return CT == SLCT_CheckedLiteral; 1965 1966 // Strftime is particular as it always uses a single 'time' argument, 1967 // so it is safe to pass a non-literal string. 1968 if (Type == FST_Strftime) 1969 return false; 1970 1971 // Do not emit diag when the string param is a macro expansion and the 1972 // format is either NSString or CFString. This is a hack to prevent 1973 // diag when using the NSLocalizedString and CFCopyLocalizedString macros 1974 // which are usually used in place of NS and CF string literals. 1975 if (Type == FST_NSString && 1976 SourceMgr.isInSystemMacro(Args[format_idx]->getLocStart())) 1977 return false; 1978 1979 // If there are no arguments specified, warn with -Wformat-security, otherwise 1980 // warn only with -Wformat-nonliteral. 1981 if (Args.size() == firstDataArg) 1982 Diag(Args[format_idx]->getLocStart(), 1983 diag::warn_format_nonliteral_noargs) 1984 << OrigFormatExpr->getSourceRange(); 1985 else 1986 Diag(Args[format_idx]->getLocStart(), 1987 diag::warn_format_nonliteral) 1988 << OrigFormatExpr->getSourceRange(); 1989 return false; 1990} 1991 1992namespace { 1993class CheckFormatHandler : public analyze_format_string::FormatStringHandler { 1994protected: 1995 Sema &S; 1996 const StringLiteral *FExpr; 1997 const Expr *OrigFormatExpr; 1998 const unsigned FirstDataArg; 1999 const unsigned NumDataArgs; 2000 const char *Beg; // Start of format string. 2001 const bool HasVAListArg; 2002 ArrayRef<const Expr *> Args; 2003 unsigned FormatIdx; 2004 llvm::BitVector CoveredArgs; 2005 bool usesPositionalArgs; 2006 bool atFirstArg; 2007 bool inFunctionCall; 2008 Sema::VariadicCallType CallType; 2009public: 2010 CheckFormatHandler(Sema &s, const StringLiteral *fexpr, 2011 const Expr *origFormatExpr, unsigned firstDataArg, 2012 unsigned numDataArgs, const char *beg, bool hasVAListArg, 2013 ArrayRef<const Expr *> Args, 2014 unsigned formatIdx, bool inFunctionCall, 2015 Sema::VariadicCallType callType) 2016 : S(s), FExpr(fexpr), OrigFormatExpr(origFormatExpr), 2017 FirstDataArg(firstDataArg), NumDataArgs(numDataArgs), 2018 Beg(beg), HasVAListArg(hasVAListArg), 2019 Args(Args), FormatIdx(formatIdx), 2020 usesPositionalArgs(false), atFirstArg(true), 2021 inFunctionCall(inFunctionCall), CallType(callType) { 2022 CoveredArgs.resize(numDataArgs); 2023 CoveredArgs.reset(); 2024 } 2025 2026 void DoneProcessing(); 2027 2028 void HandleIncompleteSpecifier(const char *startSpecifier, 2029 unsigned specifierLen); 2030 2031 void HandleInvalidLengthModifier( 2032 const analyze_format_string::FormatSpecifier &FS, 2033 const analyze_format_string::ConversionSpecifier &CS, 2034 const char *startSpecifier, unsigned specifierLen, unsigned DiagID); 2035 2036 void HandleNonStandardLengthModifier( 2037 const analyze_format_string::FormatSpecifier &FS, 2038 const char *startSpecifier, unsigned specifierLen); 2039 2040 void HandleNonStandardConversionSpecifier( 2041 const analyze_format_string::ConversionSpecifier &CS, 2042 const char *startSpecifier, unsigned specifierLen); 2043 2044 virtual void HandlePosition(const char *startPos, unsigned posLen); 2045 2046 virtual void HandleInvalidPosition(const char *startSpecifier, 2047 unsigned specifierLen, 2048 analyze_format_string::PositionContext p); 2049 2050 virtual void HandleZeroPosition(const char *startPos, unsigned posLen); 2051 2052 void HandleNullChar(const char *nullCharacter); 2053 2054 template <typename Range> 2055 static void EmitFormatDiagnostic(Sema &S, bool inFunctionCall, 2056 const Expr *ArgumentExpr, 2057 PartialDiagnostic PDiag, 2058 SourceLocation StringLoc, 2059 bool IsStringLocation, Range StringRange, 2060 ArrayRef<FixItHint> Fixit = None); 2061 2062protected: 2063 bool HandleInvalidConversionSpecifier(unsigned argIndex, SourceLocation Loc, 2064 const char *startSpec, 2065 unsigned specifierLen, 2066 const char *csStart, unsigned csLen); 2067 2068 void HandlePositionalNonpositionalArgs(SourceLocation Loc, 2069 const char *startSpec, 2070 unsigned specifierLen); 2071 2072 SourceRange getFormatStringRange(); 2073 CharSourceRange getSpecifierRange(const char *startSpecifier, 2074 unsigned specifierLen); 2075 SourceLocation getLocationOfByte(const char *x); 2076 2077 const Expr *getDataArg(unsigned i) const; 2078 2079 bool CheckNumArgs(const analyze_format_string::FormatSpecifier &FS, 2080 const analyze_format_string::ConversionSpecifier &CS, 2081 const char *startSpecifier, unsigned specifierLen, 2082 unsigned argIndex); 2083 2084 template <typename Range> 2085 void EmitFormatDiagnostic(PartialDiagnostic PDiag, SourceLocation StringLoc, 2086 bool IsStringLocation, Range StringRange, 2087 ArrayRef<FixItHint> Fixit = None); 2088 2089 void CheckPositionalAndNonpositionalArgs( 2090 const analyze_format_string::FormatSpecifier *FS); 2091}; 2092} 2093 2094SourceRange CheckFormatHandler::getFormatStringRange() { 2095 return OrigFormatExpr->getSourceRange(); 2096} 2097 2098CharSourceRange CheckFormatHandler:: 2099getSpecifierRange(const char *startSpecifier, unsigned specifierLen) { 2100 SourceLocation Start = getLocationOfByte(startSpecifier); 2101 SourceLocation End = getLocationOfByte(startSpecifier + specifierLen - 1); 2102 2103 // Advance the end SourceLocation by one due to half-open ranges. 2104 End = End.getLocWithOffset(1); 2105 2106 return CharSourceRange::getCharRange(Start, End); 2107} 2108 2109SourceLocation CheckFormatHandler::getLocationOfByte(const char *x) { 2110 return S.getLocationOfStringLiteralByte(FExpr, x - Beg); 2111} 2112 2113void CheckFormatHandler::HandleIncompleteSpecifier(const char *startSpecifier, 2114 unsigned specifierLen){ 2115 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_incomplete_specifier), 2116 getLocationOfByte(startSpecifier), 2117 /*IsStringLocation*/true, 2118 getSpecifierRange(startSpecifier, specifierLen)); 2119} 2120 2121void CheckFormatHandler::HandleInvalidLengthModifier( 2122 const analyze_format_string::FormatSpecifier &FS, 2123 const analyze_format_string::ConversionSpecifier &CS, 2124 const char *startSpecifier, unsigned specifierLen, unsigned DiagID) { 2125 using namespace analyze_format_string; 2126 2127 const LengthModifier &LM = FS.getLengthModifier(); 2128 CharSourceRange LMRange = getSpecifierRange(LM.getStart(), LM.getLength()); 2129 2130 // See if we know how to fix this length modifier. 2131 Optional<LengthModifier> FixedLM = FS.getCorrectedLengthModifier(); 2132 if (FixedLM) { 2133 EmitFormatDiagnostic(S.PDiag(DiagID) << LM.toString() << CS.toString(), 2134 getLocationOfByte(LM.getStart()), 2135 /*IsStringLocation*/true, 2136 getSpecifierRange(startSpecifier, specifierLen)); 2137 2138 S.Diag(getLocationOfByte(LM.getStart()), diag::note_format_fix_specifier) 2139 << FixedLM->toString() 2140 << FixItHint::CreateReplacement(LMRange, FixedLM->toString()); 2141 2142 } else { 2143 FixItHint Hint; 2144 if (DiagID == diag::warn_format_nonsensical_length) 2145 Hint = FixItHint::CreateRemoval(LMRange); 2146 2147 EmitFormatDiagnostic(S.PDiag(DiagID) << LM.toString() << CS.toString(), 2148 getLocationOfByte(LM.getStart()), 2149 /*IsStringLocation*/true, 2150 getSpecifierRange(startSpecifier, specifierLen), 2151 Hint); 2152 } 2153} 2154 2155void CheckFormatHandler::HandleNonStandardLengthModifier( 2156 const analyze_format_string::FormatSpecifier &FS, 2157 const char *startSpecifier, unsigned specifierLen) { 2158 using namespace analyze_format_string; 2159 2160 const LengthModifier &LM = FS.getLengthModifier(); 2161 CharSourceRange LMRange = getSpecifierRange(LM.getStart(), LM.getLength()); 2162 2163 // See if we know how to fix this length modifier. 2164 Optional<LengthModifier> FixedLM = FS.getCorrectedLengthModifier(); 2165 if (FixedLM) { 2166 EmitFormatDiagnostic(S.PDiag(diag::warn_format_non_standard) 2167 << LM.toString() << 0, 2168 getLocationOfByte(LM.getStart()), 2169 /*IsStringLocation*/true, 2170 getSpecifierRange(startSpecifier, specifierLen)); 2171 2172 S.Diag(getLocationOfByte(LM.getStart()), diag::note_format_fix_specifier) 2173 << FixedLM->toString() 2174 << FixItHint::CreateReplacement(LMRange, FixedLM->toString()); 2175 2176 } else { 2177 EmitFormatDiagnostic(S.PDiag(diag::warn_format_non_standard) 2178 << LM.toString() << 0, 2179 getLocationOfByte(LM.getStart()), 2180 /*IsStringLocation*/true, 2181 getSpecifierRange(startSpecifier, specifierLen)); 2182 } 2183} 2184 2185void CheckFormatHandler::HandleNonStandardConversionSpecifier( 2186 const analyze_format_string::ConversionSpecifier &CS, 2187 const char *startSpecifier, unsigned specifierLen) { 2188 using namespace analyze_format_string; 2189 2190 // See if we know how to fix this conversion specifier. 2191 Optional<ConversionSpecifier> FixedCS = CS.getStandardSpecifier(); 2192 if (FixedCS) { 2193 EmitFormatDiagnostic(S.PDiag(diag::warn_format_non_standard) 2194 << CS.toString() << /*conversion specifier*/1, 2195 getLocationOfByte(CS.getStart()), 2196 /*IsStringLocation*/true, 2197 getSpecifierRange(startSpecifier, specifierLen)); 2198 2199 CharSourceRange CSRange = getSpecifierRange(CS.getStart(), CS.getLength()); 2200 S.Diag(getLocationOfByte(CS.getStart()), diag::note_format_fix_specifier) 2201 << FixedCS->toString() 2202 << FixItHint::CreateReplacement(CSRange, FixedCS->toString()); 2203 } else { 2204 EmitFormatDiagnostic(S.PDiag(diag::warn_format_non_standard) 2205 << CS.toString() << /*conversion specifier*/1, 2206 getLocationOfByte(CS.getStart()), 2207 /*IsStringLocation*/true, 2208 getSpecifierRange(startSpecifier, specifierLen)); 2209 } 2210} 2211 2212void CheckFormatHandler::HandlePosition(const char *startPos, 2213 unsigned posLen) { 2214 EmitFormatDiagnostic(S.PDiag(diag::warn_format_non_standard_positional_arg), 2215 getLocationOfByte(startPos), 2216 /*IsStringLocation*/true, 2217 getSpecifierRange(startPos, posLen)); 2218} 2219 2220void 2221CheckFormatHandler::HandleInvalidPosition(const char *startPos, unsigned posLen, 2222 analyze_format_string::PositionContext p) { 2223 EmitFormatDiagnostic(S.PDiag(diag::warn_format_invalid_positional_specifier) 2224 << (unsigned) p, 2225 getLocationOfByte(startPos), /*IsStringLocation*/true, 2226 getSpecifierRange(startPos, posLen)); 2227} 2228 2229void CheckFormatHandler::HandleZeroPosition(const char *startPos, 2230 unsigned posLen) { 2231 EmitFormatDiagnostic(S.PDiag(diag::warn_format_zero_positional_specifier), 2232 getLocationOfByte(startPos), 2233 /*IsStringLocation*/true, 2234 getSpecifierRange(startPos, posLen)); 2235} 2236 2237void CheckFormatHandler::HandleNullChar(const char *nullCharacter) { 2238 if (!isa<ObjCStringLiteral>(OrigFormatExpr)) { 2239 // The presence of a null character is likely an error. 2240 EmitFormatDiagnostic( 2241 S.PDiag(diag::warn_printf_format_string_contains_null_char), 2242 getLocationOfByte(nullCharacter), /*IsStringLocation*/true, 2243 getFormatStringRange()); 2244 } 2245} 2246 2247// Note that this may return NULL if there was an error parsing or building 2248// one of the argument expressions. 2249const Expr *CheckFormatHandler::getDataArg(unsigned i) const { 2250 return Args[FirstDataArg + i]; 2251} 2252 2253void CheckFormatHandler::DoneProcessing() { 2254 // Does the number of data arguments exceed the number of 2255 // format conversions in the format string? 2256 if (!HasVAListArg) { 2257 // Find any arguments that weren't covered. 2258 CoveredArgs.flip(); 2259 signed notCoveredArg = CoveredArgs.find_first(); 2260 if (notCoveredArg >= 0) { 2261 assert((unsigned)notCoveredArg < NumDataArgs); 2262 if (const Expr *E = getDataArg((unsigned) notCoveredArg)) { 2263 SourceLocation Loc = E->getLocStart(); 2264 if (!S.getSourceManager().isInSystemMacro(Loc)) { 2265 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_data_arg_not_used), 2266 Loc, /*IsStringLocation*/false, 2267 getFormatStringRange()); 2268 } 2269 } 2270 } 2271 } 2272} 2273 2274bool 2275CheckFormatHandler::HandleInvalidConversionSpecifier(unsigned argIndex, 2276 SourceLocation Loc, 2277 const char *startSpec, 2278 unsigned specifierLen, 2279 const char *csStart, 2280 unsigned csLen) { 2281 2282 bool keepGoing = true; 2283 if (argIndex < NumDataArgs) { 2284 // Consider the argument coverered, even though the specifier doesn't 2285 // make sense. 2286 CoveredArgs.set(argIndex); 2287 } 2288 else { 2289 // If argIndex exceeds the number of data arguments we 2290 // don't issue a warning because that is just a cascade of warnings (and 2291 // they may have intended '%%' anyway). We don't want to continue processing 2292 // the format string after this point, however, as we will like just get 2293 // gibberish when trying to match arguments. 2294 keepGoing = false; 2295 } 2296 2297 EmitFormatDiagnostic(S.PDiag(diag::warn_format_invalid_conversion) 2298 << StringRef(csStart, csLen), 2299 Loc, /*IsStringLocation*/true, 2300 getSpecifierRange(startSpec, specifierLen)); 2301 2302 return keepGoing; 2303} 2304 2305void 2306CheckFormatHandler::HandlePositionalNonpositionalArgs(SourceLocation Loc, 2307 const char *startSpec, 2308 unsigned specifierLen) { 2309 EmitFormatDiagnostic( 2310 S.PDiag(diag::warn_format_mix_positional_nonpositional_args), 2311 Loc, /*isStringLoc*/true, getSpecifierRange(startSpec, specifierLen)); 2312} 2313 2314bool 2315CheckFormatHandler::CheckNumArgs( 2316 const analyze_format_string::FormatSpecifier &FS, 2317 const analyze_format_string::ConversionSpecifier &CS, 2318 const char *startSpecifier, unsigned specifierLen, unsigned argIndex) { 2319 2320 if (argIndex >= NumDataArgs) { 2321 PartialDiagnostic PDiag = FS.usesPositionalArg() 2322 ? (S.PDiag(diag::warn_printf_positional_arg_exceeds_data_args) 2323 << (argIndex+1) << NumDataArgs) 2324 : S.PDiag(diag::warn_printf_insufficient_data_args); 2325 EmitFormatDiagnostic( 2326 PDiag, getLocationOfByte(CS.getStart()), /*IsStringLocation*/true, 2327 getSpecifierRange(startSpecifier, specifierLen)); 2328 return false; 2329 } 2330 return true; 2331} 2332 2333template<typename Range> 2334void CheckFormatHandler::EmitFormatDiagnostic(PartialDiagnostic PDiag, 2335 SourceLocation Loc, 2336 bool IsStringLocation, 2337 Range StringRange, 2338 ArrayRef<FixItHint> FixIt) { 2339 EmitFormatDiagnostic(S, inFunctionCall, Args[FormatIdx], PDiag, 2340 Loc, IsStringLocation, StringRange, FixIt); 2341} 2342 2343/// \brief If the format string is not within the funcion call, emit a note 2344/// so that the function call and string are in diagnostic messages. 2345/// 2346/// \param InFunctionCall if true, the format string is within the function 2347/// call and only one diagnostic message will be produced. Otherwise, an 2348/// extra note will be emitted pointing to location of the format string. 2349/// 2350/// \param ArgumentExpr the expression that is passed as the format string 2351/// argument in the function call. Used for getting locations when two 2352/// diagnostics are emitted. 2353/// 2354/// \param PDiag the callee should already have provided any strings for the 2355/// diagnostic message. This function only adds locations and fixits 2356/// to diagnostics. 2357/// 2358/// \param Loc primary location for diagnostic. If two diagnostics are 2359/// required, one will be at Loc and a new SourceLocation will be created for 2360/// the other one. 2361/// 2362/// \param IsStringLocation if true, Loc points to the format string should be 2363/// used for the note. Otherwise, Loc points to the argument list and will 2364/// be used with PDiag. 2365/// 2366/// \param StringRange some or all of the string to highlight. This is 2367/// templated so it can accept either a CharSourceRange or a SourceRange. 2368/// 2369/// \param FixIt optional fix it hint for the format string. 2370template<typename Range> 2371void CheckFormatHandler::EmitFormatDiagnostic(Sema &S, bool InFunctionCall, 2372 const Expr *ArgumentExpr, 2373 PartialDiagnostic PDiag, 2374 SourceLocation Loc, 2375 bool IsStringLocation, 2376 Range StringRange, 2377 ArrayRef<FixItHint> FixIt) { 2378 if (InFunctionCall) { 2379 const Sema::SemaDiagnosticBuilder &D = S.Diag(Loc, PDiag); 2380 D << StringRange; 2381 for (ArrayRef<FixItHint>::iterator I = FixIt.begin(), E = FixIt.end(); 2382 I != E; ++I) { 2383 D << *I; 2384 } 2385 } else { 2386 S.Diag(IsStringLocation ? ArgumentExpr->getExprLoc() : Loc, PDiag) 2387 << ArgumentExpr->getSourceRange(); 2388 2389 const Sema::SemaDiagnosticBuilder &Note = 2390 S.Diag(IsStringLocation ? Loc : StringRange.getBegin(), 2391 diag::note_format_string_defined); 2392 2393 Note << StringRange; 2394 for (ArrayRef<FixItHint>::iterator I = FixIt.begin(), E = FixIt.end(); 2395 I != E; ++I) { 2396 Note << *I; 2397 } 2398 } 2399} 2400 2401//===--- CHECK: Printf format string checking ------------------------------===// 2402 2403namespace { 2404class CheckPrintfHandler : public CheckFormatHandler { 2405 bool ObjCContext; 2406public: 2407 CheckPrintfHandler(Sema &s, const StringLiteral *fexpr, 2408 const Expr *origFormatExpr, unsigned firstDataArg, 2409 unsigned numDataArgs, bool isObjC, 2410 const char *beg, bool hasVAListArg, 2411 ArrayRef<const Expr *> Args, 2412 unsigned formatIdx, bool inFunctionCall, 2413 Sema::VariadicCallType CallType) 2414 : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg, 2415 numDataArgs, beg, hasVAListArg, Args, 2416 formatIdx, inFunctionCall, CallType), ObjCContext(isObjC) 2417 {} 2418 2419 2420 bool HandleInvalidPrintfConversionSpecifier( 2421 const analyze_printf::PrintfSpecifier &FS, 2422 const char *startSpecifier, 2423 unsigned specifierLen); 2424 2425 bool HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier &FS, 2426 const char *startSpecifier, 2427 unsigned specifierLen); 2428 bool checkFormatExpr(const analyze_printf::PrintfSpecifier &FS, 2429 const char *StartSpecifier, 2430 unsigned SpecifierLen, 2431 const Expr *E); 2432 2433 bool HandleAmount(const analyze_format_string::OptionalAmount &Amt, unsigned k, 2434 const char *startSpecifier, unsigned specifierLen); 2435 void HandleInvalidAmount(const analyze_printf::PrintfSpecifier &FS, 2436 const analyze_printf::OptionalAmount &Amt, 2437 unsigned type, 2438 const char *startSpecifier, unsigned specifierLen); 2439 void HandleFlag(const analyze_printf::PrintfSpecifier &FS, 2440 const analyze_printf::OptionalFlag &flag, 2441 const char *startSpecifier, unsigned specifierLen); 2442 void HandleIgnoredFlag(const analyze_printf::PrintfSpecifier &FS, 2443 const analyze_printf::OptionalFlag &ignoredFlag, 2444 const analyze_printf::OptionalFlag &flag, 2445 const char *startSpecifier, unsigned specifierLen); 2446 bool checkForCStrMembers(const analyze_printf::ArgType &AT, 2447 const Expr *E, const CharSourceRange &CSR); 2448 2449}; 2450} 2451 2452bool CheckPrintfHandler::HandleInvalidPrintfConversionSpecifier( 2453 const analyze_printf::PrintfSpecifier &FS, 2454 const char *startSpecifier, 2455 unsigned specifierLen) { 2456 const analyze_printf::PrintfConversionSpecifier &CS = 2457 FS.getConversionSpecifier(); 2458 2459 return HandleInvalidConversionSpecifier(FS.getArgIndex(), 2460 getLocationOfByte(CS.getStart()), 2461 startSpecifier, specifierLen, 2462 CS.getStart(), CS.getLength()); 2463} 2464 2465bool CheckPrintfHandler::HandleAmount( 2466 const analyze_format_string::OptionalAmount &Amt, 2467 unsigned k, const char *startSpecifier, 2468 unsigned specifierLen) { 2469 2470 if (Amt.hasDataArgument()) { 2471 if (!HasVAListArg) { 2472 unsigned argIndex = Amt.getArgIndex(); 2473 if (argIndex >= NumDataArgs) { 2474 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_asterisk_missing_arg) 2475 << k, 2476 getLocationOfByte(Amt.getStart()), 2477 /*IsStringLocation*/true, 2478 getSpecifierRange(startSpecifier, specifierLen)); 2479 // Don't do any more checking. We will just emit 2480 // spurious errors. 2481 return false; 2482 } 2483 2484 // Type check the data argument. It should be an 'int'. 2485 // Although not in conformance with C99, we also allow the argument to be 2486 // an 'unsigned int' as that is a reasonably safe case. GCC also 2487 // doesn't emit a warning for that case. 2488 CoveredArgs.set(argIndex); 2489 const Expr *Arg = getDataArg(argIndex); 2490 if (!Arg) 2491 return false; 2492 2493 QualType T = Arg->getType(); 2494 2495 const analyze_printf::ArgType &AT = Amt.getArgType(S.Context); 2496 assert(AT.isValid()); 2497 2498 if (!AT.matchesType(S.Context, T)) { 2499 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_asterisk_wrong_type) 2500 << k << AT.getRepresentativeTypeName(S.Context) 2501 << T << Arg->getSourceRange(), 2502 getLocationOfByte(Amt.getStart()), 2503 /*IsStringLocation*/true, 2504 getSpecifierRange(startSpecifier, specifierLen)); 2505 // Don't do any more checking. We will just emit 2506 // spurious errors. 2507 return false; 2508 } 2509 } 2510 } 2511 return true; 2512} 2513 2514void CheckPrintfHandler::HandleInvalidAmount( 2515 const analyze_printf::PrintfSpecifier &FS, 2516 const analyze_printf::OptionalAmount &Amt, 2517 unsigned type, 2518 const char *startSpecifier, 2519 unsigned specifierLen) { 2520 const analyze_printf::PrintfConversionSpecifier &CS = 2521 FS.getConversionSpecifier(); 2522 2523 FixItHint fixit = 2524 Amt.getHowSpecified() == analyze_printf::OptionalAmount::Constant 2525 ? FixItHint::CreateRemoval(getSpecifierRange(Amt.getStart(), 2526 Amt.getConstantLength())) 2527 : FixItHint(); 2528 2529 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_nonsensical_optional_amount) 2530 << type << CS.toString(), 2531 getLocationOfByte(Amt.getStart()), 2532 /*IsStringLocation*/true, 2533 getSpecifierRange(startSpecifier, specifierLen), 2534 fixit); 2535} 2536 2537void CheckPrintfHandler::HandleFlag(const analyze_printf::PrintfSpecifier &FS, 2538 const analyze_printf::OptionalFlag &flag, 2539 const char *startSpecifier, 2540 unsigned specifierLen) { 2541 // Warn about pointless flag with a fixit removal. 2542 const analyze_printf::PrintfConversionSpecifier &CS = 2543 FS.getConversionSpecifier(); 2544 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_nonsensical_flag) 2545 << flag.toString() << CS.toString(), 2546 getLocationOfByte(flag.getPosition()), 2547 /*IsStringLocation*/true, 2548 getSpecifierRange(startSpecifier, specifierLen), 2549 FixItHint::CreateRemoval( 2550 getSpecifierRange(flag.getPosition(), 1))); 2551} 2552 2553void CheckPrintfHandler::HandleIgnoredFlag( 2554 const analyze_printf::PrintfSpecifier &FS, 2555 const analyze_printf::OptionalFlag &ignoredFlag, 2556 const analyze_printf::OptionalFlag &flag, 2557 const char *startSpecifier, 2558 unsigned specifierLen) { 2559 // Warn about ignored flag with a fixit removal. 2560 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_ignored_flag) 2561 << ignoredFlag.toString() << flag.toString(), 2562 getLocationOfByte(ignoredFlag.getPosition()), 2563 /*IsStringLocation*/true, 2564 getSpecifierRange(startSpecifier, specifierLen), 2565 FixItHint::CreateRemoval( 2566 getSpecifierRange(ignoredFlag.getPosition(), 1))); 2567} 2568 2569// Determines if the specified is a C++ class or struct containing 2570// a member with the specified name and kind (e.g. a CXXMethodDecl named 2571// "c_str()"). 2572template<typename MemberKind> 2573static llvm::SmallPtrSet<MemberKind*, 1> 2574CXXRecordMembersNamed(StringRef Name, Sema &S, QualType Ty) { 2575 const RecordType *RT = Ty->getAs<RecordType>(); 2576 llvm::SmallPtrSet<MemberKind*, 1> Results; 2577 2578 if (!RT) 2579 return Results; 2580 const CXXRecordDecl *RD = dyn_cast<CXXRecordDecl>(RT->getDecl()); 2581 if (!RD) 2582 return Results; 2583 2584 LookupResult R(S, &S.PP.getIdentifierTable().get(Name), SourceLocation(), 2585 Sema::LookupMemberName); 2586 2587 // We just need to include all members of the right kind turned up by the 2588 // filter, at this point. 2589 if (S.LookupQualifiedName(R, RT->getDecl())) 2590 for (LookupResult::iterator I = R.begin(), E = R.end(); I != E; ++I) { 2591 NamedDecl *decl = (*I)->getUnderlyingDecl(); 2592 if (MemberKind *FK = dyn_cast<MemberKind>(decl)) 2593 Results.insert(FK); 2594 } 2595 return Results; 2596} 2597 2598// Check if a (w)string was passed when a (w)char* was needed, and offer a 2599// better diagnostic if so. AT is assumed to be valid. 2600// Returns true when a c_str() conversion method is found. 2601bool CheckPrintfHandler::checkForCStrMembers( 2602 const analyze_printf::ArgType &AT, const Expr *E, 2603 const CharSourceRange &CSR) { 2604 typedef llvm::SmallPtrSet<CXXMethodDecl*, 1> MethodSet; 2605 2606 MethodSet Results = 2607 CXXRecordMembersNamed<CXXMethodDecl>("c_str", S, E->getType()); 2608 2609 for (MethodSet::iterator MI = Results.begin(), ME = Results.end(); 2610 MI != ME; ++MI) { 2611 const CXXMethodDecl *Method = *MI; 2612 if (Method->getNumParams() == 0 && 2613 AT.matchesType(S.Context, Method->getResultType())) { 2614 // FIXME: Suggest parens if the expression needs them. 2615 SourceLocation EndLoc = 2616 S.getPreprocessor().getLocForEndOfToken(E->getLocEnd()); 2617 S.Diag(E->getLocStart(), diag::note_printf_c_str) 2618 << "c_str()" 2619 << FixItHint::CreateInsertion(EndLoc, ".c_str()"); 2620 return true; 2621 } 2622 } 2623 2624 return false; 2625} 2626 2627bool 2628CheckPrintfHandler::HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier 2629 &FS, 2630 const char *startSpecifier, 2631 unsigned specifierLen) { 2632 2633 using namespace analyze_format_string; 2634 using namespace analyze_printf; 2635 const PrintfConversionSpecifier &CS = FS.getConversionSpecifier(); 2636 2637 if (FS.consumesDataArgument()) { 2638 if (atFirstArg) { 2639 atFirstArg = false; 2640 usesPositionalArgs = FS.usesPositionalArg(); 2641 } 2642 else if (usesPositionalArgs != FS.usesPositionalArg()) { 2643 HandlePositionalNonpositionalArgs(getLocationOfByte(CS.getStart()), 2644 startSpecifier, specifierLen); 2645 return false; 2646 } 2647 } 2648 2649 // First check if the field width, precision, and conversion specifier 2650 // have matching data arguments. 2651 if (!HandleAmount(FS.getFieldWidth(), /* field width */ 0, 2652 startSpecifier, specifierLen)) { 2653 return false; 2654 } 2655 2656 if (!HandleAmount(FS.getPrecision(), /* precision */ 1, 2657 startSpecifier, specifierLen)) { 2658 return false; 2659 } 2660 2661 if (!CS.consumesDataArgument()) { 2662 // FIXME: Technically specifying a precision or field width here 2663 // makes no sense. Worth issuing a warning at some point. 2664 return true; 2665 } 2666 2667 // Consume the argument. 2668 unsigned argIndex = FS.getArgIndex(); 2669 if (argIndex < NumDataArgs) { 2670 // The check to see if the argIndex is valid will come later. 2671 // We set the bit here because we may exit early from this 2672 // function if we encounter some other error. 2673 CoveredArgs.set(argIndex); 2674 } 2675 2676 // Check for using an Objective-C specific conversion specifier 2677 // in a non-ObjC literal. 2678 if (!ObjCContext && CS.isObjCArg()) { 2679 return HandleInvalidPrintfConversionSpecifier(FS, startSpecifier, 2680 specifierLen); 2681 } 2682 2683 // Check for invalid use of field width 2684 if (!FS.hasValidFieldWidth()) { 2685 HandleInvalidAmount(FS, FS.getFieldWidth(), /* field width */ 0, 2686 startSpecifier, specifierLen); 2687 } 2688 2689 // Check for invalid use of precision 2690 if (!FS.hasValidPrecision()) { 2691 HandleInvalidAmount(FS, FS.getPrecision(), /* precision */ 1, 2692 startSpecifier, specifierLen); 2693 } 2694 2695 // Check each flag does not conflict with any other component. 2696 if (!FS.hasValidThousandsGroupingPrefix()) 2697 HandleFlag(FS, FS.hasThousandsGrouping(), startSpecifier, specifierLen); 2698 if (!FS.hasValidLeadingZeros()) 2699 HandleFlag(FS, FS.hasLeadingZeros(), startSpecifier, specifierLen); 2700 if (!FS.hasValidPlusPrefix()) 2701 HandleFlag(FS, FS.hasPlusPrefix(), startSpecifier, specifierLen); 2702 if (!FS.hasValidSpacePrefix()) 2703 HandleFlag(FS, FS.hasSpacePrefix(), startSpecifier, specifierLen); 2704 if (!FS.hasValidAlternativeForm()) 2705 HandleFlag(FS, FS.hasAlternativeForm(), startSpecifier, specifierLen); 2706 if (!FS.hasValidLeftJustified()) 2707 HandleFlag(FS, FS.isLeftJustified(), startSpecifier, specifierLen); 2708 2709 // Check that flags are not ignored by another flag 2710 if (FS.hasSpacePrefix() && FS.hasPlusPrefix()) // ' ' ignored by '+' 2711 HandleIgnoredFlag(FS, FS.hasSpacePrefix(), FS.hasPlusPrefix(), 2712 startSpecifier, specifierLen); 2713 if (FS.hasLeadingZeros() && FS.isLeftJustified()) // '0' ignored by '-' 2714 HandleIgnoredFlag(FS, FS.hasLeadingZeros(), FS.isLeftJustified(), 2715 startSpecifier, specifierLen); 2716 2717 // Check the length modifier is valid with the given conversion specifier. 2718 if (!FS.hasValidLengthModifier(S.getASTContext().getTargetInfo())) 2719 HandleInvalidLengthModifier(FS, CS, startSpecifier, specifierLen, 2720 diag::warn_format_nonsensical_length); 2721 else if (!FS.hasStandardLengthModifier()) 2722 HandleNonStandardLengthModifier(FS, startSpecifier, specifierLen); 2723 else if (!FS.hasStandardLengthConversionCombination()) 2724 HandleInvalidLengthModifier(FS, CS, startSpecifier, specifierLen, 2725 diag::warn_format_non_standard_conversion_spec); 2726 2727 if (!FS.hasStandardConversionSpecifier(S.getLangOpts())) 2728 HandleNonStandardConversionSpecifier(CS, startSpecifier, specifierLen); 2729 2730 // The remaining checks depend on the data arguments. 2731 if (HasVAListArg) 2732 return true; 2733 2734 if (!CheckNumArgs(FS, CS, startSpecifier, specifierLen, argIndex)) 2735 return false; 2736 2737 const Expr *Arg = getDataArg(argIndex); 2738 if (!Arg) 2739 return true; 2740 2741 return checkFormatExpr(FS, startSpecifier, specifierLen, Arg); 2742} 2743 2744static bool requiresParensToAddCast(const Expr *E) { 2745 // FIXME: We should have a general way to reason about operator 2746 // precedence and whether parens are actually needed here. 2747 // Take care of a few common cases where they aren't. 2748 const Expr *Inside = E->IgnoreImpCasts(); 2749 if (const PseudoObjectExpr *POE = dyn_cast<PseudoObjectExpr>(Inside)) 2750 Inside = POE->getSyntacticForm()->IgnoreImpCasts(); 2751 2752 switch (Inside->getStmtClass()) { 2753 case Stmt::ArraySubscriptExprClass: 2754 case Stmt::CallExprClass: 2755 case Stmt::CharacterLiteralClass: 2756 case Stmt::CXXBoolLiteralExprClass: 2757 case Stmt::DeclRefExprClass: 2758 case Stmt::FloatingLiteralClass: 2759 case Stmt::IntegerLiteralClass: 2760 case Stmt::MemberExprClass: 2761 case Stmt::ObjCArrayLiteralClass: 2762 case Stmt::ObjCBoolLiteralExprClass: 2763 case Stmt::ObjCBoxedExprClass: 2764 case Stmt::ObjCDictionaryLiteralClass: 2765 case Stmt::ObjCEncodeExprClass: 2766 case Stmt::ObjCIvarRefExprClass: 2767 case Stmt::ObjCMessageExprClass: 2768 case Stmt::ObjCPropertyRefExprClass: 2769 case Stmt::ObjCStringLiteralClass: 2770 case Stmt::ObjCSubscriptRefExprClass: 2771 case Stmt::ParenExprClass: 2772 case Stmt::StringLiteralClass: 2773 case Stmt::UnaryOperatorClass: 2774 return false; 2775 default: 2776 return true; 2777 } 2778} 2779 2780bool 2781CheckPrintfHandler::checkFormatExpr(const analyze_printf::PrintfSpecifier &FS, 2782 const char *StartSpecifier, 2783 unsigned SpecifierLen, 2784 const Expr *E) { 2785 using namespace analyze_format_string; 2786 using namespace analyze_printf; 2787 // Now type check the data expression that matches the 2788 // format specifier. 2789 const analyze_printf::ArgType &AT = FS.getArgType(S.Context, 2790 ObjCContext); 2791 if (!AT.isValid()) 2792 return true; 2793 2794 QualType ExprTy = E->getType(); 2795 while (const TypeOfExprType *TET = dyn_cast<TypeOfExprType>(ExprTy)) { 2796 ExprTy = TET->getUnderlyingExpr()->getType(); 2797 } 2798 2799 if (AT.matchesType(S.Context, ExprTy)) 2800 return true; 2801 2802 // Look through argument promotions for our error message's reported type. 2803 // This includes the integral and floating promotions, but excludes array 2804 // and function pointer decay; seeing that an argument intended to be a 2805 // string has type 'char [6]' is probably more confusing than 'char *'. 2806 if (const ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(E)) { 2807 if (ICE->getCastKind() == CK_IntegralCast || 2808 ICE->getCastKind() == CK_FloatingCast) { 2809 E = ICE->getSubExpr(); 2810 ExprTy = E->getType(); 2811 2812 // Check if we didn't match because of an implicit cast from a 'char' 2813 // or 'short' to an 'int'. This is done because printf is a varargs 2814 // function. 2815 if (ICE->getType() == S.Context.IntTy || 2816 ICE->getType() == S.Context.UnsignedIntTy) { 2817 // All further checking is done on the subexpression. 2818 if (AT.matchesType(S.Context, ExprTy)) 2819 return true; 2820 } 2821 } 2822 } else if (const CharacterLiteral *CL = dyn_cast<CharacterLiteral>(E)) { 2823 // Special case for 'a', which has type 'int' in C. 2824 // Note, however, that we do /not/ want to treat multibyte constants like 2825 // 'MooV' as characters! This form is deprecated but still exists. 2826 if (ExprTy == S.Context.IntTy) 2827 if (llvm::isUIntN(S.Context.getCharWidth(), CL->getValue())) 2828 ExprTy = S.Context.CharTy; 2829 } 2830 2831 // %C in an Objective-C context prints a unichar, not a wchar_t. 2832 // If the argument is an integer of some kind, believe the %C and suggest 2833 // a cast instead of changing the conversion specifier. 2834 QualType IntendedTy = ExprTy; 2835 if (ObjCContext && 2836 FS.getConversionSpecifier().getKind() == ConversionSpecifier::CArg) { 2837 if (ExprTy->isIntegralOrUnscopedEnumerationType() && 2838 !ExprTy->isCharType()) { 2839 // 'unichar' is defined as a typedef of unsigned short, but we should 2840 // prefer using the typedef if it is visible. 2841 IntendedTy = S.Context.UnsignedShortTy; 2842 2843 LookupResult Result(S, &S.Context.Idents.get("unichar"), E->getLocStart(), 2844 Sema::LookupOrdinaryName); 2845 if (S.LookupName(Result, S.getCurScope())) { 2846 NamedDecl *ND = Result.getFoundDecl(); 2847 if (TypedefNameDecl *TD = dyn_cast<TypedefNameDecl>(ND)) 2848 if (TD->getUnderlyingType() == IntendedTy) 2849 IntendedTy = S.Context.getTypedefType(TD); 2850 } 2851 } 2852 } 2853 2854 // Special-case some of Darwin's platform-independence types by suggesting 2855 // casts to primitive types that are known to be large enough. 2856 bool ShouldNotPrintDirectly = false; 2857 if (S.Context.getTargetInfo().getTriple().isOSDarwin()) { 2858 // Use a 'while' to peel off layers of typedefs. 2859 QualType TyTy = IntendedTy; 2860 while (const TypedefType *UserTy = TyTy->getAs<TypedefType>()) { 2861 StringRef Name = UserTy->getDecl()->getName(); 2862 QualType CastTy = llvm::StringSwitch<QualType>(Name) 2863 .Case("NSInteger", S.Context.LongTy) 2864 .Case("NSUInteger", S.Context.UnsignedLongTy) 2865 .Case("SInt32", S.Context.IntTy) 2866 .Case("UInt32", S.Context.UnsignedIntTy) 2867 .Default(QualType()); 2868 2869 if (!CastTy.isNull()) { 2870 ShouldNotPrintDirectly = true; 2871 IntendedTy = CastTy; 2872 break; 2873 } 2874 TyTy = UserTy->desugar(); 2875 } 2876 } 2877 2878 // We may be able to offer a FixItHint if it is a supported type. 2879 PrintfSpecifier fixedFS = FS; 2880 bool success = fixedFS.fixType(IntendedTy, S.getLangOpts(), 2881 S.Context, ObjCContext); 2882 2883 if (success) { 2884 // Get the fix string from the fixed format specifier 2885 SmallString<16> buf; 2886 llvm::raw_svector_ostream os(buf); 2887 fixedFS.toString(os); 2888 2889 CharSourceRange SpecRange = getSpecifierRange(StartSpecifier, SpecifierLen); 2890 2891 if (IntendedTy == ExprTy) { 2892 // In this case, the specifier is wrong and should be changed to match 2893 // the argument. 2894 EmitFormatDiagnostic( 2895 S.PDiag(diag::warn_printf_conversion_argument_type_mismatch) 2896 << AT.getRepresentativeTypeName(S.Context) << IntendedTy 2897 << E->getSourceRange(), 2898 E->getLocStart(), 2899 /*IsStringLocation*/false, 2900 SpecRange, 2901 FixItHint::CreateReplacement(SpecRange, os.str())); 2902 2903 } else { 2904 // The canonical type for formatting this value is different from the 2905 // actual type of the expression. (This occurs, for example, with Darwin's 2906 // NSInteger on 32-bit platforms, where it is typedef'd as 'int', but 2907 // should be printed as 'long' for 64-bit compatibility.) 2908 // Rather than emitting a normal format/argument mismatch, we want to 2909 // add a cast to the recommended type (and correct the format string 2910 // if necessary). 2911 SmallString<16> CastBuf; 2912 llvm::raw_svector_ostream CastFix(CastBuf); 2913 CastFix << "("; 2914 IntendedTy.print(CastFix, S.Context.getPrintingPolicy()); 2915 CastFix << ")"; 2916 2917 SmallVector<FixItHint,4> Hints; 2918 if (!AT.matchesType(S.Context, IntendedTy)) 2919 Hints.push_back(FixItHint::CreateReplacement(SpecRange, os.str())); 2920 2921 if (const CStyleCastExpr *CCast = dyn_cast<CStyleCastExpr>(E)) { 2922 // If there's already a cast present, just replace it. 2923 SourceRange CastRange(CCast->getLParenLoc(), CCast->getRParenLoc()); 2924 Hints.push_back(FixItHint::CreateReplacement(CastRange, CastFix.str())); 2925 2926 } else if (!requiresParensToAddCast(E)) { 2927 // If the expression has high enough precedence, 2928 // just write the C-style cast. 2929 Hints.push_back(FixItHint::CreateInsertion(E->getLocStart(), 2930 CastFix.str())); 2931 } else { 2932 // Otherwise, add parens around the expression as well as the cast. 2933 CastFix << "("; 2934 Hints.push_back(FixItHint::CreateInsertion(E->getLocStart(), 2935 CastFix.str())); 2936 2937 SourceLocation After = S.PP.getLocForEndOfToken(E->getLocEnd()); 2938 Hints.push_back(FixItHint::CreateInsertion(After, ")")); 2939 } 2940 2941 if (ShouldNotPrintDirectly) { 2942 // The expression has a type that should not be printed directly. 2943 // We extract the name from the typedef because we don't want to show 2944 // the underlying type in the diagnostic. 2945 StringRef Name = cast<TypedefType>(ExprTy)->getDecl()->getName(); 2946 2947 EmitFormatDiagnostic(S.PDiag(diag::warn_format_argument_needs_cast) 2948 << Name << IntendedTy 2949 << E->getSourceRange(), 2950 E->getLocStart(), /*IsStringLocation=*/false, 2951 SpecRange, Hints); 2952 } else { 2953 // In this case, the expression could be printed using a different 2954 // specifier, but we've decided that the specifier is probably correct 2955 // and we should cast instead. Just use the normal warning message. 2956 EmitFormatDiagnostic( 2957 S.PDiag(diag::warn_printf_conversion_argument_type_mismatch) 2958 << AT.getRepresentativeTypeName(S.Context) << ExprTy 2959 << E->getSourceRange(), 2960 E->getLocStart(), /*IsStringLocation*/false, 2961 SpecRange, Hints); 2962 } 2963 } 2964 } else { 2965 const CharSourceRange &CSR = getSpecifierRange(StartSpecifier, 2966 SpecifierLen); 2967 // Since the warning for passing non-POD types to variadic functions 2968 // was deferred until now, we emit a warning for non-POD 2969 // arguments here. 2970 if (S.isValidVarArgType(ExprTy) == Sema::VAK_Invalid) { 2971 unsigned DiagKind; 2972 if (ExprTy->isObjCObjectType()) 2973 DiagKind = diag::err_cannot_pass_objc_interface_to_vararg_format; 2974 else 2975 DiagKind = diag::warn_non_pod_vararg_with_format_string; 2976 2977 EmitFormatDiagnostic( 2978 S.PDiag(DiagKind) 2979 << S.getLangOpts().CPlusPlus11 2980 << ExprTy 2981 << CallType 2982 << AT.getRepresentativeTypeName(S.Context) 2983 << CSR 2984 << E->getSourceRange(), 2985 E->getLocStart(), /*IsStringLocation*/false, CSR); 2986 2987 checkForCStrMembers(AT, E, CSR); 2988 } else 2989 EmitFormatDiagnostic( 2990 S.PDiag(diag::warn_printf_conversion_argument_type_mismatch) 2991 << AT.getRepresentativeTypeName(S.Context) << ExprTy 2992 << CSR 2993 << E->getSourceRange(), 2994 E->getLocStart(), /*IsStringLocation*/false, CSR); 2995 } 2996 2997 return true; 2998} 2999 3000//===--- CHECK: Scanf format string checking ------------------------------===// 3001 3002namespace { 3003class CheckScanfHandler : public CheckFormatHandler { 3004public: 3005 CheckScanfHandler(Sema &s, const StringLiteral *fexpr, 3006 const Expr *origFormatExpr, unsigned firstDataArg, 3007 unsigned numDataArgs, const char *beg, bool hasVAListArg, 3008 ArrayRef<const Expr *> Args, 3009 unsigned formatIdx, bool inFunctionCall, 3010 Sema::VariadicCallType CallType) 3011 : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg, 3012 numDataArgs, beg, hasVAListArg, 3013 Args, formatIdx, inFunctionCall, CallType) 3014 {} 3015 3016 bool HandleScanfSpecifier(const analyze_scanf::ScanfSpecifier &FS, 3017 const char *startSpecifier, 3018 unsigned specifierLen); 3019 3020 bool HandleInvalidScanfConversionSpecifier( 3021 const analyze_scanf::ScanfSpecifier &FS, 3022 const char *startSpecifier, 3023 unsigned specifierLen); 3024 3025 void HandleIncompleteScanList(const char *start, const char *end); 3026}; 3027} 3028 3029void CheckScanfHandler::HandleIncompleteScanList(const char *start, 3030 const char *end) { 3031 EmitFormatDiagnostic(S.PDiag(diag::warn_scanf_scanlist_incomplete), 3032 getLocationOfByte(end), /*IsStringLocation*/true, 3033 getSpecifierRange(start, end - start)); 3034} 3035 3036bool CheckScanfHandler::HandleInvalidScanfConversionSpecifier( 3037 const analyze_scanf::ScanfSpecifier &FS, 3038 const char *startSpecifier, 3039 unsigned specifierLen) { 3040 3041 const analyze_scanf::ScanfConversionSpecifier &CS = 3042 FS.getConversionSpecifier(); 3043 3044 return HandleInvalidConversionSpecifier(FS.getArgIndex(), 3045 getLocationOfByte(CS.getStart()), 3046 startSpecifier, specifierLen, 3047 CS.getStart(), CS.getLength()); 3048} 3049 3050bool CheckScanfHandler::HandleScanfSpecifier( 3051 const analyze_scanf::ScanfSpecifier &FS, 3052 const char *startSpecifier, 3053 unsigned specifierLen) { 3054 3055 using namespace analyze_scanf; 3056 using namespace analyze_format_string; 3057 3058 const ScanfConversionSpecifier &CS = FS.getConversionSpecifier(); 3059 3060 // Handle case where '%' and '*' don't consume an argument. These shouldn't 3061 // be used to decide if we are using positional arguments consistently. 3062 if (FS.consumesDataArgument()) { 3063 if (atFirstArg) { 3064 atFirstArg = false; 3065 usesPositionalArgs = FS.usesPositionalArg(); 3066 } 3067 else if (usesPositionalArgs != FS.usesPositionalArg()) { 3068 HandlePositionalNonpositionalArgs(getLocationOfByte(CS.getStart()), 3069 startSpecifier, specifierLen); 3070 return false; 3071 } 3072 } 3073 3074 // Check if the field with is non-zero. 3075 const OptionalAmount &Amt = FS.getFieldWidth(); 3076 if (Amt.getHowSpecified() == OptionalAmount::Constant) { 3077 if (Amt.getConstantAmount() == 0) { 3078 const CharSourceRange &R = getSpecifierRange(Amt.getStart(), 3079 Amt.getConstantLength()); 3080 EmitFormatDiagnostic(S.PDiag(diag::warn_scanf_nonzero_width), 3081 getLocationOfByte(Amt.getStart()), 3082 /*IsStringLocation*/true, R, 3083 FixItHint::CreateRemoval(R)); 3084 } 3085 } 3086 3087 if (!FS.consumesDataArgument()) { 3088 // FIXME: Technically specifying a precision or field width here 3089 // makes no sense. Worth issuing a warning at some point. 3090 return true; 3091 } 3092 3093 // Consume the argument. 3094 unsigned argIndex = FS.getArgIndex(); 3095 if (argIndex < NumDataArgs) { 3096 // The check to see if the argIndex is valid will come later. 3097 // We set the bit here because we may exit early from this 3098 // function if we encounter some other error. 3099 CoveredArgs.set(argIndex); 3100 } 3101 3102 // Check the length modifier is valid with the given conversion specifier. 3103 if (!FS.hasValidLengthModifier(S.getASTContext().getTargetInfo())) 3104 HandleInvalidLengthModifier(FS, CS, startSpecifier, specifierLen, 3105 diag::warn_format_nonsensical_length); 3106 else if (!FS.hasStandardLengthModifier()) 3107 HandleNonStandardLengthModifier(FS, startSpecifier, specifierLen); 3108 else if (!FS.hasStandardLengthConversionCombination()) 3109 HandleInvalidLengthModifier(FS, CS, startSpecifier, specifierLen, 3110 diag::warn_format_non_standard_conversion_spec); 3111 3112 if (!FS.hasStandardConversionSpecifier(S.getLangOpts())) 3113 HandleNonStandardConversionSpecifier(CS, startSpecifier, specifierLen); 3114 3115 // The remaining checks depend on the data arguments. 3116 if (HasVAListArg) 3117 return true; 3118 3119 if (!CheckNumArgs(FS, CS, startSpecifier, specifierLen, argIndex)) 3120 return false; 3121 3122 // Check that the argument type matches the format specifier. 3123 const Expr *Ex = getDataArg(argIndex); 3124 if (!Ex) 3125 return true; 3126 3127 const analyze_format_string::ArgType &AT = FS.getArgType(S.Context); 3128 if (AT.isValid() && !AT.matchesType(S.Context, Ex->getType())) { 3129 ScanfSpecifier fixedFS = FS; 3130 bool success = fixedFS.fixType(Ex->getType(), S.getLangOpts(), 3131 S.Context); 3132 3133 if (success) { 3134 // Get the fix string from the fixed format specifier. 3135 SmallString<128> buf; 3136 llvm::raw_svector_ostream os(buf); 3137 fixedFS.toString(os); 3138 3139 EmitFormatDiagnostic( 3140 S.PDiag(diag::warn_printf_conversion_argument_type_mismatch) 3141 << AT.getRepresentativeTypeName(S.Context) << Ex->getType() 3142 << Ex->getSourceRange(), 3143 Ex->getLocStart(), 3144 /*IsStringLocation*/false, 3145 getSpecifierRange(startSpecifier, specifierLen), 3146 FixItHint::CreateReplacement( 3147 getSpecifierRange(startSpecifier, specifierLen), 3148 os.str())); 3149 } else { 3150 EmitFormatDiagnostic( 3151 S.PDiag(diag::warn_printf_conversion_argument_type_mismatch) 3152 << AT.getRepresentativeTypeName(S.Context) << Ex->getType() 3153 << Ex->getSourceRange(), 3154 Ex->getLocStart(), 3155 /*IsStringLocation*/false, 3156 getSpecifierRange(startSpecifier, specifierLen)); 3157 } 3158 } 3159 3160 return true; 3161} 3162 3163void Sema::CheckFormatString(const StringLiteral *FExpr, 3164 const Expr *OrigFormatExpr, 3165 ArrayRef<const Expr *> Args, 3166 bool HasVAListArg, unsigned format_idx, 3167 unsigned firstDataArg, FormatStringType Type, 3168 bool inFunctionCall, VariadicCallType CallType) { 3169 3170 // CHECK: is the format string a wide literal? 3171 if (!FExpr->isAscii() && !FExpr->isUTF8()) { 3172 CheckFormatHandler::EmitFormatDiagnostic( 3173 *this, inFunctionCall, Args[format_idx], 3174 PDiag(diag::warn_format_string_is_wide_literal), FExpr->getLocStart(), 3175 /*IsStringLocation*/true, OrigFormatExpr->getSourceRange()); 3176 return; 3177 } 3178 3179 // Str - The format string. NOTE: this is NOT null-terminated! 3180 StringRef StrRef = FExpr->getString(); 3181 const char *Str = StrRef.data(); 3182 unsigned StrLen = StrRef.size(); 3183 const unsigned numDataArgs = Args.size() - firstDataArg; 3184 3185 // CHECK: empty format string? 3186 if (StrLen == 0 && numDataArgs > 0) { 3187 CheckFormatHandler::EmitFormatDiagnostic( 3188 *this, inFunctionCall, Args[format_idx], 3189 PDiag(diag::warn_empty_format_string), FExpr->getLocStart(), 3190 /*IsStringLocation*/true, OrigFormatExpr->getSourceRange()); 3191 return; 3192 } 3193 3194 if (Type == FST_Printf || Type == FST_NSString) { 3195 CheckPrintfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg, 3196 numDataArgs, (Type == FST_NSString), 3197 Str, HasVAListArg, Args, format_idx, 3198 inFunctionCall, CallType); 3199 3200 if (!analyze_format_string::ParsePrintfString(H, Str, Str + StrLen, 3201 getLangOpts(), 3202 Context.getTargetInfo())) 3203 H.DoneProcessing(); 3204 } else if (Type == FST_Scanf) { 3205 CheckScanfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg, numDataArgs, 3206 Str, HasVAListArg, Args, format_idx, 3207 inFunctionCall, CallType); 3208 3209 if (!analyze_format_string::ParseScanfString(H, Str, Str + StrLen, 3210 getLangOpts(), 3211 Context.getTargetInfo())) 3212 H.DoneProcessing(); 3213 } // TODO: handle other formats 3214} 3215 3216//===--- CHECK: Standard memory functions ---------------------------------===// 3217 3218/// \brief Determine whether the given type is a dynamic class type (e.g., 3219/// whether it has a vtable). 3220static bool isDynamicClassType(QualType T) { 3221 if (CXXRecordDecl *Record = T->getAsCXXRecordDecl()) 3222 if (CXXRecordDecl *Definition = Record->getDefinition()) 3223 if (Definition->isDynamicClass()) 3224 return true; 3225 3226 return false; 3227} 3228 3229/// \brief If E is a sizeof expression, returns its argument expression, 3230/// otherwise returns NULL. 3231static const Expr *getSizeOfExprArg(const Expr* E) { 3232 if (const UnaryExprOrTypeTraitExpr *SizeOf = 3233 dyn_cast<UnaryExprOrTypeTraitExpr>(E)) 3234 if (SizeOf->getKind() == clang::UETT_SizeOf && !SizeOf->isArgumentType()) 3235 return SizeOf->getArgumentExpr()->IgnoreParenImpCasts(); 3236 3237 return 0; 3238} 3239 3240/// \brief If E is a sizeof expression, returns its argument type. 3241static QualType getSizeOfArgType(const Expr* E) { 3242 if (const UnaryExprOrTypeTraitExpr *SizeOf = 3243 dyn_cast<UnaryExprOrTypeTraitExpr>(E)) 3244 if (SizeOf->getKind() == clang::UETT_SizeOf) 3245 return SizeOf->getTypeOfArgument(); 3246 3247 return QualType(); 3248} 3249 3250/// \brief Check for dangerous or invalid arguments to memset(). 3251/// 3252/// This issues warnings on known problematic, dangerous or unspecified 3253/// arguments to the standard 'memset', 'memcpy', 'memmove', and 'memcmp' 3254/// function calls. 3255/// 3256/// \param Call The call expression to diagnose. 3257void Sema::CheckMemaccessArguments(const CallExpr *Call, 3258 unsigned BId, 3259 IdentifierInfo *FnName) { 3260 assert(BId != 0); 3261 3262 // It is possible to have a non-standard definition of memset. Validate 3263 // we have enough arguments, and if not, abort further checking. 3264 unsigned ExpectedNumArgs = (BId == Builtin::BIstrndup ? 2 : 3); 3265 if (Call->getNumArgs() < ExpectedNumArgs) 3266 return; 3267 3268 unsigned LastArg = (BId == Builtin::BImemset || 3269 BId == Builtin::BIstrndup ? 1 : 2); 3270 unsigned LenArg = (BId == Builtin::BIstrndup ? 1 : 2); 3271 const Expr *LenExpr = Call->getArg(LenArg)->IgnoreParenImpCasts(); 3272 3273 // We have special checking when the length is a sizeof expression. 3274 QualType SizeOfArgTy = getSizeOfArgType(LenExpr); 3275 const Expr *SizeOfArg = getSizeOfExprArg(LenExpr); 3276 llvm::FoldingSetNodeID SizeOfArgID; 3277 3278 for (unsigned ArgIdx = 0; ArgIdx != LastArg; ++ArgIdx) { 3279 const Expr *Dest = Call->getArg(ArgIdx)->IgnoreParenImpCasts(); 3280 SourceRange ArgRange = Call->getArg(ArgIdx)->getSourceRange(); 3281 3282 QualType DestTy = Dest->getType(); 3283 if (const PointerType *DestPtrTy = DestTy->getAs<PointerType>()) { 3284 QualType PointeeTy = DestPtrTy->getPointeeType(); 3285 3286 // Never warn about void type pointers. This can be used to suppress 3287 // false positives. 3288 if (PointeeTy->isVoidType()) 3289 continue; 3290 3291 // Catch "memset(p, 0, sizeof(p))" -- needs to be sizeof(*p). Do this by 3292 // actually comparing the expressions for equality. Because computing the 3293 // expression IDs can be expensive, we only do this if the diagnostic is 3294 // enabled. 3295 if (SizeOfArg && 3296 Diags.getDiagnosticLevel(diag::warn_sizeof_pointer_expr_memaccess, 3297 SizeOfArg->getExprLoc())) { 3298 // We only compute IDs for expressions if the warning is enabled, and 3299 // cache the sizeof arg's ID. 3300 if (SizeOfArgID == llvm::FoldingSetNodeID()) 3301 SizeOfArg->Profile(SizeOfArgID, Context, true); 3302 llvm::FoldingSetNodeID DestID; 3303 Dest->Profile(DestID, Context, true); 3304 if (DestID == SizeOfArgID) { 3305 // TODO: For strncpy() and friends, this could suggest sizeof(dst) 3306 // over sizeof(src) as well. 3307 unsigned ActionIdx = 0; // Default is to suggest dereferencing. 3308 StringRef ReadableName = FnName->getName(); 3309 3310 if (const UnaryOperator *UnaryOp = dyn_cast<UnaryOperator>(Dest)) 3311 if (UnaryOp->getOpcode() == UO_AddrOf) 3312 ActionIdx = 1; // If its an address-of operator, just remove it. 3313 if (!PointeeTy->isIncompleteType() && 3314 (Context.getTypeSize(PointeeTy) == Context.getCharWidth())) 3315 ActionIdx = 2; // If the pointee's size is sizeof(char), 3316 // suggest an explicit length. 3317 3318 // If the function is defined as a builtin macro, do not show macro 3319 // expansion. 3320 SourceLocation SL = SizeOfArg->getExprLoc(); 3321 SourceRange DSR = Dest->getSourceRange(); 3322 SourceRange SSR = SizeOfArg->getSourceRange(); 3323 SourceManager &SM = PP.getSourceManager(); 3324 3325 if (SM.isMacroArgExpansion(SL)) { 3326 ReadableName = Lexer::getImmediateMacroName(SL, SM, LangOpts); 3327 SL = SM.getSpellingLoc(SL); 3328 DSR = SourceRange(SM.getSpellingLoc(DSR.getBegin()), 3329 SM.getSpellingLoc(DSR.getEnd())); 3330 SSR = SourceRange(SM.getSpellingLoc(SSR.getBegin()), 3331 SM.getSpellingLoc(SSR.getEnd())); 3332 } 3333 3334 DiagRuntimeBehavior(SL, SizeOfArg, 3335 PDiag(diag::warn_sizeof_pointer_expr_memaccess) 3336 << ReadableName 3337 << PointeeTy 3338 << DestTy 3339 << DSR 3340 << SSR); 3341 DiagRuntimeBehavior(SL, SizeOfArg, 3342 PDiag(diag::warn_sizeof_pointer_expr_memaccess_note) 3343 << ActionIdx 3344 << SSR); 3345 3346 break; 3347 } 3348 } 3349 3350 // Also check for cases where the sizeof argument is the exact same 3351 // type as the memory argument, and where it points to a user-defined 3352 // record type. 3353 if (SizeOfArgTy != QualType()) { 3354 if (PointeeTy->isRecordType() && 3355 Context.typesAreCompatible(SizeOfArgTy, DestTy)) { 3356 DiagRuntimeBehavior(LenExpr->getExprLoc(), Dest, 3357 PDiag(diag::warn_sizeof_pointer_type_memaccess) 3358 << FnName << SizeOfArgTy << ArgIdx 3359 << PointeeTy << Dest->getSourceRange() 3360 << LenExpr->getSourceRange()); 3361 break; 3362 } 3363 } 3364 3365 // Always complain about dynamic classes. 3366 if (isDynamicClassType(PointeeTy)) { 3367 3368 unsigned OperationType = 0; 3369 // "overwritten" if we're warning about the destination for any call 3370 // but memcmp; otherwise a verb appropriate to the call. 3371 if (ArgIdx != 0 || BId == Builtin::BImemcmp) { 3372 if (BId == Builtin::BImemcpy) 3373 OperationType = 1; 3374 else if(BId == Builtin::BImemmove) 3375 OperationType = 2; 3376 else if (BId == Builtin::BImemcmp) 3377 OperationType = 3; 3378 } 3379 3380 DiagRuntimeBehavior( 3381 Dest->getExprLoc(), Dest, 3382 PDiag(diag::warn_dyn_class_memaccess) 3383 << (BId == Builtin::BImemcmp ? ArgIdx + 2 : ArgIdx) 3384 << FnName << PointeeTy 3385 << OperationType 3386 << Call->getCallee()->getSourceRange()); 3387 } else if (PointeeTy.hasNonTrivialObjCLifetime() && 3388 BId != Builtin::BImemset) 3389 DiagRuntimeBehavior( 3390 Dest->getExprLoc(), Dest, 3391 PDiag(diag::warn_arc_object_memaccess) 3392 << ArgIdx << FnName << PointeeTy 3393 << Call->getCallee()->getSourceRange()); 3394 else 3395 continue; 3396 3397 DiagRuntimeBehavior( 3398 Dest->getExprLoc(), Dest, 3399 PDiag(diag::note_bad_memaccess_silence) 3400 << FixItHint::CreateInsertion(ArgRange.getBegin(), "(void*)")); 3401 break; 3402 } 3403 } 3404} 3405 3406// A little helper routine: ignore addition and subtraction of integer literals. 3407// This intentionally does not ignore all integer constant expressions because 3408// we don't want to remove sizeof(). 3409static const Expr *ignoreLiteralAdditions(const Expr *Ex, ASTContext &Ctx) { 3410 Ex = Ex->IgnoreParenCasts(); 3411 3412 for (;;) { 3413 const BinaryOperator * BO = dyn_cast<BinaryOperator>(Ex); 3414 if (!BO || !BO->isAdditiveOp()) 3415 break; 3416 3417 const Expr *RHS = BO->getRHS()->IgnoreParenCasts(); 3418 const Expr *LHS = BO->getLHS()->IgnoreParenCasts(); 3419 3420 if (isa<IntegerLiteral>(RHS)) 3421 Ex = LHS; 3422 else if (isa<IntegerLiteral>(LHS)) 3423 Ex = RHS; 3424 else 3425 break; 3426 } 3427 3428 return Ex; 3429} 3430 3431static bool isConstantSizeArrayWithMoreThanOneElement(QualType Ty, 3432 ASTContext &Context) { 3433 // Only handle constant-sized or VLAs, but not flexible members. 3434 if (const ConstantArrayType *CAT = Context.getAsConstantArrayType(Ty)) { 3435 // Only issue the FIXIT for arrays of size > 1. 3436 if (CAT->getSize().getSExtValue() <= 1) 3437 return false; 3438 } else if (!Ty->isVariableArrayType()) { 3439 return false; 3440 } 3441 return true; 3442} 3443 3444// Warn if the user has made the 'size' argument to strlcpy or strlcat 3445// be the size of the source, instead of the destination. 3446void Sema::CheckStrlcpycatArguments(const CallExpr *Call, 3447 IdentifierInfo *FnName) { 3448 3449 // Don't crash if the user has the wrong number of arguments 3450 if (Call->getNumArgs() != 3) 3451 return; 3452 3453 const Expr *SrcArg = ignoreLiteralAdditions(Call->getArg(1), Context); 3454 const Expr *SizeArg = ignoreLiteralAdditions(Call->getArg(2), Context); 3455 const Expr *CompareWithSrc = NULL; 3456 3457 // Look for 'strlcpy(dst, x, sizeof(x))' 3458 if (const Expr *Ex = getSizeOfExprArg(SizeArg)) 3459 CompareWithSrc = Ex; 3460 else { 3461 // Look for 'strlcpy(dst, x, strlen(x))' 3462 if (const CallExpr *SizeCall = dyn_cast<CallExpr>(SizeArg)) { 3463 if (SizeCall->isBuiltinCall() == Builtin::BIstrlen 3464 && SizeCall->getNumArgs() == 1) 3465 CompareWithSrc = ignoreLiteralAdditions(SizeCall->getArg(0), Context); 3466 } 3467 } 3468 3469 if (!CompareWithSrc) 3470 return; 3471 3472 // Determine if the argument to sizeof/strlen is equal to the source 3473 // argument. In principle there's all kinds of things you could do 3474 // here, for instance creating an == expression and evaluating it with 3475 // EvaluateAsBooleanCondition, but this uses a more direct technique: 3476 const DeclRefExpr *SrcArgDRE = dyn_cast<DeclRefExpr>(SrcArg); 3477 if (!SrcArgDRE) 3478 return; 3479 3480 const DeclRefExpr *CompareWithSrcDRE = dyn_cast<DeclRefExpr>(CompareWithSrc); 3481 if (!CompareWithSrcDRE || 3482 SrcArgDRE->getDecl() != CompareWithSrcDRE->getDecl()) 3483 return; 3484 3485 const Expr *OriginalSizeArg = Call->getArg(2); 3486 Diag(CompareWithSrcDRE->getLocStart(), diag::warn_strlcpycat_wrong_size) 3487 << OriginalSizeArg->getSourceRange() << FnName; 3488 3489 // Output a FIXIT hint if the destination is an array (rather than a 3490 // pointer to an array). This could be enhanced to handle some 3491 // pointers if we know the actual size, like if DstArg is 'array+2' 3492 // we could say 'sizeof(array)-2'. 3493 const Expr *DstArg = Call->getArg(0)->IgnoreParenImpCasts(); 3494 if (!isConstantSizeArrayWithMoreThanOneElement(DstArg->getType(), Context)) 3495 return; 3496 3497 SmallString<128> sizeString; 3498 llvm::raw_svector_ostream OS(sizeString); 3499 OS << "sizeof("; 3500 DstArg->printPretty(OS, 0, getPrintingPolicy()); 3501 OS << ")"; 3502 3503 Diag(OriginalSizeArg->getLocStart(), diag::note_strlcpycat_wrong_size) 3504 << FixItHint::CreateReplacement(OriginalSizeArg->getSourceRange(), 3505 OS.str()); 3506} 3507 3508/// Check if two expressions refer to the same declaration. 3509static bool referToTheSameDecl(const Expr *E1, const Expr *E2) { 3510 if (const DeclRefExpr *D1 = dyn_cast_or_null<DeclRefExpr>(E1)) 3511 if (const DeclRefExpr *D2 = dyn_cast_or_null<DeclRefExpr>(E2)) 3512 return D1->getDecl() == D2->getDecl(); 3513 return false; 3514} 3515 3516static const Expr *getStrlenExprArg(const Expr *E) { 3517 if (const CallExpr *CE = dyn_cast<CallExpr>(E)) { 3518 const FunctionDecl *FD = CE->getDirectCallee(); 3519 if (!FD || FD->getMemoryFunctionKind() != Builtin::BIstrlen) 3520 return 0; 3521 return CE->getArg(0)->IgnoreParenCasts(); 3522 } 3523 return 0; 3524} 3525 3526// Warn on anti-patterns as the 'size' argument to strncat. 3527// The correct size argument should look like following: 3528// strncat(dst, src, sizeof(dst) - strlen(dest) - 1); 3529void Sema::CheckStrncatArguments(const CallExpr *CE, 3530 IdentifierInfo *FnName) { 3531 // Don't crash if the user has the wrong number of arguments. 3532 if (CE->getNumArgs() < 3) 3533 return; 3534 const Expr *DstArg = CE->getArg(0)->IgnoreParenCasts(); 3535 const Expr *SrcArg = CE->getArg(1)->IgnoreParenCasts(); 3536 const Expr *LenArg = CE->getArg(2)->IgnoreParenCasts(); 3537 3538 // Identify common expressions, which are wrongly used as the size argument 3539 // to strncat and may lead to buffer overflows. 3540 unsigned PatternType = 0; 3541 if (const Expr *SizeOfArg = getSizeOfExprArg(LenArg)) { 3542 // - sizeof(dst) 3543 if (referToTheSameDecl(SizeOfArg, DstArg)) 3544 PatternType = 1; 3545 // - sizeof(src) 3546 else if (referToTheSameDecl(SizeOfArg, SrcArg)) 3547 PatternType = 2; 3548 } else if (const BinaryOperator *BE = dyn_cast<BinaryOperator>(LenArg)) { 3549 if (BE->getOpcode() == BO_Sub) { 3550 const Expr *L = BE->getLHS()->IgnoreParenCasts(); 3551 const Expr *R = BE->getRHS()->IgnoreParenCasts(); 3552 // - sizeof(dst) - strlen(dst) 3553 if (referToTheSameDecl(DstArg, getSizeOfExprArg(L)) && 3554 referToTheSameDecl(DstArg, getStrlenExprArg(R))) 3555 PatternType = 1; 3556 // - sizeof(src) - (anything) 3557 else if (referToTheSameDecl(SrcArg, getSizeOfExprArg(L))) 3558 PatternType = 2; 3559 } 3560 } 3561 3562 if (PatternType == 0) 3563 return; 3564 3565 // Generate the diagnostic. 3566 SourceLocation SL = LenArg->getLocStart(); 3567 SourceRange SR = LenArg->getSourceRange(); 3568 SourceManager &SM = PP.getSourceManager(); 3569 3570 // If the function is defined as a builtin macro, do not show macro expansion. 3571 if (SM.isMacroArgExpansion(SL)) { 3572 SL = SM.getSpellingLoc(SL); 3573 SR = SourceRange(SM.getSpellingLoc(SR.getBegin()), 3574 SM.getSpellingLoc(SR.getEnd())); 3575 } 3576 3577 // Check if the destination is an array (rather than a pointer to an array). 3578 QualType DstTy = DstArg->getType(); 3579 bool isKnownSizeArray = isConstantSizeArrayWithMoreThanOneElement(DstTy, 3580 Context); 3581 if (!isKnownSizeArray) { 3582 if (PatternType == 1) 3583 Diag(SL, diag::warn_strncat_wrong_size) << SR; 3584 else 3585 Diag(SL, diag::warn_strncat_src_size) << SR; 3586 return; 3587 } 3588 3589 if (PatternType == 1) 3590 Diag(SL, diag::warn_strncat_large_size) << SR; 3591 else 3592 Diag(SL, diag::warn_strncat_src_size) << SR; 3593 3594 SmallString<128> sizeString; 3595 llvm::raw_svector_ostream OS(sizeString); 3596 OS << "sizeof("; 3597 DstArg->printPretty(OS, 0, getPrintingPolicy()); 3598 OS << ") - "; 3599 OS << "strlen("; 3600 DstArg->printPretty(OS, 0, getPrintingPolicy()); 3601 OS << ") - 1"; 3602 3603 Diag(SL, diag::note_strncat_wrong_size) 3604 << FixItHint::CreateReplacement(SR, OS.str()); 3605} 3606 3607//===--- CHECK: Return Address of Stack Variable --------------------------===// 3608 3609static Expr *EvalVal(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars, 3610 Decl *ParentDecl); 3611static Expr *EvalAddr(Expr* E, SmallVectorImpl<DeclRefExpr *> &refVars, 3612 Decl *ParentDecl); 3613 3614/// CheckReturnStackAddr - Check if a return statement returns the address 3615/// of a stack variable. 3616void 3617Sema::CheckReturnStackAddr(Expr *RetValExp, QualType lhsType, 3618 SourceLocation ReturnLoc) { 3619 3620 Expr *stackE = 0; 3621 SmallVector<DeclRefExpr *, 8> refVars; 3622 3623 // Perform checking for returned stack addresses, local blocks, 3624 // label addresses or references to temporaries. 3625 if (lhsType->isPointerType() || 3626 (!getLangOpts().ObjCAutoRefCount && lhsType->isBlockPointerType())) { 3627 stackE = EvalAddr(RetValExp, refVars, /*ParentDecl=*/0); 3628 } else if (lhsType->isReferenceType()) { 3629 stackE = EvalVal(RetValExp, refVars, /*ParentDecl=*/0); 3630 } 3631 3632 if (stackE == 0) 3633 return; // Nothing suspicious was found. 3634 3635 SourceLocation diagLoc; 3636 SourceRange diagRange; 3637 if (refVars.empty()) { 3638 diagLoc = stackE->getLocStart(); 3639 diagRange = stackE->getSourceRange(); 3640 } else { 3641 // We followed through a reference variable. 'stackE' contains the 3642 // problematic expression but we will warn at the return statement pointing 3643 // at the reference variable. We will later display the "trail" of 3644 // reference variables using notes. 3645 diagLoc = refVars[0]->getLocStart(); 3646 diagRange = refVars[0]->getSourceRange(); 3647 } 3648 3649 if (DeclRefExpr *DR = dyn_cast<DeclRefExpr>(stackE)) { //address of local var. 3650 Diag(diagLoc, lhsType->isReferenceType() ? diag::warn_ret_stack_ref 3651 : diag::warn_ret_stack_addr) 3652 << DR->getDecl()->getDeclName() << diagRange; 3653 } else if (isa<BlockExpr>(stackE)) { // local block. 3654 Diag(diagLoc, diag::err_ret_local_block) << diagRange; 3655 } else if (isa<AddrLabelExpr>(stackE)) { // address of label. 3656 Diag(diagLoc, diag::warn_ret_addr_label) << diagRange; 3657 } else { // local temporary. 3658 Diag(diagLoc, lhsType->isReferenceType() ? diag::warn_ret_local_temp_ref 3659 : diag::warn_ret_local_temp_addr) 3660 << diagRange; 3661 } 3662 3663 // Display the "trail" of reference variables that we followed until we 3664 // found the problematic expression using notes. 3665 for (unsigned i = 0, e = refVars.size(); i != e; ++i) { 3666 VarDecl *VD = cast<VarDecl>(refVars[i]->getDecl()); 3667 // If this var binds to another reference var, show the range of the next 3668 // var, otherwise the var binds to the problematic expression, in which case 3669 // show the range of the expression. 3670 SourceRange range = (i < e-1) ? refVars[i+1]->getSourceRange() 3671 : stackE->getSourceRange(); 3672 Diag(VD->getLocation(), diag::note_ref_var_local_bind) 3673 << VD->getDeclName() << range; 3674 } 3675} 3676 3677/// EvalAddr - EvalAddr and EvalVal are mutually recursive functions that 3678/// check if the expression in a return statement evaluates to an address 3679/// to a location on the stack, a local block, an address of a label, or a 3680/// reference to local temporary. The recursion is used to traverse the 3681/// AST of the return expression, with recursion backtracking when we 3682/// encounter a subexpression that (1) clearly does not lead to one of the 3683/// above problematic expressions (2) is something we cannot determine leads to 3684/// a problematic expression based on such local checking. 3685/// 3686/// Both EvalAddr and EvalVal follow through reference variables to evaluate 3687/// the expression that they point to. Such variables are added to the 3688/// 'refVars' vector so that we know what the reference variable "trail" was. 3689/// 3690/// EvalAddr processes expressions that are pointers that are used as 3691/// references (and not L-values). EvalVal handles all other values. 3692/// At the base case of the recursion is a check for the above problematic 3693/// expressions. 3694/// 3695/// This implementation handles: 3696/// 3697/// * pointer-to-pointer casts 3698/// * implicit conversions from array references to pointers 3699/// * taking the address of fields 3700/// * arbitrary interplay between "&" and "*" operators 3701/// * pointer arithmetic from an address of a stack variable 3702/// * taking the address of an array element where the array is on the stack 3703static Expr *EvalAddr(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars, 3704 Decl *ParentDecl) { 3705 if (E->isTypeDependent()) 3706 return NULL; 3707 3708 // We should only be called for evaluating pointer expressions. 3709 assert((E->getType()->isAnyPointerType() || 3710 E->getType()->isBlockPointerType() || 3711 E->getType()->isObjCQualifiedIdType()) && 3712 "EvalAddr only works on pointers"); 3713 3714 E = E->IgnoreParens(); 3715 3716 // Our "symbolic interpreter" is just a dispatch off the currently 3717 // viewed AST node. We then recursively traverse the AST by calling 3718 // EvalAddr and EvalVal appropriately. 3719 switch (E->getStmtClass()) { 3720 case Stmt::DeclRefExprClass: { 3721 DeclRefExpr *DR = cast<DeclRefExpr>(E); 3722 3723 if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl())) 3724 // If this is a reference variable, follow through to the expression that 3725 // it points to. 3726 if (V->hasLocalStorage() && 3727 V->getType()->isReferenceType() && V->hasInit()) { 3728 // Add the reference variable to the "trail". 3729 refVars.push_back(DR); 3730 return EvalAddr(V->getInit(), refVars, ParentDecl); 3731 } 3732 3733 return NULL; 3734 } 3735 3736 case Stmt::UnaryOperatorClass: { 3737 // The only unary operator that make sense to handle here 3738 // is AddrOf. All others don't make sense as pointers. 3739 UnaryOperator *U = cast<UnaryOperator>(E); 3740 3741 if (U->getOpcode() == UO_AddrOf) 3742 return EvalVal(U->getSubExpr(), refVars, ParentDecl); 3743 else 3744 return NULL; 3745 } 3746 3747 case Stmt::BinaryOperatorClass: { 3748 // Handle pointer arithmetic. All other binary operators are not valid 3749 // in this context. 3750 BinaryOperator *B = cast<BinaryOperator>(E); 3751 BinaryOperatorKind op = B->getOpcode(); 3752 3753 if (op != BO_Add && op != BO_Sub) 3754 return NULL; 3755 3756 Expr *Base = B->getLHS(); 3757 3758 // Determine which argument is the real pointer base. It could be 3759 // the RHS argument instead of the LHS. 3760 if (!Base->getType()->isPointerType()) Base = B->getRHS(); 3761 3762 assert (Base->getType()->isPointerType()); 3763 return EvalAddr(Base, refVars, ParentDecl); 3764 } 3765 3766 // For conditional operators we need to see if either the LHS or RHS are 3767 // valid DeclRefExpr*s. If one of them is valid, we return it. 3768 case Stmt::ConditionalOperatorClass: { 3769 ConditionalOperator *C = cast<ConditionalOperator>(E); 3770 3771 // Handle the GNU extension for missing LHS. 3772 if (Expr *lhsExpr = C->getLHS()) { 3773 // In C++, we can have a throw-expression, which has 'void' type. 3774 if (!lhsExpr->getType()->isVoidType()) 3775 if (Expr* LHS = EvalAddr(lhsExpr, refVars, ParentDecl)) 3776 return LHS; 3777 } 3778 3779 // In C++, we can have a throw-expression, which has 'void' type. 3780 if (C->getRHS()->getType()->isVoidType()) 3781 return NULL; 3782 3783 return EvalAddr(C->getRHS(), refVars, ParentDecl); 3784 } 3785 3786 case Stmt::BlockExprClass: 3787 if (cast<BlockExpr>(E)->getBlockDecl()->hasCaptures()) 3788 return E; // local block. 3789 return NULL; 3790 3791 case Stmt::AddrLabelExprClass: 3792 return E; // address of label. 3793 3794 case Stmt::ExprWithCleanupsClass: 3795 return EvalAddr(cast<ExprWithCleanups>(E)->getSubExpr(), refVars, 3796 ParentDecl); 3797 3798 // For casts, we need to handle conversions from arrays to 3799 // pointer values, and pointer-to-pointer conversions. 3800 case Stmt::ImplicitCastExprClass: 3801 case Stmt::CStyleCastExprClass: 3802 case Stmt::CXXFunctionalCastExprClass: 3803 case Stmt::ObjCBridgedCastExprClass: 3804 case Stmt::CXXStaticCastExprClass: 3805 case Stmt::CXXDynamicCastExprClass: 3806 case Stmt::CXXConstCastExprClass: 3807 case Stmt::CXXReinterpretCastExprClass: { 3808 Expr* SubExpr = cast<CastExpr>(E)->getSubExpr(); 3809 switch (cast<CastExpr>(E)->getCastKind()) { 3810 case CK_BitCast: 3811 case CK_LValueToRValue: 3812 case CK_NoOp: 3813 case CK_BaseToDerived: 3814 case CK_DerivedToBase: 3815 case CK_UncheckedDerivedToBase: 3816 case CK_Dynamic: 3817 case CK_CPointerToObjCPointerCast: 3818 case CK_BlockPointerToObjCPointerCast: 3819 case CK_AnyPointerToBlockPointerCast: 3820 return EvalAddr(SubExpr, refVars, ParentDecl); 3821 3822 case CK_ArrayToPointerDecay: 3823 return EvalVal(SubExpr, refVars, ParentDecl); 3824 3825 default: 3826 return 0; 3827 } 3828 } 3829 3830 case Stmt::MaterializeTemporaryExprClass: 3831 if (Expr *Result = EvalAddr( 3832 cast<MaterializeTemporaryExpr>(E)->GetTemporaryExpr(), 3833 refVars, ParentDecl)) 3834 return Result; 3835 3836 return E; 3837 3838 // Everything else: we simply don't reason about them. 3839 default: 3840 return NULL; 3841 } 3842} 3843 3844 3845/// EvalVal - This function is complements EvalAddr in the mutual recursion. 3846/// See the comments for EvalAddr for more details. 3847static Expr *EvalVal(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars, 3848 Decl *ParentDecl) { 3849do { 3850 // We should only be called for evaluating non-pointer expressions, or 3851 // expressions with a pointer type that are not used as references but instead 3852 // are l-values (e.g., DeclRefExpr with a pointer type). 3853 3854 // Our "symbolic interpreter" is just a dispatch off the currently 3855 // viewed AST node. We then recursively traverse the AST by calling 3856 // EvalAddr and EvalVal appropriately. 3857 3858 E = E->IgnoreParens(); 3859 switch (E->getStmtClass()) { 3860 case Stmt::ImplicitCastExprClass: { 3861 ImplicitCastExpr *IE = cast<ImplicitCastExpr>(E); 3862 if (IE->getValueKind() == VK_LValue) { 3863 E = IE->getSubExpr(); 3864 continue; 3865 } 3866 return NULL; 3867 } 3868 3869 case Stmt::ExprWithCleanupsClass: 3870 return EvalVal(cast<ExprWithCleanups>(E)->getSubExpr(), refVars,ParentDecl); 3871 3872 case Stmt::DeclRefExprClass: { 3873 // When we hit a DeclRefExpr we are looking at code that refers to a 3874 // variable's name. If it's not a reference variable we check if it has 3875 // local storage within the function, and if so, return the expression. 3876 DeclRefExpr *DR = cast<DeclRefExpr>(E); 3877 3878 if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl())) { 3879 // Check if it refers to itself, e.g. "int& i = i;". 3880 if (V == ParentDecl) 3881 return DR; 3882 3883 if (V->hasLocalStorage()) { 3884 if (!V->getType()->isReferenceType()) 3885 return DR; 3886 3887 // Reference variable, follow through to the expression that 3888 // it points to. 3889 if (V->hasInit()) { 3890 // Add the reference variable to the "trail". 3891 refVars.push_back(DR); 3892 return EvalVal(V->getInit(), refVars, V); 3893 } 3894 } 3895 } 3896 3897 return NULL; 3898 } 3899 3900 case Stmt::UnaryOperatorClass: { 3901 // The only unary operator that make sense to handle here 3902 // is Deref. All others don't resolve to a "name." This includes 3903 // handling all sorts of rvalues passed to a unary operator. 3904 UnaryOperator *U = cast<UnaryOperator>(E); 3905 3906 if (U->getOpcode() == UO_Deref) 3907 return EvalAddr(U->getSubExpr(), refVars, ParentDecl); 3908 3909 return NULL; 3910 } 3911 3912 case Stmt::ArraySubscriptExprClass: { 3913 // Array subscripts are potential references to data on the stack. We 3914 // retrieve the DeclRefExpr* for the array variable if it indeed 3915 // has local storage. 3916 return EvalAddr(cast<ArraySubscriptExpr>(E)->getBase(), refVars,ParentDecl); 3917 } 3918 3919 case Stmt::ConditionalOperatorClass: { 3920 // For conditional operators we need to see if either the LHS or RHS are 3921 // non-NULL Expr's. If one is non-NULL, we return it. 3922 ConditionalOperator *C = cast<ConditionalOperator>(E); 3923 3924 // Handle the GNU extension for missing LHS. 3925 if (Expr *lhsExpr = C->getLHS()) 3926 if (Expr *LHS = EvalVal(lhsExpr, refVars, ParentDecl)) 3927 return LHS; 3928 3929 return EvalVal(C->getRHS(), refVars, ParentDecl); 3930 } 3931 3932 // Accesses to members are potential references to data on the stack. 3933 case Stmt::MemberExprClass: { 3934 MemberExpr *M = cast<MemberExpr>(E); 3935 3936 // Check for indirect access. We only want direct field accesses. 3937 if (M->isArrow()) 3938 return NULL; 3939 3940 // Check whether the member type is itself a reference, in which case 3941 // we're not going to refer to the member, but to what the member refers to. 3942 if (M->getMemberDecl()->getType()->isReferenceType()) 3943 return NULL; 3944 3945 return EvalVal(M->getBase(), refVars, ParentDecl); 3946 } 3947 3948 case Stmt::MaterializeTemporaryExprClass: 3949 if (Expr *Result = EvalVal( 3950 cast<MaterializeTemporaryExpr>(E)->GetTemporaryExpr(), 3951 refVars, ParentDecl)) 3952 return Result; 3953 3954 return E; 3955 3956 default: 3957 // Check that we don't return or take the address of a reference to a 3958 // temporary. This is only useful in C++. 3959 if (!E->isTypeDependent() && E->isRValue()) 3960 return E; 3961 3962 // Everything else: we simply don't reason about them. 3963 return NULL; 3964 } 3965} while (true); 3966} 3967 3968//===--- CHECK: Floating-Point comparisons (-Wfloat-equal) ---------------===// 3969 3970/// Check for comparisons of floating point operands using != and ==. 3971/// Issue a warning if these are no self-comparisons, as they are not likely 3972/// to do what the programmer intended. 3973void Sema::CheckFloatComparison(SourceLocation Loc, Expr* LHS, Expr *RHS) { 3974 Expr* LeftExprSansParen = LHS->IgnoreParenImpCasts(); 3975 Expr* RightExprSansParen = RHS->IgnoreParenImpCasts(); 3976 3977 // Special case: check for x == x (which is OK). 3978 // Do not emit warnings for such cases. 3979 if (DeclRefExpr* DRL = dyn_cast<DeclRefExpr>(LeftExprSansParen)) 3980 if (DeclRefExpr* DRR = dyn_cast<DeclRefExpr>(RightExprSansParen)) 3981 if (DRL->getDecl() == DRR->getDecl()) 3982 return; 3983 3984 3985 // Special case: check for comparisons against literals that can be exactly 3986 // represented by APFloat. In such cases, do not emit a warning. This 3987 // is a heuristic: often comparison against such literals are used to 3988 // detect if a value in a variable has not changed. This clearly can 3989 // lead to false negatives. 3990 if (FloatingLiteral* FLL = dyn_cast<FloatingLiteral>(LeftExprSansParen)) { 3991 if (FLL->isExact()) 3992 return; 3993 } else 3994 if (FloatingLiteral* FLR = dyn_cast<FloatingLiteral>(RightExprSansParen)) 3995 if (FLR->isExact()) 3996 return; 3997 3998 // Check for comparisons with builtin types. 3999 if (CallExpr* CL = dyn_cast<CallExpr>(LeftExprSansParen)) 4000 if (CL->isBuiltinCall()) 4001 return; 4002 4003 if (CallExpr* CR = dyn_cast<CallExpr>(RightExprSansParen)) 4004 if (CR->isBuiltinCall()) 4005 return; 4006 4007 // Emit the diagnostic. 4008 Diag(Loc, diag::warn_floatingpoint_eq) 4009 << LHS->getSourceRange() << RHS->getSourceRange(); 4010} 4011 4012//===--- CHECK: Integer mixed-sign comparisons (-Wsign-compare) --------===// 4013//===--- CHECK: Lossy implicit conversions (-Wconversion) --------------===// 4014 4015namespace { 4016 4017/// Structure recording the 'active' range of an integer-valued 4018/// expression. 4019struct IntRange { 4020 /// The number of bits active in the int. 4021 unsigned Width; 4022 4023 /// True if the int is known not to have negative values. 4024 bool NonNegative; 4025 4026 IntRange(unsigned Width, bool NonNegative) 4027 : Width(Width), NonNegative(NonNegative) 4028 {} 4029 4030 /// Returns the range of the bool type. 4031 static IntRange forBoolType() { 4032 return IntRange(1, true); 4033 } 4034 4035 /// Returns the range of an opaque value of the given integral type. 4036 static IntRange forValueOfType(ASTContext &C, QualType T) { 4037 return forValueOfCanonicalType(C, 4038 T->getCanonicalTypeInternal().getTypePtr()); 4039 } 4040 4041 /// Returns the range of an opaque value of a canonical integral type. 4042 static IntRange forValueOfCanonicalType(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 4050 // For enum types, use the known bit width of the enumerators. 4051 if (const EnumType *ET = dyn_cast<EnumType>(T)) { 4052 EnumDecl *Enum = ET->getDecl(); 4053 if (!Enum->isCompleteDefinition()) 4054 return IntRange(C.getIntWidth(QualType(T, 0)), false); 4055 4056 unsigned NumPositive = Enum->getNumPositiveBits(); 4057 unsigned NumNegative = Enum->getNumNegativeBits(); 4058 4059 if (NumNegative == 0) 4060 return IntRange(NumPositive, true/*NonNegative*/); 4061 else 4062 return IntRange(std::max(NumPositive + 1, NumNegative), 4063 false/*NonNegative*/); 4064 } 4065 4066 const BuiltinType *BT = cast<BuiltinType>(T); 4067 assert(BT->isInteger()); 4068 4069 return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger()); 4070 } 4071 4072 /// Returns the "target" range of a canonical integral type, i.e. 4073 /// the range of values expressible in the type. 4074 /// 4075 /// This matches forValueOfCanonicalType except that enums have the 4076 /// full range of their type, not the range of their enumerators. 4077 static IntRange forTargetOfCanonicalType(ASTContext &C, const Type *T) { 4078 assert(T->isCanonicalUnqualified()); 4079 4080 if (const VectorType *VT = dyn_cast<VectorType>(T)) 4081 T = VT->getElementType().getTypePtr(); 4082 if (const ComplexType *CT = dyn_cast<ComplexType>(T)) 4083 T = CT->getElementType().getTypePtr(); 4084 if (const EnumType *ET = dyn_cast<EnumType>(T)) 4085 T = C.getCanonicalType(ET->getDecl()->getIntegerType()).getTypePtr(); 4086 4087 const BuiltinType *BT = cast<BuiltinType>(T); 4088 assert(BT->isInteger()); 4089 4090 return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger()); 4091 } 4092 4093 /// Returns the supremum of two ranges: i.e. their conservative merge. 4094 static IntRange join(IntRange L, IntRange R) { 4095 return IntRange(std::max(L.Width, R.Width), 4096 L.NonNegative && R.NonNegative); 4097 } 4098 4099 /// Returns the infinum of two ranges: i.e. their aggressive merge. 4100 static IntRange meet(IntRange L, IntRange R) { 4101 return IntRange(std::min(L.Width, R.Width), 4102 L.NonNegative || R.NonNegative); 4103 } 4104}; 4105 4106static IntRange GetValueRange(ASTContext &C, llvm::APSInt &value, 4107 unsigned MaxWidth) { 4108 if (value.isSigned() && value.isNegative()) 4109 return IntRange(value.getMinSignedBits(), false); 4110 4111 if (value.getBitWidth() > MaxWidth) 4112 value = value.trunc(MaxWidth); 4113 4114 // isNonNegative() just checks the sign bit without considering 4115 // signedness. 4116 return IntRange(value.getActiveBits(), true); 4117} 4118 4119static IntRange GetValueRange(ASTContext &C, APValue &result, QualType Ty, 4120 unsigned MaxWidth) { 4121 if (result.isInt()) 4122 return GetValueRange(C, result.getInt(), MaxWidth); 4123 4124 if (result.isVector()) { 4125 IntRange R = GetValueRange(C, result.getVectorElt(0), Ty, MaxWidth); 4126 for (unsigned i = 1, e = result.getVectorLength(); i != e; ++i) { 4127 IntRange El = GetValueRange(C, result.getVectorElt(i), Ty, MaxWidth); 4128 R = IntRange::join(R, El); 4129 } 4130 return R; 4131 } 4132 4133 if (result.isComplexInt()) { 4134 IntRange R = GetValueRange(C, result.getComplexIntReal(), MaxWidth); 4135 IntRange I = GetValueRange(C, result.getComplexIntImag(), MaxWidth); 4136 return IntRange::join(R, I); 4137 } 4138 4139 // This can happen with lossless casts to intptr_t of "based" lvalues. 4140 // Assume it might use arbitrary bits. 4141 // FIXME: The only reason we need to pass the type in here is to get 4142 // the sign right on this one case. It would be nice if APValue 4143 // preserved this. 4144 assert(result.isLValue() || result.isAddrLabelDiff()); 4145 return IntRange(MaxWidth, Ty->isUnsignedIntegerOrEnumerationType()); 4146} 4147 4148static QualType GetExprType(Expr *E) { 4149 QualType Ty = E->getType(); 4150 if (const AtomicType *AtomicRHS = Ty->getAs<AtomicType>()) 4151 Ty = AtomicRHS->getValueType(); 4152 return Ty; 4153} 4154 4155/// Pseudo-evaluate the given integer expression, estimating the 4156/// range of values it might take. 4157/// 4158/// \param MaxWidth - the width to which the value will be truncated 4159static IntRange GetExprRange(ASTContext &C, Expr *E, unsigned MaxWidth) { 4160 E = E->IgnoreParens(); 4161 4162 // Try a full evaluation first. 4163 Expr::EvalResult result; 4164 if (E->EvaluateAsRValue(result, C)) 4165 return GetValueRange(C, result.Val, GetExprType(E), MaxWidth); 4166 4167 // I think we only want to look through implicit casts here; if the 4168 // user has an explicit widening cast, we should treat the value as 4169 // being of the new, wider type. 4170 if (ImplicitCastExpr *CE = dyn_cast<ImplicitCastExpr>(E)) { 4171 if (CE->getCastKind() == CK_NoOp || CE->getCastKind() == CK_LValueToRValue) 4172 return GetExprRange(C, CE->getSubExpr(), MaxWidth); 4173 4174 IntRange OutputTypeRange = IntRange::forValueOfType(C, GetExprType(CE)); 4175 4176 bool isIntegerCast = (CE->getCastKind() == CK_IntegralCast); 4177 4178 // Assume that non-integer casts can span the full range of the type. 4179 if (!isIntegerCast) 4180 return OutputTypeRange; 4181 4182 IntRange SubRange 4183 = GetExprRange(C, CE->getSubExpr(), 4184 std::min(MaxWidth, OutputTypeRange.Width)); 4185 4186 // Bail out if the subexpr's range is as wide as the cast type. 4187 if (SubRange.Width >= OutputTypeRange.Width) 4188 return OutputTypeRange; 4189 4190 // Otherwise, we take the smaller width, and we're non-negative if 4191 // either the output type or the subexpr is. 4192 return IntRange(SubRange.Width, 4193 SubRange.NonNegative || OutputTypeRange.NonNegative); 4194 } 4195 4196 if (ConditionalOperator *CO = dyn_cast<ConditionalOperator>(E)) { 4197 // If we can fold the condition, just take that operand. 4198 bool CondResult; 4199 if (CO->getCond()->EvaluateAsBooleanCondition(CondResult, C)) 4200 return GetExprRange(C, CondResult ? CO->getTrueExpr() 4201 : CO->getFalseExpr(), 4202 MaxWidth); 4203 4204 // Otherwise, conservatively merge. 4205 IntRange L = GetExprRange(C, CO->getTrueExpr(), MaxWidth); 4206 IntRange R = GetExprRange(C, CO->getFalseExpr(), MaxWidth); 4207 return IntRange::join(L, R); 4208 } 4209 4210 if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) { 4211 switch (BO->getOpcode()) { 4212 4213 // Boolean-valued operations are single-bit and positive. 4214 case BO_LAnd: 4215 case BO_LOr: 4216 case BO_LT: 4217 case BO_GT: 4218 case BO_LE: 4219 case BO_GE: 4220 case BO_EQ: 4221 case BO_NE: 4222 return IntRange::forBoolType(); 4223 4224 // The type of the assignments is the type of the LHS, so the RHS 4225 // is not necessarily the same type. 4226 case BO_MulAssign: 4227 case BO_DivAssign: 4228 case BO_RemAssign: 4229 case BO_AddAssign: 4230 case BO_SubAssign: 4231 case BO_XorAssign: 4232 case BO_OrAssign: 4233 // TODO: bitfields? 4234 return IntRange::forValueOfType(C, GetExprType(E)); 4235 4236 // Simple assignments just pass through the RHS, which will have 4237 // been coerced to the LHS type. 4238 case BO_Assign: 4239 // TODO: bitfields? 4240 return GetExprRange(C, BO->getRHS(), MaxWidth); 4241 4242 // Operations with opaque sources are black-listed. 4243 case BO_PtrMemD: 4244 case BO_PtrMemI: 4245 return IntRange::forValueOfType(C, GetExprType(E)); 4246 4247 // Bitwise-and uses the *infinum* of the two source ranges. 4248 case BO_And: 4249 case BO_AndAssign: 4250 return IntRange::meet(GetExprRange(C, BO->getLHS(), MaxWidth), 4251 GetExprRange(C, BO->getRHS(), MaxWidth)); 4252 4253 // Left shift gets black-listed based on a judgement call. 4254 case BO_Shl: 4255 // ...except that we want to treat '1 << (blah)' as logically 4256 // positive. It's an important idiom. 4257 if (IntegerLiteral *I 4258 = dyn_cast<IntegerLiteral>(BO->getLHS()->IgnoreParenCasts())) { 4259 if (I->getValue() == 1) { 4260 IntRange R = IntRange::forValueOfType(C, GetExprType(E)); 4261 return IntRange(R.Width, /*NonNegative*/ true); 4262 } 4263 } 4264 // fallthrough 4265 4266 case BO_ShlAssign: 4267 return IntRange::forValueOfType(C, GetExprType(E)); 4268 4269 // Right shift by a constant can narrow its left argument. 4270 case BO_Shr: 4271 case BO_ShrAssign: { 4272 IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth); 4273 4274 // If the shift amount is a positive constant, drop the width by 4275 // that much. 4276 llvm::APSInt shift; 4277 if (BO->getRHS()->isIntegerConstantExpr(shift, C) && 4278 shift.isNonNegative()) { 4279 unsigned zext = shift.getZExtValue(); 4280 if (zext >= L.Width) 4281 L.Width = (L.NonNegative ? 0 : 1); 4282 else 4283 L.Width -= zext; 4284 } 4285 4286 return L; 4287 } 4288 4289 // Comma acts as its right operand. 4290 case BO_Comma: 4291 return GetExprRange(C, BO->getRHS(), MaxWidth); 4292 4293 // Black-list pointer subtractions. 4294 case BO_Sub: 4295 if (BO->getLHS()->getType()->isPointerType()) 4296 return IntRange::forValueOfType(C, GetExprType(E)); 4297 break; 4298 4299 // The width of a division result is mostly determined by the size 4300 // of the LHS. 4301 case BO_Div: { 4302 // Don't 'pre-truncate' the operands. 4303 unsigned opWidth = C.getIntWidth(GetExprType(E)); 4304 IntRange L = GetExprRange(C, BO->getLHS(), opWidth); 4305 4306 // If the divisor is constant, use that. 4307 llvm::APSInt divisor; 4308 if (BO->getRHS()->isIntegerConstantExpr(divisor, C)) { 4309 unsigned log2 = divisor.logBase2(); // floor(log_2(divisor)) 4310 if (log2 >= L.Width) 4311 L.Width = (L.NonNegative ? 0 : 1); 4312 else 4313 L.Width = std::min(L.Width - log2, MaxWidth); 4314 return L; 4315 } 4316 4317 // Otherwise, just use the LHS's width. 4318 IntRange R = GetExprRange(C, BO->getRHS(), opWidth); 4319 return IntRange(L.Width, L.NonNegative && R.NonNegative); 4320 } 4321 4322 // The result of a remainder can't be larger than the result of 4323 // either side. 4324 case BO_Rem: { 4325 // Don't 'pre-truncate' the operands. 4326 unsigned opWidth = C.getIntWidth(GetExprType(E)); 4327 IntRange L = GetExprRange(C, BO->getLHS(), opWidth); 4328 IntRange R = GetExprRange(C, BO->getRHS(), opWidth); 4329 4330 IntRange meet = IntRange::meet(L, R); 4331 meet.Width = std::min(meet.Width, MaxWidth); 4332 return meet; 4333 } 4334 4335 // The default behavior is okay for these. 4336 case BO_Mul: 4337 case BO_Add: 4338 case BO_Xor: 4339 case BO_Or: 4340 break; 4341 } 4342 4343 // The default case is to treat the operation as if it were closed 4344 // on the narrowest type that encompasses both operands. 4345 IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth); 4346 IntRange R = GetExprRange(C, BO->getRHS(), MaxWidth); 4347 return IntRange::join(L, R); 4348 } 4349 4350 if (UnaryOperator *UO = dyn_cast<UnaryOperator>(E)) { 4351 switch (UO->getOpcode()) { 4352 // Boolean-valued operations are white-listed. 4353 case UO_LNot: 4354 return IntRange::forBoolType(); 4355 4356 // Operations with opaque sources are black-listed. 4357 case UO_Deref: 4358 case UO_AddrOf: // should be impossible 4359 return IntRange::forValueOfType(C, GetExprType(E)); 4360 4361 default: 4362 return GetExprRange(C, UO->getSubExpr(), MaxWidth); 4363 } 4364 } 4365 4366 if (FieldDecl *BitField = E->getSourceBitField()) 4367 return IntRange(BitField->getBitWidthValue(C), 4368 BitField->getType()->isUnsignedIntegerOrEnumerationType()); 4369 4370 return IntRange::forValueOfType(C, GetExprType(E)); 4371} 4372 4373static IntRange GetExprRange(ASTContext &C, Expr *E) { 4374 return GetExprRange(C, E, C.getIntWidth(GetExprType(E))); 4375} 4376 4377/// Checks whether the given value, which currently has the given 4378/// source semantics, has the same value when coerced through the 4379/// target semantics. 4380static bool IsSameFloatAfterCast(const llvm::APFloat &value, 4381 const llvm::fltSemantics &Src, 4382 const llvm::fltSemantics &Tgt) { 4383 llvm::APFloat truncated = value; 4384 4385 bool ignored; 4386 truncated.convert(Src, llvm::APFloat::rmNearestTiesToEven, &ignored); 4387 truncated.convert(Tgt, llvm::APFloat::rmNearestTiesToEven, &ignored); 4388 4389 return truncated.bitwiseIsEqual(value); 4390} 4391 4392/// Checks whether the given value, which currently has the given 4393/// source semantics, has the same value when coerced through the 4394/// target semantics. 4395/// 4396/// The value might be a vector of floats (or a complex number). 4397static bool IsSameFloatAfterCast(const APValue &value, 4398 const llvm::fltSemantics &Src, 4399 const llvm::fltSemantics &Tgt) { 4400 if (value.isFloat()) 4401 return IsSameFloatAfterCast(value.getFloat(), Src, Tgt); 4402 4403 if (value.isVector()) { 4404 for (unsigned i = 0, e = value.getVectorLength(); i != e; ++i) 4405 if (!IsSameFloatAfterCast(value.getVectorElt(i), Src, Tgt)) 4406 return false; 4407 return true; 4408 } 4409 4410 assert(value.isComplexFloat()); 4411 return (IsSameFloatAfterCast(value.getComplexFloatReal(), Src, Tgt) && 4412 IsSameFloatAfterCast(value.getComplexFloatImag(), Src, Tgt)); 4413} 4414 4415static void AnalyzeImplicitConversions(Sema &S, Expr *E, SourceLocation CC); 4416 4417static bool IsZero(Sema &S, Expr *E) { 4418 // Suppress cases where we are comparing against an enum constant. 4419 if (const DeclRefExpr *DR = 4420 dyn_cast<DeclRefExpr>(E->IgnoreParenImpCasts())) 4421 if (isa<EnumConstantDecl>(DR->getDecl())) 4422 return false; 4423 4424 // Suppress cases where the '0' value is expanded from a macro. 4425 if (E->getLocStart().isMacroID()) 4426 return false; 4427 4428 llvm::APSInt Value; 4429 return E->isIntegerConstantExpr(Value, S.Context) && Value == 0; 4430} 4431 4432static bool HasEnumType(Expr *E) { 4433 // Strip off implicit integral promotions. 4434 while (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(E)) { 4435 if (ICE->getCastKind() != CK_IntegralCast && 4436 ICE->getCastKind() != CK_NoOp) 4437 break; 4438 E = ICE->getSubExpr(); 4439 } 4440 4441 return E->getType()->isEnumeralType(); 4442} 4443 4444static void CheckTrivialUnsignedComparison(Sema &S, BinaryOperator *E) { 4445 BinaryOperatorKind op = E->getOpcode(); 4446 if (E->isValueDependent()) 4447 return; 4448 4449 if (op == BO_LT && IsZero(S, E->getRHS())) { 4450 S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison) 4451 << "< 0" << "false" << HasEnumType(E->getLHS()) 4452 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 4453 } else if (op == BO_GE && IsZero(S, E->getRHS())) { 4454 S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison) 4455 << ">= 0" << "true" << HasEnumType(E->getLHS()) 4456 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 4457 } else if (op == BO_GT && IsZero(S, E->getLHS())) { 4458 S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison) 4459 << "0 >" << "false" << HasEnumType(E->getRHS()) 4460 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 4461 } else if (op == BO_LE && IsZero(S, E->getLHS())) { 4462 S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison) 4463 << "0 <=" << "true" << HasEnumType(E->getRHS()) 4464 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 4465 } 4466} 4467 4468static void DiagnoseOutOfRangeComparison(Sema &S, BinaryOperator *E, 4469 Expr *Constant, Expr *Other, 4470 llvm::APSInt Value, 4471 bool RhsConstant) { 4472 // 0 values are handled later by CheckTrivialUnsignedComparison(). 4473 if (Value == 0) 4474 return; 4475 4476 BinaryOperatorKind op = E->getOpcode(); 4477 QualType OtherT = Other->getType(); 4478 QualType ConstantT = Constant->getType(); 4479 QualType CommonT = E->getLHS()->getType(); 4480 if (S.Context.hasSameUnqualifiedType(OtherT, ConstantT)) 4481 return; 4482 assert((OtherT->isIntegerType() && ConstantT->isIntegerType()) 4483 && "comparison with non-integer type"); 4484 4485 bool ConstantSigned = ConstantT->isSignedIntegerType(); 4486 bool CommonSigned = CommonT->isSignedIntegerType(); 4487 4488 bool EqualityOnly = false; 4489 4490 // TODO: Investigate using GetExprRange() to get tighter bounds on 4491 // on the bit ranges. 4492 IntRange OtherRange = IntRange::forValueOfType(S.Context, OtherT); 4493 unsigned OtherWidth = OtherRange.Width; 4494 4495 if (CommonSigned) { 4496 // The common type is signed, therefore no signed to unsigned conversion. 4497 if (!OtherRange.NonNegative) { 4498 // Check that the constant is representable in type OtherT. 4499 if (ConstantSigned) { 4500 if (OtherWidth >= Value.getMinSignedBits()) 4501 return; 4502 } else { // !ConstantSigned 4503 if (OtherWidth >= Value.getActiveBits() + 1) 4504 return; 4505 } 4506 } else { // !OtherSigned 4507 // Check that the constant is representable in type OtherT. 4508 // Negative values are out of range. 4509 if (ConstantSigned) { 4510 if (Value.isNonNegative() && OtherWidth >= Value.getActiveBits()) 4511 return; 4512 } else { // !ConstantSigned 4513 if (OtherWidth >= Value.getActiveBits()) 4514 return; 4515 } 4516 } 4517 } else { // !CommonSigned 4518 if (OtherRange.NonNegative) { 4519 if (OtherWidth >= Value.getActiveBits()) 4520 return; 4521 } else if (!OtherRange.NonNegative && !ConstantSigned) { 4522 // Check to see if the constant is representable in OtherT. 4523 if (OtherWidth > Value.getActiveBits()) 4524 return; 4525 // Check to see if the constant is equivalent to a negative value 4526 // cast to CommonT. 4527 if (S.Context.getIntWidth(ConstantT) == S.Context.getIntWidth(CommonT) && 4528 Value.isNegative() && Value.getMinSignedBits() <= OtherWidth) 4529 return; 4530 // The constant value rests between values that OtherT can represent after 4531 // conversion. Relational comparison still works, but equality 4532 // comparisons will be tautological. 4533 EqualityOnly = true; 4534 } else { // OtherSigned && ConstantSigned 4535 assert(0 && "Two signed types converted to unsigned types."); 4536 } 4537 } 4538 4539 bool PositiveConstant = !ConstantSigned || Value.isNonNegative(); 4540 4541 bool IsTrue = true; 4542 if (op == BO_EQ || op == BO_NE) { 4543 IsTrue = op == BO_NE; 4544 } else if (EqualityOnly) { 4545 return; 4546 } else if (RhsConstant) { 4547 if (op == BO_GT || op == BO_GE) 4548 IsTrue = !PositiveConstant; 4549 else // op == BO_LT || op == BO_LE 4550 IsTrue = PositiveConstant; 4551 } else { 4552 if (op == BO_LT || op == BO_LE) 4553 IsTrue = !PositiveConstant; 4554 else // op == BO_GT || op == BO_GE 4555 IsTrue = PositiveConstant; 4556 } 4557 4558 // If this is a comparison to an enum constant, include that 4559 // constant in the diagnostic. 4560 const EnumConstantDecl *ED = 0; 4561 if (const DeclRefExpr *DR = dyn_cast<DeclRefExpr>(Constant)) 4562 ED = dyn_cast<EnumConstantDecl>(DR->getDecl()); 4563 4564 SmallString<64> PrettySourceValue; 4565 llvm::raw_svector_ostream OS(PrettySourceValue); 4566 if (ED) 4567 OS << '\'' << *ED << "' (" << Value << ")"; 4568 else 4569 OS << Value; 4570 4571 S.Diag(E->getOperatorLoc(), diag::warn_out_of_range_compare) 4572 << OS.str() << OtherT << IsTrue 4573 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 4574} 4575 4576/// Analyze the operands of the given comparison. Implements the 4577/// fallback case from AnalyzeComparison. 4578static void AnalyzeImpConvsInComparison(Sema &S, BinaryOperator *E) { 4579 AnalyzeImplicitConversions(S, E->getLHS(), E->getOperatorLoc()); 4580 AnalyzeImplicitConversions(S, E->getRHS(), E->getOperatorLoc()); 4581} 4582 4583/// \brief Implements -Wsign-compare. 4584/// 4585/// \param E the binary operator to check for warnings 4586static void AnalyzeComparison(Sema &S, BinaryOperator *E) { 4587 // The type the comparison is being performed in. 4588 QualType T = E->getLHS()->getType(); 4589 assert(S.Context.hasSameUnqualifiedType(T, E->getRHS()->getType()) 4590 && "comparison with mismatched types"); 4591 if (E->isValueDependent()) 4592 return AnalyzeImpConvsInComparison(S, E); 4593 4594 Expr *LHS = E->getLHS()->IgnoreParenImpCasts(); 4595 Expr *RHS = E->getRHS()->IgnoreParenImpCasts(); 4596 4597 bool IsComparisonConstant = false; 4598 4599 // Check whether an integer constant comparison results in a value 4600 // of 'true' or 'false'. 4601 if (T->isIntegralType(S.Context)) { 4602 llvm::APSInt RHSValue; 4603 bool IsRHSIntegralLiteral = 4604 RHS->isIntegerConstantExpr(RHSValue, S.Context); 4605 llvm::APSInt LHSValue; 4606 bool IsLHSIntegralLiteral = 4607 LHS->isIntegerConstantExpr(LHSValue, S.Context); 4608 if (IsRHSIntegralLiteral && !IsLHSIntegralLiteral) 4609 DiagnoseOutOfRangeComparison(S, E, RHS, LHS, RHSValue, true); 4610 else if (!IsRHSIntegralLiteral && IsLHSIntegralLiteral) 4611 DiagnoseOutOfRangeComparison(S, E, LHS, RHS, LHSValue, false); 4612 else 4613 IsComparisonConstant = 4614 (IsRHSIntegralLiteral && IsLHSIntegralLiteral); 4615 } else if (!T->hasUnsignedIntegerRepresentation()) 4616 IsComparisonConstant = E->isIntegerConstantExpr(S.Context); 4617 4618 // We don't do anything special if this isn't an unsigned integral 4619 // comparison: we're only interested in integral comparisons, and 4620 // signed comparisons only happen in cases we don't care to warn about. 4621 // 4622 // We also don't care about value-dependent expressions or expressions 4623 // whose result is a constant. 4624 if (!T->hasUnsignedIntegerRepresentation() || IsComparisonConstant) 4625 return AnalyzeImpConvsInComparison(S, E); 4626 4627 // Check to see if one of the (unmodified) operands is of different 4628 // signedness. 4629 Expr *signedOperand, *unsignedOperand; 4630 if (LHS->getType()->hasSignedIntegerRepresentation()) { 4631 assert(!RHS->getType()->hasSignedIntegerRepresentation() && 4632 "unsigned comparison between two signed integer expressions?"); 4633 signedOperand = LHS; 4634 unsignedOperand = RHS; 4635 } else if (RHS->getType()->hasSignedIntegerRepresentation()) { 4636 signedOperand = RHS; 4637 unsignedOperand = LHS; 4638 } else { 4639 CheckTrivialUnsignedComparison(S, E); 4640 return AnalyzeImpConvsInComparison(S, E); 4641 } 4642 4643 // Otherwise, calculate the effective range of the signed operand. 4644 IntRange signedRange = GetExprRange(S.Context, signedOperand); 4645 4646 // Go ahead and analyze implicit conversions in the operands. Note 4647 // that we skip the implicit conversions on both sides. 4648 AnalyzeImplicitConversions(S, LHS, E->getOperatorLoc()); 4649 AnalyzeImplicitConversions(S, RHS, E->getOperatorLoc()); 4650 4651 // If the signed range is non-negative, -Wsign-compare won't fire, 4652 // but we should still check for comparisons which are always true 4653 // or false. 4654 if (signedRange.NonNegative) 4655 return CheckTrivialUnsignedComparison(S, E); 4656 4657 // For (in)equality comparisons, if the unsigned operand is a 4658 // constant which cannot collide with a overflowed signed operand, 4659 // then reinterpreting the signed operand as unsigned will not 4660 // change the result of the comparison. 4661 if (E->isEqualityOp()) { 4662 unsigned comparisonWidth = S.Context.getIntWidth(T); 4663 IntRange unsignedRange = GetExprRange(S.Context, unsignedOperand); 4664 4665 // We should never be unable to prove that the unsigned operand is 4666 // non-negative. 4667 assert(unsignedRange.NonNegative && "unsigned range includes negative?"); 4668 4669 if (unsignedRange.Width < comparisonWidth) 4670 return; 4671 } 4672 4673 S.DiagRuntimeBehavior(E->getOperatorLoc(), E, 4674 S.PDiag(diag::warn_mixed_sign_comparison) 4675 << LHS->getType() << RHS->getType() 4676 << LHS->getSourceRange() << RHS->getSourceRange()); 4677} 4678 4679/// Analyzes an attempt to assign the given value to a bitfield. 4680/// 4681/// Returns true if there was something fishy about the attempt. 4682static bool AnalyzeBitFieldAssignment(Sema &S, FieldDecl *Bitfield, Expr *Init, 4683 SourceLocation InitLoc) { 4684 assert(Bitfield->isBitField()); 4685 if (Bitfield->isInvalidDecl()) 4686 return false; 4687 4688 // White-list bool bitfields. 4689 if (Bitfield->getType()->isBooleanType()) 4690 return false; 4691 4692 // Ignore value- or type-dependent expressions. 4693 if (Bitfield->getBitWidth()->isValueDependent() || 4694 Bitfield->getBitWidth()->isTypeDependent() || 4695 Init->isValueDependent() || 4696 Init->isTypeDependent()) 4697 return false; 4698 4699 Expr *OriginalInit = Init->IgnoreParenImpCasts(); 4700 4701 llvm::APSInt Value; 4702 if (!OriginalInit->EvaluateAsInt(Value, S.Context, Expr::SE_AllowSideEffects)) 4703 return false; 4704 4705 unsigned OriginalWidth = Value.getBitWidth(); 4706 unsigned FieldWidth = Bitfield->getBitWidthValue(S.Context); 4707 4708 if (OriginalWidth <= FieldWidth) 4709 return false; 4710 4711 // Compute the value which the bitfield will contain. 4712 llvm::APSInt TruncatedValue = Value.trunc(FieldWidth); 4713 TruncatedValue.setIsSigned(Bitfield->getType()->isSignedIntegerType()); 4714 4715 // Check whether the stored value is equal to the original value. 4716 TruncatedValue = TruncatedValue.extend(OriginalWidth); 4717 if (llvm::APSInt::isSameValue(Value, TruncatedValue)) 4718 return false; 4719 4720 // Special-case bitfields of width 1: booleans are naturally 0/1, and 4721 // therefore don't strictly fit into a signed bitfield of width 1. 4722 if (FieldWidth == 1 && Value == 1) 4723 return false; 4724 4725 std::string PrettyValue = Value.toString(10); 4726 std::string PrettyTrunc = TruncatedValue.toString(10); 4727 4728 S.Diag(InitLoc, diag::warn_impcast_bitfield_precision_constant) 4729 << PrettyValue << PrettyTrunc << OriginalInit->getType() 4730 << Init->getSourceRange(); 4731 4732 return true; 4733} 4734 4735/// Analyze the given simple or compound assignment for warning-worthy 4736/// operations. 4737static void AnalyzeAssignment(Sema &S, BinaryOperator *E) { 4738 // Just recurse on the LHS. 4739 AnalyzeImplicitConversions(S, E->getLHS(), E->getOperatorLoc()); 4740 4741 // We want to recurse on the RHS as normal unless we're assigning to 4742 // a bitfield. 4743 if (FieldDecl *Bitfield = E->getLHS()->getSourceBitField()) { 4744 if (AnalyzeBitFieldAssignment(S, Bitfield, E->getRHS(), 4745 E->getOperatorLoc())) { 4746 // Recurse, ignoring any implicit conversions on the RHS. 4747 return AnalyzeImplicitConversions(S, E->getRHS()->IgnoreParenImpCasts(), 4748 E->getOperatorLoc()); 4749 } 4750 } 4751 4752 AnalyzeImplicitConversions(S, E->getRHS(), E->getOperatorLoc()); 4753} 4754 4755/// Diagnose an implicit cast; purely a helper for CheckImplicitConversion. 4756static void DiagnoseImpCast(Sema &S, Expr *E, QualType SourceType, QualType T, 4757 SourceLocation CContext, unsigned diag, 4758 bool pruneControlFlow = false) { 4759 if (pruneControlFlow) { 4760 S.DiagRuntimeBehavior(E->getExprLoc(), E, 4761 S.PDiag(diag) 4762 << SourceType << T << E->getSourceRange() 4763 << SourceRange(CContext)); 4764 return; 4765 } 4766 S.Diag(E->getExprLoc(), diag) 4767 << SourceType << T << E->getSourceRange() << SourceRange(CContext); 4768} 4769 4770/// Diagnose an implicit cast; purely a helper for CheckImplicitConversion. 4771static void DiagnoseImpCast(Sema &S, Expr *E, QualType T, 4772 SourceLocation CContext, unsigned diag, 4773 bool pruneControlFlow = false) { 4774 DiagnoseImpCast(S, E, E->getType(), T, CContext, diag, pruneControlFlow); 4775} 4776 4777/// Diagnose an implicit cast from a literal expression. Does not warn when the 4778/// cast wouldn't lose information. 4779void DiagnoseFloatingLiteralImpCast(Sema &S, FloatingLiteral *FL, QualType T, 4780 SourceLocation CContext) { 4781 // Try to convert the literal exactly to an integer. If we can, don't warn. 4782 bool isExact = false; 4783 const llvm::APFloat &Value = FL->getValue(); 4784 llvm::APSInt IntegerValue(S.Context.getIntWidth(T), 4785 T->hasUnsignedIntegerRepresentation()); 4786 if (Value.convertToInteger(IntegerValue, 4787 llvm::APFloat::rmTowardZero, &isExact) 4788 == llvm::APFloat::opOK && isExact) 4789 return; 4790 4791 SmallString<16> PrettySourceValue; 4792 Value.toString(PrettySourceValue); 4793 SmallString<16> PrettyTargetValue; 4794 if (T->isSpecificBuiltinType(BuiltinType::Bool)) 4795 PrettyTargetValue = IntegerValue == 0 ? "false" : "true"; 4796 else 4797 IntegerValue.toString(PrettyTargetValue); 4798 4799 S.Diag(FL->getExprLoc(), diag::warn_impcast_literal_float_to_integer) 4800 << FL->getType() << T.getUnqualifiedType() << PrettySourceValue 4801 << PrettyTargetValue << FL->getSourceRange() << SourceRange(CContext); 4802} 4803 4804std::string PrettyPrintInRange(const llvm::APSInt &Value, IntRange Range) { 4805 if (!Range.Width) return "0"; 4806 4807 llvm::APSInt ValueInRange = Value; 4808 ValueInRange.setIsSigned(!Range.NonNegative); 4809 ValueInRange = ValueInRange.trunc(Range.Width); 4810 return ValueInRange.toString(10); 4811} 4812 4813static bool IsImplicitBoolFloatConversion(Sema &S, Expr *Ex, bool ToBool) { 4814 if (!isa<ImplicitCastExpr>(Ex)) 4815 return false; 4816 4817 Expr *InnerE = Ex->IgnoreParenImpCasts(); 4818 const Type *Target = S.Context.getCanonicalType(Ex->getType()).getTypePtr(); 4819 const Type *Source = 4820 S.Context.getCanonicalType(InnerE->getType()).getTypePtr(); 4821 if (Target->isDependentType()) 4822 return false; 4823 4824 const BuiltinType *FloatCandidateBT = 4825 dyn_cast<BuiltinType>(ToBool ? Source : Target); 4826 const Type *BoolCandidateType = ToBool ? Target : Source; 4827 4828 return (BoolCandidateType->isSpecificBuiltinType(BuiltinType::Bool) && 4829 FloatCandidateBT && (FloatCandidateBT->isFloatingPoint())); 4830} 4831 4832void CheckImplicitArgumentConversions(Sema &S, CallExpr *TheCall, 4833 SourceLocation CC) { 4834 unsigned NumArgs = TheCall->getNumArgs(); 4835 for (unsigned i = 0; i < NumArgs; ++i) { 4836 Expr *CurrA = TheCall->getArg(i); 4837 if (!IsImplicitBoolFloatConversion(S, CurrA, true)) 4838 continue; 4839 4840 bool IsSwapped = ((i > 0) && 4841 IsImplicitBoolFloatConversion(S, TheCall->getArg(i - 1), false)); 4842 IsSwapped |= ((i < (NumArgs - 1)) && 4843 IsImplicitBoolFloatConversion(S, TheCall->getArg(i + 1), false)); 4844 if (IsSwapped) { 4845 // Warn on this floating-point to bool conversion. 4846 DiagnoseImpCast(S, CurrA->IgnoreParenImpCasts(), 4847 CurrA->getType(), CC, 4848 diag::warn_impcast_floating_point_to_bool); 4849 } 4850 } 4851} 4852 4853void CheckImplicitConversion(Sema &S, Expr *E, QualType T, 4854 SourceLocation CC, bool *ICContext = 0) { 4855 if (E->isTypeDependent() || E->isValueDependent()) return; 4856 4857 const Type *Source = S.Context.getCanonicalType(E->getType()).getTypePtr(); 4858 const Type *Target = S.Context.getCanonicalType(T).getTypePtr(); 4859 if (Source == Target) return; 4860 if (Target->isDependentType()) return; 4861 4862 // If the conversion context location is invalid don't complain. We also 4863 // don't want to emit a warning if the issue occurs from the expansion of 4864 // a system macro. The problem is that 'getSpellingLoc()' is slow, so we 4865 // delay this check as long as possible. Once we detect we are in that 4866 // scenario, we just return. 4867 if (CC.isInvalid()) 4868 return; 4869 4870 // Diagnose implicit casts to bool. 4871 if (Target->isSpecificBuiltinType(BuiltinType::Bool)) { 4872 if (isa<StringLiteral>(E)) 4873 // Warn on string literal to bool. Checks for string literals in logical 4874 // expressions, for instances, assert(0 && "error here"), is prevented 4875 // by a check in AnalyzeImplicitConversions(). 4876 return DiagnoseImpCast(S, E, T, CC, 4877 diag::warn_impcast_string_literal_to_bool); 4878 if (Source->isFunctionType()) { 4879 // Warn on function to bool. Checks free functions and static member 4880 // functions. Weakly imported functions are excluded from the check, 4881 // since it's common to test their value to check whether the linker 4882 // found a definition for them. 4883 ValueDecl *D = 0; 4884 if (DeclRefExpr* R = dyn_cast<DeclRefExpr>(E)) { 4885 D = R->getDecl(); 4886 } else if (MemberExpr *M = dyn_cast<MemberExpr>(E)) { 4887 D = M->getMemberDecl(); 4888 } 4889 4890 if (D && !D->isWeak()) { 4891 if (FunctionDecl* F = dyn_cast<FunctionDecl>(D)) { 4892 S.Diag(E->getExprLoc(), diag::warn_impcast_function_to_bool) 4893 << F << E->getSourceRange() << SourceRange(CC); 4894 S.Diag(E->getExprLoc(), diag::note_function_to_bool_silence) 4895 << FixItHint::CreateInsertion(E->getExprLoc(), "&"); 4896 QualType ReturnType; 4897 UnresolvedSet<4> NonTemplateOverloads; 4898 S.tryExprAsCall(*E, ReturnType, NonTemplateOverloads); 4899 if (!ReturnType.isNull() 4900 && ReturnType->isSpecificBuiltinType(BuiltinType::Bool)) 4901 S.Diag(E->getExprLoc(), diag::note_function_to_bool_call) 4902 << FixItHint::CreateInsertion( 4903 S.getPreprocessor().getLocForEndOfToken(E->getLocEnd()), "()"); 4904 return; 4905 } 4906 } 4907 } 4908 } 4909 4910 // Strip vector types. 4911 if (isa<VectorType>(Source)) { 4912 if (!isa<VectorType>(Target)) { 4913 if (S.SourceMgr.isInSystemMacro(CC)) 4914 return; 4915 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_vector_scalar); 4916 } 4917 4918 // If the vector cast is cast between two vectors of the same size, it is 4919 // a bitcast, not a conversion. 4920 if (S.Context.getTypeSize(Source) == S.Context.getTypeSize(Target)) 4921 return; 4922 4923 Source = cast<VectorType>(Source)->getElementType().getTypePtr(); 4924 Target = cast<VectorType>(Target)->getElementType().getTypePtr(); 4925 } 4926 4927 // Strip complex types. 4928 if (isa<ComplexType>(Source)) { 4929 if (!isa<ComplexType>(Target)) { 4930 if (S.SourceMgr.isInSystemMacro(CC)) 4931 return; 4932 4933 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_complex_scalar); 4934 } 4935 4936 Source = cast<ComplexType>(Source)->getElementType().getTypePtr(); 4937 Target = cast<ComplexType>(Target)->getElementType().getTypePtr(); 4938 } 4939 4940 const BuiltinType *SourceBT = dyn_cast<BuiltinType>(Source); 4941 const BuiltinType *TargetBT = dyn_cast<BuiltinType>(Target); 4942 4943 // If the source is floating point... 4944 if (SourceBT && SourceBT->isFloatingPoint()) { 4945 // ...and the target is floating point... 4946 if (TargetBT && TargetBT->isFloatingPoint()) { 4947 // ...then warn if we're dropping FP rank. 4948 4949 // Builtin FP kinds are ordered by increasing FP rank. 4950 if (SourceBT->getKind() > TargetBT->getKind()) { 4951 // Don't warn about float constants that are precisely 4952 // representable in the target type. 4953 Expr::EvalResult result; 4954 if (E->EvaluateAsRValue(result, S.Context)) { 4955 // Value might be a float, a float vector, or a float complex. 4956 if (IsSameFloatAfterCast(result.Val, 4957 S.Context.getFloatTypeSemantics(QualType(TargetBT, 0)), 4958 S.Context.getFloatTypeSemantics(QualType(SourceBT, 0)))) 4959 return; 4960 } 4961 4962 if (S.SourceMgr.isInSystemMacro(CC)) 4963 return; 4964 4965 DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_float_precision); 4966 } 4967 return; 4968 } 4969 4970 // If the target is integral, always warn. 4971 if (TargetBT && TargetBT->isInteger()) { 4972 if (S.SourceMgr.isInSystemMacro(CC)) 4973 return; 4974 4975 Expr *InnerE = E->IgnoreParenImpCasts(); 4976 // We also want to warn on, e.g., "int i = -1.234" 4977 if (UnaryOperator *UOp = dyn_cast<UnaryOperator>(InnerE)) 4978 if (UOp->getOpcode() == UO_Minus || UOp->getOpcode() == UO_Plus) 4979 InnerE = UOp->getSubExpr()->IgnoreParenImpCasts(); 4980 4981 if (FloatingLiteral *FL = dyn_cast<FloatingLiteral>(InnerE)) { 4982 DiagnoseFloatingLiteralImpCast(S, FL, T, CC); 4983 } else { 4984 DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_float_integer); 4985 } 4986 } 4987 4988 // If the target is bool, warn if expr is a function or method call. 4989 if (Target->isSpecificBuiltinType(BuiltinType::Bool) && 4990 isa<CallExpr>(E)) { 4991 // Check last argument of function call to see if it is an 4992 // implicit cast from a type matching the type the result 4993 // is being cast to. 4994 CallExpr *CEx = cast<CallExpr>(E); 4995 unsigned NumArgs = CEx->getNumArgs(); 4996 if (NumArgs > 0) { 4997 Expr *LastA = CEx->getArg(NumArgs - 1); 4998 Expr *InnerE = LastA->IgnoreParenImpCasts(); 4999 const Type *InnerType = 5000 S.Context.getCanonicalType(InnerE->getType()).getTypePtr(); 5001 if (isa<ImplicitCastExpr>(LastA) && (InnerType == Target)) { 5002 // Warn on this floating-point to bool conversion 5003 DiagnoseImpCast(S, E, T, CC, 5004 diag::warn_impcast_floating_point_to_bool); 5005 } 5006 } 5007 } 5008 return; 5009 } 5010 5011 if ((E->isNullPointerConstant(S.Context, Expr::NPC_ValueDependentIsNotNull) 5012 == Expr::NPCK_GNUNull) && !Target->isAnyPointerType() 5013 && !Target->isBlockPointerType() && !Target->isMemberPointerType() 5014 && Target->isScalarType() && !Target->isNullPtrType()) { 5015 SourceLocation Loc = E->getSourceRange().getBegin(); 5016 if (Loc.isMacroID()) 5017 Loc = S.SourceMgr.getImmediateExpansionRange(Loc).first; 5018 if (!Loc.isMacroID() || CC.isMacroID()) 5019 S.Diag(Loc, diag::warn_impcast_null_pointer_to_integer) 5020 << T << clang::SourceRange(CC) 5021 << FixItHint::CreateReplacement(Loc, S.getFixItZeroLiteralForType(T)); 5022 } 5023 5024 if (!Source->isIntegerType() || !Target->isIntegerType()) 5025 return; 5026 5027 // TODO: remove this early return once the false positives for constant->bool 5028 // in templates, macros, etc, are reduced or removed. 5029 if (Target->isSpecificBuiltinType(BuiltinType::Bool)) 5030 return; 5031 5032 IntRange SourceRange = GetExprRange(S.Context, E); 5033 IntRange TargetRange = IntRange::forTargetOfCanonicalType(S.Context, Target); 5034 5035 if (SourceRange.Width > TargetRange.Width) { 5036 // If the source is a constant, use a default-on diagnostic. 5037 // TODO: this should happen for bitfield stores, too. 5038 llvm::APSInt Value(32); 5039 if (E->isIntegerConstantExpr(Value, S.Context)) { 5040 if (S.SourceMgr.isInSystemMacro(CC)) 5041 return; 5042 5043 std::string PrettySourceValue = Value.toString(10); 5044 std::string PrettyTargetValue = PrettyPrintInRange(Value, TargetRange); 5045 5046 S.DiagRuntimeBehavior(E->getExprLoc(), E, 5047 S.PDiag(diag::warn_impcast_integer_precision_constant) 5048 << PrettySourceValue << PrettyTargetValue 5049 << E->getType() << T << E->getSourceRange() 5050 << clang::SourceRange(CC)); 5051 return; 5052 } 5053 5054 // People want to build with -Wshorten-64-to-32 and not -Wconversion. 5055 if (S.SourceMgr.isInSystemMacro(CC)) 5056 return; 5057 5058 if (TargetRange.Width == 32 && S.Context.getIntWidth(E->getType()) == 64) 5059 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_integer_64_32, 5060 /* pruneControlFlow */ true); 5061 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_integer_precision); 5062 } 5063 5064 if ((TargetRange.NonNegative && !SourceRange.NonNegative) || 5065 (!TargetRange.NonNegative && SourceRange.NonNegative && 5066 SourceRange.Width == TargetRange.Width)) { 5067 5068 if (S.SourceMgr.isInSystemMacro(CC)) 5069 return; 5070 5071 unsigned DiagID = diag::warn_impcast_integer_sign; 5072 5073 // Traditionally, gcc has warned about this under -Wsign-compare. 5074 // We also want to warn about it in -Wconversion. 5075 // So if -Wconversion is off, use a completely identical diagnostic 5076 // in the sign-compare group. 5077 // The conditional-checking code will 5078 if (ICContext) { 5079 DiagID = diag::warn_impcast_integer_sign_conditional; 5080 *ICContext = true; 5081 } 5082 5083 return DiagnoseImpCast(S, E, T, CC, DiagID); 5084 } 5085 5086 // Diagnose conversions between different enumeration types. 5087 // In C, we pretend that the type of an EnumConstantDecl is its enumeration 5088 // type, to give us better diagnostics. 5089 QualType SourceType = E->getType(); 5090 if (!S.getLangOpts().CPlusPlus) { 5091 if (DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(E)) 5092 if (EnumConstantDecl *ECD = dyn_cast<EnumConstantDecl>(DRE->getDecl())) { 5093 EnumDecl *Enum = cast<EnumDecl>(ECD->getDeclContext()); 5094 SourceType = S.Context.getTypeDeclType(Enum); 5095 Source = S.Context.getCanonicalType(SourceType).getTypePtr(); 5096 } 5097 } 5098 5099 if (const EnumType *SourceEnum = Source->getAs<EnumType>()) 5100 if (const EnumType *TargetEnum = Target->getAs<EnumType>()) 5101 if (SourceEnum->getDecl()->hasNameForLinkage() && 5102 TargetEnum->getDecl()->hasNameForLinkage() && 5103 SourceEnum != TargetEnum) { 5104 if (S.SourceMgr.isInSystemMacro(CC)) 5105 return; 5106 5107 return DiagnoseImpCast(S, E, SourceType, T, CC, 5108 diag::warn_impcast_different_enum_types); 5109 } 5110 5111 return; 5112} 5113 5114void CheckConditionalOperator(Sema &S, ConditionalOperator *E, 5115 SourceLocation CC, QualType T); 5116 5117void CheckConditionalOperand(Sema &S, Expr *E, QualType T, 5118 SourceLocation CC, bool &ICContext) { 5119 E = E->IgnoreParenImpCasts(); 5120 5121 if (isa<ConditionalOperator>(E)) 5122 return CheckConditionalOperator(S, cast<ConditionalOperator>(E), CC, T); 5123 5124 AnalyzeImplicitConversions(S, E, CC); 5125 if (E->getType() != T) 5126 return CheckImplicitConversion(S, E, T, CC, &ICContext); 5127 return; 5128} 5129 5130void CheckConditionalOperator(Sema &S, ConditionalOperator *E, 5131 SourceLocation CC, QualType T) { 5132 AnalyzeImplicitConversions(S, E->getCond(), CC); 5133 5134 bool Suspicious = false; 5135 CheckConditionalOperand(S, E->getTrueExpr(), T, CC, Suspicious); 5136 CheckConditionalOperand(S, E->getFalseExpr(), T, CC, Suspicious); 5137 5138 // If -Wconversion would have warned about either of the candidates 5139 // for a signedness conversion to the context type... 5140 if (!Suspicious) return; 5141 5142 // ...but it's currently ignored... 5143 if (S.Diags.getDiagnosticLevel(diag::warn_impcast_integer_sign_conditional, 5144 CC)) 5145 return; 5146 5147 // ...then check whether it would have warned about either of the 5148 // candidates for a signedness conversion to the condition type. 5149 if (E->getType() == T) return; 5150 5151 Suspicious = false; 5152 CheckImplicitConversion(S, E->getTrueExpr()->IgnoreParenImpCasts(), 5153 E->getType(), CC, &Suspicious); 5154 if (!Suspicious) 5155 CheckImplicitConversion(S, E->getFalseExpr()->IgnoreParenImpCasts(), 5156 E->getType(), CC, &Suspicious); 5157} 5158 5159/// AnalyzeImplicitConversions - Find and report any interesting 5160/// implicit conversions in the given expression. There are a couple 5161/// of competing diagnostics here, -Wconversion and -Wsign-compare. 5162void AnalyzeImplicitConversions(Sema &S, Expr *OrigE, SourceLocation CC) { 5163 QualType T = OrigE->getType(); 5164 Expr *E = OrigE->IgnoreParenImpCasts(); 5165 5166 if (E->isTypeDependent() || E->isValueDependent()) 5167 return; 5168 5169 // For conditional operators, we analyze the arguments as if they 5170 // were being fed directly into the output. 5171 if (isa<ConditionalOperator>(E)) { 5172 ConditionalOperator *CO = cast<ConditionalOperator>(E); 5173 CheckConditionalOperator(S, CO, CC, T); 5174 return; 5175 } 5176 5177 // Check implicit argument conversions for function calls. 5178 if (CallExpr *Call = dyn_cast<CallExpr>(E)) 5179 CheckImplicitArgumentConversions(S, Call, CC); 5180 5181 // Go ahead and check any implicit conversions we might have skipped. 5182 // The non-canonical typecheck is just an optimization; 5183 // CheckImplicitConversion will filter out dead implicit conversions. 5184 if (E->getType() != T) 5185 CheckImplicitConversion(S, E, T, CC); 5186 5187 // Now continue drilling into this expression. 5188 5189 if (PseudoObjectExpr * POE = dyn_cast<PseudoObjectExpr>(E)) { 5190 if (POE->getResultExpr()) 5191 E = POE->getResultExpr(); 5192 } 5193 5194 if (const OpaqueValueExpr *OVE = dyn_cast<OpaqueValueExpr>(E)) 5195 return AnalyzeImplicitConversions(S, OVE->getSourceExpr(), CC); 5196 5197 // Skip past explicit casts. 5198 if (isa<ExplicitCastExpr>(E)) { 5199 E = cast<ExplicitCastExpr>(E)->getSubExpr()->IgnoreParenImpCasts(); 5200 return AnalyzeImplicitConversions(S, E, CC); 5201 } 5202 5203 if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) { 5204 // Do a somewhat different check with comparison operators. 5205 if (BO->isComparisonOp()) 5206 return AnalyzeComparison(S, BO); 5207 5208 // And with simple assignments. 5209 if (BO->getOpcode() == BO_Assign) 5210 return AnalyzeAssignment(S, BO); 5211 } 5212 5213 // These break the otherwise-useful invariant below. Fortunately, 5214 // we don't really need to recurse into them, because any internal 5215 // expressions should have been analyzed already when they were 5216 // built into statements. 5217 if (isa<StmtExpr>(E)) return; 5218 5219 // Don't descend into unevaluated contexts. 5220 if (isa<UnaryExprOrTypeTraitExpr>(E)) return; 5221 5222 // Now just recurse over the expression's children. 5223 CC = E->getExprLoc(); 5224 BinaryOperator *BO = dyn_cast<BinaryOperator>(E); 5225 bool IsLogicalOperator = BO && BO->isLogicalOp(); 5226 for (Stmt::child_range I = E->children(); I; ++I) { 5227 Expr *ChildExpr = dyn_cast_or_null<Expr>(*I); 5228 if (!ChildExpr) 5229 continue; 5230 5231 if (IsLogicalOperator && 5232 isa<StringLiteral>(ChildExpr->IgnoreParenImpCasts())) 5233 // Ignore checking string literals that are in logical operators. 5234 continue; 5235 AnalyzeImplicitConversions(S, ChildExpr, CC); 5236 } 5237} 5238 5239} // end anonymous namespace 5240 5241/// Diagnoses "dangerous" implicit conversions within the given 5242/// expression (which is a full expression). Implements -Wconversion 5243/// and -Wsign-compare. 5244/// 5245/// \param CC the "context" location of the implicit conversion, i.e. 5246/// the most location of the syntactic entity requiring the implicit 5247/// conversion 5248void Sema::CheckImplicitConversions(Expr *E, SourceLocation CC) { 5249 // Don't diagnose in unevaluated contexts. 5250 if (isUnevaluatedContext()) 5251 return; 5252 5253 // Don't diagnose for value- or type-dependent expressions. 5254 if (E->isTypeDependent() || E->isValueDependent()) 5255 return; 5256 5257 // Check for array bounds violations in cases where the check isn't triggered 5258 // elsewhere for other Expr types (like BinaryOperators), e.g. when an 5259 // ArraySubscriptExpr is on the RHS of a variable initialization. 5260 CheckArrayAccess(E); 5261 5262 // This is not the right CC for (e.g.) a variable initialization. 5263 AnalyzeImplicitConversions(*this, E, CC); 5264} 5265 5266/// Diagnose when expression is an integer constant expression and its evaluation 5267/// results in integer overflow 5268void Sema::CheckForIntOverflow (Expr *E) { 5269 if (isa<BinaryOperator>(E->IgnoreParens())) { 5270 llvm::SmallVector<PartialDiagnosticAt, 4> Diags; 5271 E->EvaluateForOverflow(Context, &Diags); 5272 } 5273} 5274 5275namespace { 5276/// \brief Visitor for expressions which looks for unsequenced operations on the 5277/// same object. 5278class SequenceChecker : public EvaluatedExprVisitor<SequenceChecker> { 5279 typedef EvaluatedExprVisitor<SequenceChecker> Base; 5280 5281 /// \brief A tree of sequenced regions within an expression. Two regions are 5282 /// unsequenced if one is an ancestor or a descendent of the other. When we 5283 /// finish processing an expression with sequencing, such as a comma 5284 /// expression, we fold its tree nodes into its parent, since they are 5285 /// unsequenced with respect to nodes we will visit later. 5286 class SequenceTree { 5287 struct Value { 5288 explicit Value(unsigned Parent) : Parent(Parent), Merged(false) {} 5289 unsigned Parent : 31; 5290 bool Merged : 1; 5291 }; 5292 llvm::SmallVector<Value, 8> Values; 5293 5294 public: 5295 /// \brief A region within an expression which may be sequenced with respect 5296 /// to some other region. 5297 class Seq { 5298 explicit Seq(unsigned N) : Index(N) {} 5299 unsigned Index; 5300 friend class SequenceTree; 5301 public: 5302 Seq() : Index(0) {} 5303 }; 5304 5305 SequenceTree() { Values.push_back(Value(0)); } 5306 Seq root() const { return Seq(0); } 5307 5308 /// \brief Create a new sequence of operations, which is an unsequenced 5309 /// subset of \p Parent. This sequence of operations is sequenced with 5310 /// respect to other children of \p Parent. 5311 Seq allocate(Seq Parent) { 5312 Values.push_back(Value(Parent.Index)); 5313 return Seq(Values.size() - 1); 5314 } 5315 5316 /// \brief Merge a sequence of operations into its parent. 5317 void merge(Seq S) { 5318 Values[S.Index].Merged = true; 5319 } 5320 5321 /// \brief Determine whether two operations are unsequenced. This operation 5322 /// is asymmetric: \p Cur should be the more recent sequence, and \p Old 5323 /// should have been merged into its parent as appropriate. 5324 bool isUnsequenced(Seq Cur, Seq Old) { 5325 unsigned C = representative(Cur.Index); 5326 unsigned Target = representative(Old.Index); 5327 while (C >= Target) { 5328 if (C == Target) 5329 return true; 5330 C = Values[C].Parent; 5331 } 5332 return false; 5333 } 5334 5335 private: 5336 /// \brief Pick a representative for a sequence. 5337 unsigned representative(unsigned K) { 5338 if (Values[K].Merged) 5339 // Perform path compression as we go. 5340 return Values[K].Parent = representative(Values[K].Parent); 5341 return K; 5342 } 5343 }; 5344 5345 /// An object for which we can track unsequenced uses. 5346 typedef NamedDecl *Object; 5347 5348 /// Different flavors of object usage which we track. We only track the 5349 /// least-sequenced usage of each kind. 5350 enum UsageKind { 5351 /// A read of an object. Multiple unsequenced reads are OK. 5352 UK_Use, 5353 /// A modification of an object which is sequenced before the value 5354 /// computation of the expression, such as ++n in C++. 5355 UK_ModAsValue, 5356 /// A modification of an object which is not sequenced before the value 5357 /// computation of the expression, such as n++. 5358 UK_ModAsSideEffect, 5359 5360 UK_Count = UK_ModAsSideEffect + 1 5361 }; 5362 5363 struct Usage { 5364 Usage() : Use(0), Seq() {} 5365 Expr *Use; 5366 SequenceTree::Seq Seq; 5367 }; 5368 5369 struct UsageInfo { 5370 UsageInfo() : Diagnosed(false) {} 5371 Usage Uses[UK_Count]; 5372 /// Have we issued a diagnostic for this variable already? 5373 bool Diagnosed; 5374 }; 5375 typedef llvm::SmallDenseMap<Object, UsageInfo, 16> UsageInfoMap; 5376 5377 Sema &SemaRef; 5378 /// Sequenced regions within the expression. 5379 SequenceTree Tree; 5380 /// Declaration modifications and references which we have seen. 5381 UsageInfoMap UsageMap; 5382 /// The region we are currently within. 5383 SequenceTree::Seq Region; 5384 /// Filled in with declarations which were modified as a side-effect 5385 /// (that is, post-increment operations). 5386 llvm::SmallVectorImpl<std::pair<Object, Usage> > *ModAsSideEffect; 5387 /// Expressions to check later. We defer checking these to reduce 5388 /// stack usage. 5389 llvm::SmallVectorImpl<Expr*> &WorkList; 5390 5391 /// RAII object wrapping the visitation of a sequenced subexpression of an 5392 /// expression. At the end of this process, the side-effects of the evaluation 5393 /// become sequenced with respect to the value computation of the result, so 5394 /// we downgrade any UK_ModAsSideEffect within the evaluation to 5395 /// UK_ModAsValue. 5396 struct SequencedSubexpression { 5397 SequencedSubexpression(SequenceChecker &Self) 5398 : Self(Self), OldModAsSideEffect(Self.ModAsSideEffect) { 5399 Self.ModAsSideEffect = &ModAsSideEffect; 5400 } 5401 ~SequencedSubexpression() { 5402 for (unsigned I = 0, E = ModAsSideEffect.size(); I != E; ++I) { 5403 UsageInfo &U = Self.UsageMap[ModAsSideEffect[I].first]; 5404 U.Uses[UK_ModAsSideEffect] = ModAsSideEffect[I].second; 5405 Self.addUsage(U, ModAsSideEffect[I].first, 5406 ModAsSideEffect[I].second.Use, UK_ModAsValue); 5407 } 5408 Self.ModAsSideEffect = OldModAsSideEffect; 5409 } 5410 5411 SequenceChecker &Self; 5412 llvm::SmallVector<std::pair<Object, Usage>, 4> ModAsSideEffect; 5413 llvm::SmallVectorImpl<std::pair<Object, Usage> > *OldModAsSideEffect; 5414 }; 5415 5416 /// RAII object wrapping the visitation of a subexpression which we might 5417 /// choose to evaluate as a constant. If any subexpression is evaluated and 5418 /// found to be non-constant, this allows us to suppress the evaluation of 5419 /// the outer expression. 5420 class EvaluationTracker { 5421 public: 5422 EvaluationTracker(SequenceChecker &Self) 5423 : Self(Self), Prev(Self.EvalTracker), EvalOK(true) { 5424 Self.EvalTracker = this; 5425 } 5426 ~EvaluationTracker() { 5427 Self.EvalTracker = Prev; 5428 if (Prev) 5429 Prev->EvalOK &= EvalOK; 5430 } 5431 5432 bool evaluate(const Expr *E, bool &Result) { 5433 if (!EvalOK || E->isValueDependent()) 5434 return false; 5435 EvalOK = E->EvaluateAsBooleanCondition(Result, Self.SemaRef.Context); 5436 return EvalOK; 5437 } 5438 5439 private: 5440 SequenceChecker &Self; 5441 EvaluationTracker *Prev; 5442 bool EvalOK; 5443 } *EvalTracker; 5444 5445 /// \brief Find the object which is produced by the specified expression, 5446 /// if any. 5447 Object getObject(Expr *E, bool Mod) const { 5448 E = E->IgnoreParenCasts(); 5449 if (UnaryOperator *UO = dyn_cast<UnaryOperator>(E)) { 5450 if (Mod && (UO->getOpcode() == UO_PreInc || UO->getOpcode() == UO_PreDec)) 5451 return getObject(UO->getSubExpr(), Mod); 5452 } else if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) { 5453 if (BO->getOpcode() == BO_Comma) 5454 return getObject(BO->getRHS(), Mod); 5455 if (Mod && BO->isAssignmentOp()) 5456 return getObject(BO->getLHS(), Mod); 5457 } else if (MemberExpr *ME = dyn_cast<MemberExpr>(E)) { 5458 // FIXME: Check for more interesting cases, like "x.n = ++x.n". 5459 if (isa<CXXThisExpr>(ME->getBase()->IgnoreParenCasts())) 5460 return ME->getMemberDecl(); 5461 } else if (DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(E)) 5462 // FIXME: If this is a reference, map through to its value. 5463 return DRE->getDecl(); 5464 return 0; 5465 } 5466 5467 /// \brief Note that an object was modified or used by an expression. 5468 void addUsage(UsageInfo &UI, Object O, Expr *Ref, UsageKind UK) { 5469 Usage &U = UI.Uses[UK]; 5470 if (!U.Use || !Tree.isUnsequenced(Region, U.Seq)) { 5471 if (UK == UK_ModAsSideEffect && ModAsSideEffect) 5472 ModAsSideEffect->push_back(std::make_pair(O, U)); 5473 U.Use = Ref; 5474 U.Seq = Region; 5475 } 5476 } 5477 /// \brief Check whether a modification or use conflicts with a prior usage. 5478 void checkUsage(Object O, UsageInfo &UI, Expr *Ref, UsageKind OtherKind, 5479 bool IsModMod) { 5480 if (UI.Diagnosed) 5481 return; 5482 5483 const Usage &U = UI.Uses[OtherKind]; 5484 if (!U.Use || !Tree.isUnsequenced(Region, U.Seq)) 5485 return; 5486 5487 Expr *Mod = U.Use; 5488 Expr *ModOrUse = Ref; 5489 if (OtherKind == UK_Use) 5490 std::swap(Mod, ModOrUse); 5491 5492 SemaRef.Diag(Mod->getExprLoc(), 5493 IsModMod ? diag::warn_unsequenced_mod_mod 5494 : diag::warn_unsequenced_mod_use) 5495 << O << SourceRange(ModOrUse->getExprLoc()); 5496 UI.Diagnosed = true; 5497 } 5498 5499 void notePreUse(Object O, Expr *Use) { 5500 UsageInfo &U = UsageMap[O]; 5501 // Uses conflict with other modifications. 5502 checkUsage(O, U, Use, UK_ModAsValue, false); 5503 } 5504 void notePostUse(Object O, Expr *Use) { 5505 UsageInfo &U = UsageMap[O]; 5506 checkUsage(O, U, Use, UK_ModAsSideEffect, false); 5507 addUsage(U, O, Use, UK_Use); 5508 } 5509 5510 void notePreMod(Object O, Expr *Mod) { 5511 UsageInfo &U = UsageMap[O]; 5512 // Modifications conflict with other modifications and with uses. 5513 checkUsage(O, U, Mod, UK_ModAsValue, true); 5514 checkUsage(O, U, Mod, UK_Use, false); 5515 } 5516 void notePostMod(Object O, Expr *Use, UsageKind UK) { 5517 UsageInfo &U = UsageMap[O]; 5518 checkUsage(O, U, Use, UK_ModAsSideEffect, true); 5519 addUsage(U, O, Use, UK); 5520 } 5521 5522public: 5523 SequenceChecker(Sema &S, Expr *E, 5524 llvm::SmallVectorImpl<Expr*> &WorkList) 5525 : Base(S.Context), SemaRef(S), Region(Tree.root()), 5526 ModAsSideEffect(0), WorkList(WorkList), EvalTracker(0) { 5527 Visit(E); 5528 } 5529 5530 void VisitStmt(Stmt *S) { 5531 // Skip all statements which aren't expressions for now. 5532 } 5533 5534 void VisitExpr(Expr *E) { 5535 // By default, just recurse to evaluated subexpressions. 5536 Base::VisitStmt(E); 5537 } 5538 5539 void VisitCastExpr(CastExpr *E) { 5540 Object O = Object(); 5541 if (E->getCastKind() == CK_LValueToRValue) 5542 O = getObject(E->getSubExpr(), false); 5543 5544 if (O) 5545 notePreUse(O, E); 5546 VisitExpr(E); 5547 if (O) 5548 notePostUse(O, E); 5549 } 5550 5551 void VisitBinComma(BinaryOperator *BO) { 5552 // C++11 [expr.comma]p1: 5553 // Every value computation and side effect associated with the left 5554 // expression is sequenced before every value computation and side 5555 // effect associated with the right expression. 5556 SequenceTree::Seq LHS = Tree.allocate(Region); 5557 SequenceTree::Seq RHS = Tree.allocate(Region); 5558 SequenceTree::Seq OldRegion = Region; 5559 5560 { 5561 SequencedSubexpression SeqLHS(*this); 5562 Region = LHS; 5563 Visit(BO->getLHS()); 5564 } 5565 5566 Region = RHS; 5567 Visit(BO->getRHS()); 5568 5569 Region = OldRegion; 5570 5571 // Forget that LHS and RHS are sequenced. They are both unsequenced 5572 // with respect to other stuff. 5573 Tree.merge(LHS); 5574 Tree.merge(RHS); 5575 } 5576 5577 void VisitBinAssign(BinaryOperator *BO) { 5578 // The modification is sequenced after the value computation of the LHS 5579 // and RHS, so check it before inspecting the operands and update the 5580 // map afterwards. 5581 Object O = getObject(BO->getLHS(), true); 5582 if (!O) 5583 return VisitExpr(BO); 5584 5585 notePreMod(O, BO); 5586 5587 // C++11 [expr.ass]p7: 5588 // E1 op= E2 is equivalent to E1 = E1 op E2, except that E1 is evaluated 5589 // only once. 5590 // 5591 // Therefore, for a compound assignment operator, O is considered used 5592 // everywhere except within the evaluation of E1 itself. 5593 if (isa<CompoundAssignOperator>(BO)) 5594 notePreUse(O, BO); 5595 5596 Visit(BO->getLHS()); 5597 5598 if (isa<CompoundAssignOperator>(BO)) 5599 notePostUse(O, BO); 5600 5601 Visit(BO->getRHS()); 5602 5603 // C++11 [expr.ass]p1: 5604 // the assignment is sequenced [...] before the value computation of the 5605 // assignment expression. 5606 // C11 6.5.16/3 has no such rule. 5607 notePostMod(O, BO, SemaRef.getLangOpts().CPlusPlus ? UK_ModAsValue 5608 : UK_ModAsSideEffect); 5609 } 5610 void VisitCompoundAssignOperator(CompoundAssignOperator *CAO) { 5611 VisitBinAssign(CAO); 5612 } 5613 5614 void VisitUnaryPreInc(UnaryOperator *UO) { VisitUnaryPreIncDec(UO); } 5615 void VisitUnaryPreDec(UnaryOperator *UO) { VisitUnaryPreIncDec(UO); } 5616 void VisitUnaryPreIncDec(UnaryOperator *UO) { 5617 Object O = getObject(UO->getSubExpr(), true); 5618 if (!O) 5619 return VisitExpr(UO); 5620 5621 notePreMod(O, UO); 5622 Visit(UO->getSubExpr()); 5623 // C++11 [expr.pre.incr]p1: 5624 // the expression ++x is equivalent to x+=1 5625 notePostMod(O, UO, SemaRef.getLangOpts().CPlusPlus ? UK_ModAsValue 5626 : UK_ModAsSideEffect); 5627 } 5628 5629 void VisitUnaryPostInc(UnaryOperator *UO) { VisitUnaryPostIncDec(UO); } 5630 void VisitUnaryPostDec(UnaryOperator *UO) { VisitUnaryPostIncDec(UO); } 5631 void VisitUnaryPostIncDec(UnaryOperator *UO) { 5632 Object O = getObject(UO->getSubExpr(), true); 5633 if (!O) 5634 return VisitExpr(UO); 5635 5636 notePreMod(O, UO); 5637 Visit(UO->getSubExpr()); 5638 notePostMod(O, UO, UK_ModAsSideEffect); 5639 } 5640 5641 /// Don't visit the RHS of '&&' or '||' if it might not be evaluated. 5642 void VisitBinLOr(BinaryOperator *BO) { 5643 // The side-effects of the LHS of an '&&' are sequenced before the 5644 // value computation of the RHS, and hence before the value computation 5645 // of the '&&' itself, unless the LHS evaluates to zero. We treat them 5646 // as if they were unconditionally sequenced. 5647 EvaluationTracker Eval(*this); 5648 { 5649 SequencedSubexpression Sequenced(*this); 5650 Visit(BO->getLHS()); 5651 } 5652 5653 bool Result; 5654 if (Eval.evaluate(BO->getLHS(), Result)) { 5655 if (!Result) 5656 Visit(BO->getRHS()); 5657 } else { 5658 // Check for unsequenced operations in the RHS, treating it as an 5659 // entirely separate evaluation. 5660 // 5661 // FIXME: If there are operations in the RHS which are unsequenced 5662 // with respect to operations outside the RHS, and those operations 5663 // are unconditionally evaluated, diagnose them. 5664 WorkList.push_back(BO->getRHS()); 5665 } 5666 } 5667 void VisitBinLAnd(BinaryOperator *BO) { 5668 EvaluationTracker Eval(*this); 5669 { 5670 SequencedSubexpression Sequenced(*this); 5671 Visit(BO->getLHS()); 5672 } 5673 5674 bool Result; 5675 if (Eval.evaluate(BO->getLHS(), Result)) { 5676 if (Result) 5677 Visit(BO->getRHS()); 5678 } else { 5679 WorkList.push_back(BO->getRHS()); 5680 } 5681 } 5682 5683 // Only visit the condition, unless we can be sure which subexpression will 5684 // be chosen. 5685 void VisitAbstractConditionalOperator(AbstractConditionalOperator *CO) { 5686 EvaluationTracker Eval(*this); 5687 { 5688 SequencedSubexpression Sequenced(*this); 5689 Visit(CO->getCond()); 5690 } 5691 5692 bool Result; 5693 if (Eval.evaluate(CO->getCond(), Result)) 5694 Visit(Result ? CO->getTrueExpr() : CO->getFalseExpr()); 5695 else { 5696 WorkList.push_back(CO->getTrueExpr()); 5697 WorkList.push_back(CO->getFalseExpr()); 5698 } 5699 } 5700 5701 void VisitCallExpr(CallExpr *CE) { 5702 // C++11 [intro.execution]p15: 5703 // When calling a function [...], every value computation and side effect 5704 // associated with any argument expression, or with the postfix expression 5705 // designating the called function, is sequenced before execution of every 5706 // expression or statement in the body of the function [and thus before 5707 // the value computation of its result]. 5708 SequencedSubexpression Sequenced(*this); 5709 Base::VisitCallExpr(CE); 5710 5711 // FIXME: CXXNewExpr and CXXDeleteExpr implicitly call functions. 5712 } 5713 5714 void VisitCXXConstructExpr(CXXConstructExpr *CCE) { 5715 // This is a call, so all subexpressions are sequenced before the result. 5716 SequencedSubexpression Sequenced(*this); 5717 5718 if (!CCE->isListInitialization()) 5719 return VisitExpr(CCE); 5720 5721 // In C++11, list initializations are sequenced. 5722 llvm::SmallVector<SequenceTree::Seq, 32> Elts; 5723 SequenceTree::Seq Parent = Region; 5724 for (CXXConstructExpr::arg_iterator I = CCE->arg_begin(), 5725 E = CCE->arg_end(); 5726 I != E; ++I) { 5727 Region = Tree.allocate(Parent); 5728 Elts.push_back(Region); 5729 Visit(*I); 5730 } 5731 5732 // Forget that the initializers are sequenced. 5733 Region = Parent; 5734 for (unsigned I = 0; I < Elts.size(); ++I) 5735 Tree.merge(Elts[I]); 5736 } 5737 5738 void VisitInitListExpr(InitListExpr *ILE) { 5739 if (!SemaRef.getLangOpts().CPlusPlus11) 5740 return VisitExpr(ILE); 5741 5742 // In C++11, list initializations are sequenced. 5743 llvm::SmallVector<SequenceTree::Seq, 32> Elts; 5744 SequenceTree::Seq Parent = Region; 5745 for (unsigned I = 0; I < ILE->getNumInits(); ++I) { 5746 Expr *E = ILE->getInit(I); 5747 if (!E) continue; 5748 Region = Tree.allocate(Parent); 5749 Elts.push_back(Region); 5750 Visit(E); 5751 } 5752 5753 // Forget that the initializers are sequenced. 5754 Region = Parent; 5755 for (unsigned I = 0; I < Elts.size(); ++I) 5756 Tree.merge(Elts[I]); 5757 } 5758}; 5759} 5760 5761void Sema::CheckUnsequencedOperations(Expr *E) { 5762 llvm::SmallVector<Expr*, 8> WorkList; 5763 WorkList.push_back(E); 5764 while (!WorkList.empty()) { 5765 Expr *Item = WorkList.back(); 5766 WorkList.pop_back(); 5767 SequenceChecker(*this, Item, WorkList); 5768 } 5769} 5770 5771void Sema::CheckCompletedExpr(Expr *E, SourceLocation CheckLoc, 5772 bool IsConstexpr) { 5773 CheckImplicitConversions(E, CheckLoc); 5774 CheckUnsequencedOperations(E); 5775 if (!IsConstexpr && !E->isValueDependent()) 5776 CheckForIntOverflow(E); 5777} 5778 5779void Sema::CheckBitFieldInitialization(SourceLocation InitLoc, 5780 FieldDecl *BitField, 5781 Expr *Init) { 5782 (void) AnalyzeBitFieldAssignment(*this, BitField, Init, InitLoc); 5783} 5784 5785/// CheckParmsForFunctionDef - Check that the parameters of the given 5786/// function are appropriate for the definition of a function. This 5787/// takes care of any checks that cannot be performed on the 5788/// declaration itself, e.g., that the types of each of the function 5789/// parameters are complete. 5790bool Sema::CheckParmsForFunctionDef(ParmVarDecl *const *P, 5791 ParmVarDecl *const *PEnd, 5792 bool CheckParameterNames) { 5793 bool HasInvalidParm = false; 5794 for (; P != PEnd; ++P) { 5795 ParmVarDecl *Param = *P; 5796 5797 // C99 6.7.5.3p4: the parameters in a parameter type list in a 5798 // function declarator that is part of a function definition of 5799 // that function shall not have incomplete type. 5800 // 5801 // This is also C++ [dcl.fct]p6. 5802 if (!Param->isInvalidDecl() && 5803 RequireCompleteType(Param->getLocation(), Param->getType(), 5804 diag::err_typecheck_decl_incomplete_type)) { 5805 Param->setInvalidDecl(); 5806 HasInvalidParm = true; 5807 } 5808 5809 // C99 6.9.1p5: If the declarator includes a parameter type list, the 5810 // declaration of each parameter shall include an identifier. 5811 if (CheckParameterNames && 5812 Param->getIdentifier() == 0 && 5813 !Param->isImplicit() && 5814 !getLangOpts().CPlusPlus) 5815 Diag(Param->getLocation(), diag::err_parameter_name_omitted); 5816 5817 // C99 6.7.5.3p12: 5818 // If the function declarator is not part of a definition of that 5819 // function, parameters may have incomplete type and may use the [*] 5820 // notation in their sequences of declarator specifiers to specify 5821 // variable length array types. 5822 QualType PType = Param->getOriginalType(); 5823 while (const ArrayType *AT = Context.getAsArrayType(PType)) { 5824 if (AT->getSizeModifier() == ArrayType::Star) { 5825 // FIXME: This diagnostic should point the '[*]' if source-location 5826 // information is added for it. 5827 Diag(Param->getLocation(), diag::err_array_star_in_function_definition); 5828 break; 5829 } 5830 PType= AT->getElementType(); 5831 } 5832 5833 // MSVC destroys objects passed by value in the callee. Therefore a 5834 // function definition which takes such a parameter must be able to call the 5835 // object's destructor. 5836 if (getLangOpts().CPlusPlus && 5837 Context.getTargetInfo().getCXXABI().isArgumentDestroyedByCallee()) { 5838 if (const RecordType *RT = Param->getType()->getAs<RecordType>()) 5839 FinalizeVarWithDestructor(Param, RT); 5840 } 5841 } 5842 5843 return HasInvalidParm; 5844} 5845 5846/// CheckCastAlign - Implements -Wcast-align, which warns when a 5847/// pointer cast increases the alignment requirements. 5848void Sema::CheckCastAlign(Expr *Op, QualType T, SourceRange TRange) { 5849 // This is actually a lot of work to potentially be doing on every 5850 // cast; don't do it if we're ignoring -Wcast_align (as is the default). 5851 if (getDiagnostics().getDiagnosticLevel(diag::warn_cast_align, 5852 TRange.getBegin()) 5853 == DiagnosticsEngine::Ignored) 5854 return; 5855 5856 // Ignore dependent types. 5857 if (T->isDependentType() || Op->getType()->isDependentType()) 5858 return; 5859 5860 // Require that the destination be a pointer type. 5861 const PointerType *DestPtr = T->getAs<PointerType>(); 5862 if (!DestPtr) return; 5863 5864 // If the destination has alignment 1, we're done. 5865 QualType DestPointee = DestPtr->getPointeeType(); 5866 if (DestPointee->isIncompleteType()) return; 5867 CharUnits DestAlign = Context.getTypeAlignInChars(DestPointee); 5868 if (DestAlign.isOne()) return; 5869 5870 // Require that the source be a pointer type. 5871 const PointerType *SrcPtr = Op->getType()->getAs<PointerType>(); 5872 if (!SrcPtr) return; 5873 QualType SrcPointee = SrcPtr->getPointeeType(); 5874 5875 // Whitelist casts from cv void*. We already implicitly 5876 // whitelisted casts to cv void*, since they have alignment 1. 5877 // Also whitelist casts involving incomplete types, which implicitly 5878 // includes 'void'. 5879 if (SrcPointee->isIncompleteType()) return; 5880 5881 CharUnits SrcAlign = Context.getTypeAlignInChars(SrcPointee); 5882 if (SrcAlign >= DestAlign) return; 5883 5884 Diag(TRange.getBegin(), diag::warn_cast_align) 5885 << Op->getType() << T 5886 << static_cast<unsigned>(SrcAlign.getQuantity()) 5887 << static_cast<unsigned>(DestAlign.getQuantity()) 5888 << TRange << Op->getSourceRange(); 5889} 5890 5891static const Type* getElementType(const Expr *BaseExpr) { 5892 const Type* EltType = BaseExpr->getType().getTypePtr(); 5893 if (EltType->isAnyPointerType()) 5894 return EltType->getPointeeType().getTypePtr(); 5895 else if (EltType->isArrayType()) 5896 return EltType->getBaseElementTypeUnsafe(); 5897 return EltType; 5898} 5899 5900/// \brief Check whether this array fits the idiom of a size-one tail padded 5901/// array member of a struct. 5902/// 5903/// We avoid emitting out-of-bounds access warnings for such arrays as they are 5904/// commonly used to emulate flexible arrays in C89 code. 5905static bool IsTailPaddedMemberArray(Sema &S, llvm::APInt Size, 5906 const NamedDecl *ND) { 5907 if (Size != 1 || !ND) return false; 5908 5909 const FieldDecl *FD = dyn_cast<FieldDecl>(ND); 5910 if (!FD) return false; 5911 5912 // Don't consider sizes resulting from macro expansions or template argument 5913 // substitution to form C89 tail-padded arrays. 5914 5915 TypeSourceInfo *TInfo = FD->getTypeSourceInfo(); 5916 while (TInfo) { 5917 TypeLoc TL = TInfo->getTypeLoc(); 5918 // Look through typedefs. 5919 if (TypedefTypeLoc TTL = TL.getAs<TypedefTypeLoc>()) { 5920 const TypedefNameDecl *TDL = TTL.getTypedefNameDecl(); 5921 TInfo = TDL->getTypeSourceInfo(); 5922 continue; 5923 } 5924 if (ConstantArrayTypeLoc CTL = TL.getAs<ConstantArrayTypeLoc>()) { 5925 const Expr *SizeExpr = dyn_cast<IntegerLiteral>(CTL.getSizeExpr()); 5926 if (!SizeExpr || SizeExpr->getExprLoc().isMacroID()) 5927 return false; 5928 } 5929 break; 5930 } 5931 5932 const RecordDecl *RD = dyn_cast<RecordDecl>(FD->getDeclContext()); 5933 if (!RD) return false; 5934 if (RD->isUnion()) return false; 5935 if (const CXXRecordDecl *CRD = dyn_cast<CXXRecordDecl>(RD)) { 5936 if (!CRD->isStandardLayout()) return false; 5937 } 5938 5939 // See if this is the last field decl in the record. 5940 const Decl *D = FD; 5941 while ((D = D->getNextDeclInContext())) 5942 if (isa<FieldDecl>(D)) 5943 return false; 5944 return true; 5945} 5946 5947void Sema::CheckArrayAccess(const Expr *BaseExpr, const Expr *IndexExpr, 5948 const ArraySubscriptExpr *ASE, 5949 bool AllowOnePastEnd, bool IndexNegated) { 5950 IndexExpr = IndexExpr->IgnoreParenImpCasts(); 5951 if (IndexExpr->isValueDependent()) 5952 return; 5953 5954 const Type *EffectiveType = getElementType(BaseExpr); 5955 BaseExpr = BaseExpr->IgnoreParenCasts(); 5956 const ConstantArrayType *ArrayTy = 5957 Context.getAsConstantArrayType(BaseExpr->getType()); 5958 if (!ArrayTy) 5959 return; 5960 5961 llvm::APSInt index; 5962 if (!IndexExpr->EvaluateAsInt(index, Context)) 5963 return; 5964 if (IndexNegated) 5965 index = -index; 5966 5967 const NamedDecl *ND = NULL; 5968 if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(BaseExpr)) 5969 ND = dyn_cast<NamedDecl>(DRE->getDecl()); 5970 if (const MemberExpr *ME = dyn_cast<MemberExpr>(BaseExpr)) 5971 ND = dyn_cast<NamedDecl>(ME->getMemberDecl()); 5972 5973 if (index.isUnsigned() || !index.isNegative()) { 5974 llvm::APInt size = ArrayTy->getSize(); 5975 if (!size.isStrictlyPositive()) 5976 return; 5977 5978 const Type* BaseType = getElementType(BaseExpr); 5979 if (BaseType != EffectiveType) { 5980 // Make sure we're comparing apples to apples when comparing index to size 5981 uint64_t ptrarith_typesize = Context.getTypeSize(EffectiveType); 5982 uint64_t array_typesize = Context.getTypeSize(BaseType); 5983 // Handle ptrarith_typesize being zero, such as when casting to void* 5984 if (!ptrarith_typesize) ptrarith_typesize = 1; 5985 if (ptrarith_typesize != array_typesize) { 5986 // There's a cast to a different size type involved 5987 uint64_t ratio = array_typesize / ptrarith_typesize; 5988 // TODO: Be smarter about handling cases where array_typesize is not a 5989 // multiple of ptrarith_typesize 5990 if (ptrarith_typesize * ratio == array_typesize) 5991 size *= llvm::APInt(size.getBitWidth(), ratio); 5992 } 5993 } 5994 5995 if (size.getBitWidth() > index.getBitWidth()) 5996 index = index.zext(size.getBitWidth()); 5997 else if (size.getBitWidth() < index.getBitWidth()) 5998 size = size.zext(index.getBitWidth()); 5999 6000 // For array subscripting the index must be less than size, but for pointer 6001 // arithmetic also allow the index (offset) to be equal to size since 6002 // computing the next address after the end of the array is legal and 6003 // commonly done e.g. in C++ iterators and range-based for loops. 6004 if (AllowOnePastEnd ? index.ule(size) : index.ult(size)) 6005 return; 6006 6007 // Also don't warn for arrays of size 1 which are members of some 6008 // structure. These are often used to approximate flexible arrays in C89 6009 // code. 6010 if (IsTailPaddedMemberArray(*this, size, ND)) 6011 return; 6012 6013 // Suppress the warning if the subscript expression (as identified by the 6014 // ']' location) and the index expression are both from macro expansions 6015 // within a system header. 6016 if (ASE) { 6017 SourceLocation RBracketLoc = SourceMgr.getSpellingLoc( 6018 ASE->getRBracketLoc()); 6019 if (SourceMgr.isInSystemHeader(RBracketLoc)) { 6020 SourceLocation IndexLoc = SourceMgr.getSpellingLoc( 6021 IndexExpr->getLocStart()); 6022 if (SourceMgr.isFromSameFile(RBracketLoc, IndexLoc)) 6023 return; 6024 } 6025 } 6026 6027 unsigned DiagID = diag::warn_ptr_arith_exceeds_bounds; 6028 if (ASE) 6029 DiagID = diag::warn_array_index_exceeds_bounds; 6030 6031 DiagRuntimeBehavior(BaseExpr->getLocStart(), BaseExpr, 6032 PDiag(DiagID) << index.toString(10, true) 6033 << size.toString(10, true) 6034 << (unsigned)size.getLimitedValue(~0U) 6035 << IndexExpr->getSourceRange()); 6036 } else { 6037 unsigned DiagID = diag::warn_array_index_precedes_bounds; 6038 if (!ASE) { 6039 DiagID = diag::warn_ptr_arith_precedes_bounds; 6040 if (index.isNegative()) index = -index; 6041 } 6042 6043 DiagRuntimeBehavior(BaseExpr->getLocStart(), BaseExpr, 6044 PDiag(DiagID) << index.toString(10, true) 6045 << IndexExpr->getSourceRange()); 6046 } 6047 6048 if (!ND) { 6049 // Try harder to find a NamedDecl to point at in the note. 6050 while (const ArraySubscriptExpr *ASE = 6051 dyn_cast<ArraySubscriptExpr>(BaseExpr)) 6052 BaseExpr = ASE->getBase()->IgnoreParenCasts(); 6053 if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(BaseExpr)) 6054 ND = dyn_cast<NamedDecl>(DRE->getDecl()); 6055 if (const MemberExpr *ME = dyn_cast<MemberExpr>(BaseExpr)) 6056 ND = dyn_cast<NamedDecl>(ME->getMemberDecl()); 6057 } 6058 6059 if (ND) 6060 DiagRuntimeBehavior(ND->getLocStart(), BaseExpr, 6061 PDiag(diag::note_array_index_out_of_bounds) 6062 << ND->getDeclName()); 6063} 6064 6065void Sema::CheckArrayAccess(const Expr *expr) { 6066 int AllowOnePastEnd = 0; 6067 while (expr) { 6068 expr = expr->IgnoreParenImpCasts(); 6069 switch (expr->getStmtClass()) { 6070 case Stmt::ArraySubscriptExprClass: { 6071 const ArraySubscriptExpr *ASE = cast<ArraySubscriptExpr>(expr); 6072 CheckArrayAccess(ASE->getBase(), ASE->getIdx(), ASE, 6073 AllowOnePastEnd > 0); 6074 return; 6075 } 6076 case Stmt::UnaryOperatorClass: { 6077 // Only unwrap the * and & unary operators 6078 const UnaryOperator *UO = cast<UnaryOperator>(expr); 6079 expr = UO->getSubExpr(); 6080 switch (UO->getOpcode()) { 6081 case UO_AddrOf: 6082 AllowOnePastEnd++; 6083 break; 6084 case UO_Deref: 6085 AllowOnePastEnd--; 6086 break; 6087 default: 6088 return; 6089 } 6090 break; 6091 } 6092 case Stmt::ConditionalOperatorClass: { 6093 const ConditionalOperator *cond = cast<ConditionalOperator>(expr); 6094 if (const Expr *lhs = cond->getLHS()) 6095 CheckArrayAccess(lhs); 6096 if (const Expr *rhs = cond->getRHS()) 6097 CheckArrayAccess(rhs); 6098 return; 6099 } 6100 default: 6101 return; 6102 } 6103 } 6104} 6105 6106//===--- CHECK: Objective-C retain cycles ----------------------------------// 6107 6108namespace { 6109 struct RetainCycleOwner { 6110 RetainCycleOwner() : Variable(0), Indirect(false) {} 6111 VarDecl *Variable; 6112 SourceRange Range; 6113 SourceLocation Loc; 6114 bool Indirect; 6115 6116 void setLocsFrom(Expr *e) { 6117 Loc = e->getExprLoc(); 6118 Range = e->getSourceRange(); 6119 } 6120 }; 6121} 6122 6123/// Consider whether capturing the given variable can possibly lead to 6124/// a retain cycle. 6125static bool considerVariable(VarDecl *var, Expr *ref, RetainCycleOwner &owner) { 6126 // In ARC, it's captured strongly iff the variable has __strong 6127 // lifetime. In MRR, it's captured strongly if the variable is 6128 // __block and has an appropriate type. 6129 if (var->getType().getObjCLifetime() != Qualifiers::OCL_Strong) 6130 return false; 6131 6132 owner.Variable = var; 6133 if (ref) 6134 owner.setLocsFrom(ref); 6135 return true; 6136} 6137 6138static bool findRetainCycleOwner(Sema &S, Expr *e, RetainCycleOwner &owner) { 6139 while (true) { 6140 e = e->IgnoreParens(); 6141 if (CastExpr *cast = dyn_cast<CastExpr>(e)) { 6142 switch (cast->getCastKind()) { 6143 case CK_BitCast: 6144 case CK_LValueBitCast: 6145 case CK_LValueToRValue: 6146 case CK_ARCReclaimReturnedObject: 6147 e = cast->getSubExpr(); 6148 continue; 6149 6150 default: 6151 return false; 6152 } 6153 } 6154 6155 if (ObjCIvarRefExpr *ref = dyn_cast<ObjCIvarRefExpr>(e)) { 6156 ObjCIvarDecl *ivar = ref->getDecl(); 6157 if (ivar->getType().getObjCLifetime() != Qualifiers::OCL_Strong) 6158 return false; 6159 6160 // Try to find a retain cycle in the base. 6161 if (!findRetainCycleOwner(S, ref->getBase(), owner)) 6162 return false; 6163 6164 if (ref->isFreeIvar()) owner.setLocsFrom(ref); 6165 owner.Indirect = true; 6166 return true; 6167 } 6168 6169 if (DeclRefExpr *ref = dyn_cast<DeclRefExpr>(e)) { 6170 VarDecl *var = dyn_cast<VarDecl>(ref->getDecl()); 6171 if (!var) return false; 6172 return considerVariable(var, ref, owner); 6173 } 6174 6175 if (MemberExpr *member = dyn_cast<MemberExpr>(e)) { 6176 if (member->isArrow()) return false; 6177 6178 // Don't count this as an indirect ownership. 6179 e = member->getBase(); 6180 continue; 6181 } 6182 6183 if (PseudoObjectExpr *pseudo = dyn_cast<PseudoObjectExpr>(e)) { 6184 // Only pay attention to pseudo-objects on property references. 6185 ObjCPropertyRefExpr *pre 6186 = dyn_cast<ObjCPropertyRefExpr>(pseudo->getSyntacticForm() 6187 ->IgnoreParens()); 6188 if (!pre) return false; 6189 if (pre->isImplicitProperty()) return false; 6190 ObjCPropertyDecl *property = pre->getExplicitProperty(); 6191 if (!property->isRetaining() && 6192 !(property->getPropertyIvarDecl() && 6193 property->getPropertyIvarDecl()->getType() 6194 .getObjCLifetime() == Qualifiers::OCL_Strong)) 6195 return false; 6196 6197 owner.Indirect = true; 6198 if (pre->isSuperReceiver()) { 6199 owner.Variable = S.getCurMethodDecl()->getSelfDecl(); 6200 if (!owner.Variable) 6201 return false; 6202 owner.Loc = pre->getLocation(); 6203 owner.Range = pre->getSourceRange(); 6204 return true; 6205 } 6206 e = const_cast<Expr*>(cast<OpaqueValueExpr>(pre->getBase()) 6207 ->getSourceExpr()); 6208 continue; 6209 } 6210 6211 // Array ivars? 6212 6213 return false; 6214 } 6215} 6216 6217namespace { 6218 struct FindCaptureVisitor : EvaluatedExprVisitor<FindCaptureVisitor> { 6219 FindCaptureVisitor(ASTContext &Context, VarDecl *variable) 6220 : EvaluatedExprVisitor<FindCaptureVisitor>(Context), 6221 Variable(variable), Capturer(0) {} 6222 6223 VarDecl *Variable; 6224 Expr *Capturer; 6225 6226 void VisitDeclRefExpr(DeclRefExpr *ref) { 6227 if (ref->getDecl() == Variable && !Capturer) 6228 Capturer = ref; 6229 } 6230 6231 void VisitObjCIvarRefExpr(ObjCIvarRefExpr *ref) { 6232 if (Capturer) return; 6233 Visit(ref->getBase()); 6234 if (Capturer && ref->isFreeIvar()) 6235 Capturer = ref; 6236 } 6237 6238 void VisitBlockExpr(BlockExpr *block) { 6239 // Look inside nested blocks 6240 if (block->getBlockDecl()->capturesVariable(Variable)) 6241 Visit(block->getBlockDecl()->getBody()); 6242 } 6243 6244 void VisitOpaqueValueExpr(OpaqueValueExpr *OVE) { 6245 if (Capturer) return; 6246 if (OVE->getSourceExpr()) 6247 Visit(OVE->getSourceExpr()); 6248 } 6249 }; 6250} 6251 6252/// Check whether the given argument is a block which captures a 6253/// variable. 6254static Expr *findCapturingExpr(Sema &S, Expr *e, RetainCycleOwner &owner) { 6255 assert(owner.Variable && owner.Loc.isValid()); 6256 6257 e = e->IgnoreParenCasts(); 6258 6259 // Look through [^{...} copy] and Block_copy(^{...}). 6260 if (ObjCMessageExpr *ME = dyn_cast<ObjCMessageExpr>(e)) { 6261 Selector Cmd = ME->getSelector(); 6262 if (Cmd.isUnarySelector() && Cmd.getNameForSlot(0) == "copy") { 6263 e = ME->getInstanceReceiver(); 6264 if (!e) 6265 return 0; 6266 e = e->IgnoreParenCasts(); 6267 } 6268 } else if (CallExpr *CE = dyn_cast<CallExpr>(e)) { 6269 if (CE->getNumArgs() == 1) { 6270 FunctionDecl *Fn = dyn_cast_or_null<FunctionDecl>(CE->getCalleeDecl()); 6271 if (Fn) { 6272 const IdentifierInfo *FnI = Fn->getIdentifier(); 6273 if (FnI && FnI->isStr("_Block_copy")) { 6274 e = CE->getArg(0)->IgnoreParenCasts(); 6275 } 6276 } 6277 } 6278 } 6279 6280 BlockExpr *block = dyn_cast<BlockExpr>(e); 6281 if (!block || !block->getBlockDecl()->capturesVariable(owner.Variable)) 6282 return 0; 6283 6284 FindCaptureVisitor visitor(S.Context, owner.Variable); 6285 visitor.Visit(block->getBlockDecl()->getBody()); 6286 return visitor.Capturer; 6287} 6288 6289static void diagnoseRetainCycle(Sema &S, Expr *capturer, 6290 RetainCycleOwner &owner) { 6291 assert(capturer); 6292 assert(owner.Variable && owner.Loc.isValid()); 6293 6294 S.Diag(capturer->getExprLoc(), diag::warn_arc_retain_cycle) 6295 << owner.Variable << capturer->getSourceRange(); 6296 S.Diag(owner.Loc, diag::note_arc_retain_cycle_owner) 6297 << owner.Indirect << owner.Range; 6298} 6299 6300/// Check for a keyword selector that starts with the word 'add' or 6301/// 'set'. 6302static bool isSetterLikeSelector(Selector sel) { 6303 if (sel.isUnarySelector()) return false; 6304 6305 StringRef str = sel.getNameForSlot(0); 6306 while (!str.empty() && str.front() == '_') str = str.substr(1); 6307 if (str.startswith("set")) 6308 str = str.substr(3); 6309 else if (str.startswith("add")) { 6310 // Specially whitelist 'addOperationWithBlock:'. 6311 if (sel.getNumArgs() == 1 && str.startswith("addOperationWithBlock")) 6312 return false; 6313 str = str.substr(3); 6314 } 6315 else 6316 return false; 6317 6318 if (str.empty()) return true; 6319 return !isLowercase(str.front()); 6320} 6321 6322/// Check a message send to see if it's likely to cause a retain cycle. 6323void Sema::checkRetainCycles(ObjCMessageExpr *msg) { 6324 // Only check instance methods whose selector looks like a setter. 6325 if (!msg->isInstanceMessage() || !isSetterLikeSelector(msg->getSelector())) 6326 return; 6327 6328 // Try to find a variable that the receiver is strongly owned by. 6329 RetainCycleOwner owner; 6330 if (msg->getReceiverKind() == ObjCMessageExpr::Instance) { 6331 if (!findRetainCycleOwner(*this, msg->getInstanceReceiver(), owner)) 6332 return; 6333 } else { 6334 assert(msg->getReceiverKind() == ObjCMessageExpr::SuperInstance); 6335 owner.Variable = getCurMethodDecl()->getSelfDecl(); 6336 owner.Loc = msg->getSuperLoc(); 6337 owner.Range = msg->getSuperLoc(); 6338 } 6339 6340 // Check whether the receiver is captured by any of the arguments. 6341 for (unsigned i = 0, e = msg->getNumArgs(); i != e; ++i) 6342 if (Expr *capturer = findCapturingExpr(*this, msg->getArg(i), owner)) 6343 return diagnoseRetainCycle(*this, capturer, owner); 6344} 6345 6346/// Check a property assign to see if it's likely to cause a retain cycle. 6347void Sema::checkRetainCycles(Expr *receiver, Expr *argument) { 6348 RetainCycleOwner owner; 6349 if (!findRetainCycleOwner(*this, receiver, owner)) 6350 return; 6351 6352 if (Expr *capturer = findCapturingExpr(*this, argument, owner)) 6353 diagnoseRetainCycle(*this, capturer, owner); 6354} 6355 6356void Sema::checkRetainCycles(VarDecl *Var, Expr *Init) { 6357 RetainCycleOwner Owner; 6358 if (!considerVariable(Var, /*DeclRefExpr=*/0, Owner)) 6359 return; 6360 6361 // Because we don't have an expression for the variable, we have to set the 6362 // location explicitly here. 6363 Owner.Loc = Var->getLocation(); 6364 Owner.Range = Var->getSourceRange(); 6365 6366 if (Expr *Capturer = findCapturingExpr(*this, Init, Owner)) 6367 diagnoseRetainCycle(*this, Capturer, Owner); 6368} 6369 6370static bool checkUnsafeAssignLiteral(Sema &S, SourceLocation Loc, 6371 Expr *RHS, bool isProperty) { 6372 // Check if RHS is an Objective-C object literal, which also can get 6373 // immediately zapped in a weak reference. Note that we explicitly 6374 // allow ObjCStringLiterals, since those are designed to never really die. 6375 RHS = RHS->IgnoreParenImpCasts(); 6376 6377 // This enum needs to match with the 'select' in 6378 // warn_objc_arc_literal_assign (off-by-1). 6379 Sema::ObjCLiteralKind Kind = S.CheckLiteralKind(RHS); 6380 if (Kind == Sema::LK_String || Kind == Sema::LK_None) 6381 return false; 6382 6383 S.Diag(Loc, diag::warn_arc_literal_assign) 6384 << (unsigned) Kind 6385 << (isProperty ? 0 : 1) 6386 << RHS->getSourceRange(); 6387 6388 return true; 6389} 6390 6391static bool checkUnsafeAssignObject(Sema &S, SourceLocation Loc, 6392 Qualifiers::ObjCLifetime LT, 6393 Expr *RHS, bool isProperty) { 6394 // Strip off any implicit cast added to get to the one ARC-specific. 6395 while (ImplicitCastExpr *cast = dyn_cast<ImplicitCastExpr>(RHS)) { 6396 if (cast->getCastKind() == CK_ARCConsumeObject) { 6397 S.Diag(Loc, diag::warn_arc_retained_assign) 6398 << (LT == Qualifiers::OCL_ExplicitNone) 6399 << (isProperty ? 0 : 1) 6400 << RHS->getSourceRange(); 6401 return true; 6402 } 6403 RHS = cast->getSubExpr(); 6404 } 6405 6406 if (LT == Qualifiers::OCL_Weak && 6407 checkUnsafeAssignLiteral(S, Loc, RHS, isProperty)) 6408 return true; 6409 6410 return false; 6411} 6412 6413bool Sema::checkUnsafeAssigns(SourceLocation Loc, 6414 QualType LHS, Expr *RHS) { 6415 Qualifiers::ObjCLifetime LT = LHS.getObjCLifetime(); 6416 6417 if (LT != Qualifiers::OCL_Weak && LT != Qualifiers::OCL_ExplicitNone) 6418 return false; 6419 6420 if (checkUnsafeAssignObject(*this, Loc, LT, RHS, false)) 6421 return true; 6422 6423 return false; 6424} 6425 6426void Sema::checkUnsafeExprAssigns(SourceLocation Loc, 6427 Expr *LHS, Expr *RHS) { 6428 QualType LHSType; 6429 // PropertyRef on LHS type need be directly obtained from 6430 // its declaration as it has a PsuedoType. 6431 ObjCPropertyRefExpr *PRE 6432 = dyn_cast<ObjCPropertyRefExpr>(LHS->IgnoreParens()); 6433 if (PRE && !PRE->isImplicitProperty()) { 6434 const ObjCPropertyDecl *PD = PRE->getExplicitProperty(); 6435 if (PD) 6436 LHSType = PD->getType(); 6437 } 6438 6439 if (LHSType.isNull()) 6440 LHSType = LHS->getType(); 6441 6442 Qualifiers::ObjCLifetime LT = LHSType.getObjCLifetime(); 6443 6444 if (LT == Qualifiers::OCL_Weak) { 6445 DiagnosticsEngine::Level Level = 6446 Diags.getDiagnosticLevel(diag::warn_arc_repeated_use_of_weak, Loc); 6447 if (Level != DiagnosticsEngine::Ignored) 6448 getCurFunction()->markSafeWeakUse(LHS); 6449 } 6450 6451 if (checkUnsafeAssigns(Loc, LHSType, RHS)) 6452 return; 6453 6454 // FIXME. Check for other life times. 6455 if (LT != Qualifiers::OCL_None) 6456 return; 6457 6458 if (PRE) { 6459 if (PRE->isImplicitProperty()) 6460 return; 6461 const ObjCPropertyDecl *PD = PRE->getExplicitProperty(); 6462 if (!PD) 6463 return; 6464 6465 unsigned Attributes = PD->getPropertyAttributes(); 6466 if (Attributes & ObjCPropertyDecl::OBJC_PR_assign) { 6467 // when 'assign' attribute was not explicitly specified 6468 // by user, ignore it and rely on property type itself 6469 // for lifetime info. 6470 unsigned AsWrittenAttr = PD->getPropertyAttributesAsWritten(); 6471 if (!(AsWrittenAttr & ObjCPropertyDecl::OBJC_PR_assign) && 6472 LHSType->isObjCRetainableType()) 6473 return; 6474 6475 while (ImplicitCastExpr *cast = dyn_cast<ImplicitCastExpr>(RHS)) { 6476 if (cast->getCastKind() == CK_ARCConsumeObject) { 6477 Diag(Loc, diag::warn_arc_retained_property_assign) 6478 << RHS->getSourceRange(); 6479 return; 6480 } 6481 RHS = cast->getSubExpr(); 6482 } 6483 } 6484 else if (Attributes & ObjCPropertyDecl::OBJC_PR_weak) { 6485 if (checkUnsafeAssignObject(*this, Loc, Qualifiers::OCL_Weak, RHS, true)) 6486 return; 6487 } 6488 } 6489} 6490 6491//===--- CHECK: Empty statement body (-Wempty-body) ---------------------===// 6492 6493namespace { 6494bool ShouldDiagnoseEmptyStmtBody(const SourceManager &SourceMgr, 6495 SourceLocation StmtLoc, 6496 const NullStmt *Body) { 6497 // Do not warn if the body is a macro that expands to nothing, e.g: 6498 // 6499 // #define CALL(x) 6500 // if (condition) 6501 // CALL(0); 6502 // 6503 if (Body->hasLeadingEmptyMacro()) 6504 return false; 6505 6506 // Get line numbers of statement and body. 6507 bool StmtLineInvalid; 6508 unsigned StmtLine = SourceMgr.getSpellingLineNumber(StmtLoc, 6509 &StmtLineInvalid); 6510 if (StmtLineInvalid) 6511 return false; 6512 6513 bool BodyLineInvalid; 6514 unsigned BodyLine = SourceMgr.getSpellingLineNumber(Body->getSemiLoc(), 6515 &BodyLineInvalid); 6516 if (BodyLineInvalid) 6517 return false; 6518 6519 // Warn if null statement and body are on the same line. 6520 if (StmtLine != BodyLine) 6521 return false; 6522 6523 return true; 6524} 6525} // Unnamed namespace 6526 6527void Sema::DiagnoseEmptyStmtBody(SourceLocation StmtLoc, 6528 const Stmt *Body, 6529 unsigned DiagID) { 6530 // Since this is a syntactic check, don't emit diagnostic for template 6531 // instantiations, this just adds noise. 6532 if (CurrentInstantiationScope) 6533 return; 6534 6535 // The body should be a null statement. 6536 const NullStmt *NBody = dyn_cast<NullStmt>(Body); 6537 if (!NBody) 6538 return; 6539 6540 // Do the usual checks. 6541 if (!ShouldDiagnoseEmptyStmtBody(SourceMgr, StmtLoc, NBody)) 6542 return; 6543 6544 Diag(NBody->getSemiLoc(), DiagID); 6545 Diag(NBody->getSemiLoc(), diag::note_empty_body_on_separate_line); 6546} 6547 6548void Sema::DiagnoseEmptyLoopBody(const Stmt *S, 6549 const Stmt *PossibleBody) { 6550 assert(!CurrentInstantiationScope); // Ensured by caller 6551 6552 SourceLocation StmtLoc; 6553 const Stmt *Body; 6554 unsigned DiagID; 6555 if (const ForStmt *FS = dyn_cast<ForStmt>(S)) { 6556 StmtLoc = FS->getRParenLoc(); 6557 Body = FS->getBody(); 6558 DiagID = diag::warn_empty_for_body; 6559 } else if (const WhileStmt *WS = dyn_cast<WhileStmt>(S)) { 6560 StmtLoc = WS->getCond()->getSourceRange().getEnd(); 6561 Body = WS->getBody(); 6562 DiagID = diag::warn_empty_while_body; 6563 } else 6564 return; // Neither `for' nor `while'. 6565 6566 // The body should be a null statement. 6567 const NullStmt *NBody = dyn_cast<NullStmt>(Body); 6568 if (!NBody) 6569 return; 6570 6571 // Skip expensive checks if diagnostic is disabled. 6572 if (Diags.getDiagnosticLevel(DiagID, NBody->getSemiLoc()) == 6573 DiagnosticsEngine::Ignored) 6574 return; 6575 6576 // Do the usual checks. 6577 if (!ShouldDiagnoseEmptyStmtBody(SourceMgr, StmtLoc, NBody)) 6578 return; 6579 6580 // `for(...);' and `while(...);' are popular idioms, so in order to keep 6581 // noise level low, emit diagnostics only if for/while is followed by a 6582 // CompoundStmt, e.g.: 6583 // for (int i = 0; i < n; i++); 6584 // { 6585 // a(i); 6586 // } 6587 // or if for/while is followed by a statement with more indentation 6588 // than for/while itself: 6589 // for (int i = 0; i < n; i++); 6590 // a(i); 6591 bool ProbableTypo = isa<CompoundStmt>(PossibleBody); 6592 if (!ProbableTypo) { 6593 bool BodyColInvalid; 6594 unsigned BodyCol = SourceMgr.getPresumedColumnNumber( 6595 PossibleBody->getLocStart(), 6596 &BodyColInvalid); 6597 if (BodyColInvalid) 6598 return; 6599 6600 bool StmtColInvalid; 6601 unsigned StmtCol = SourceMgr.getPresumedColumnNumber( 6602 S->getLocStart(), 6603 &StmtColInvalid); 6604 if (StmtColInvalid) 6605 return; 6606 6607 if (BodyCol > StmtCol) 6608 ProbableTypo = true; 6609 } 6610 6611 if (ProbableTypo) { 6612 Diag(NBody->getSemiLoc(), DiagID); 6613 Diag(NBody->getSemiLoc(), diag::note_empty_body_on_separate_line); 6614 } 6615} 6616 6617//===--- Layout compatibility ----------------------------------------------// 6618 6619namespace { 6620 6621bool isLayoutCompatible(ASTContext &C, QualType T1, QualType T2); 6622 6623/// \brief Check if two enumeration types are layout-compatible. 6624bool isLayoutCompatible(ASTContext &C, EnumDecl *ED1, EnumDecl *ED2) { 6625 // C++11 [dcl.enum] p8: 6626 // Two enumeration types are layout-compatible if they have the same 6627 // underlying type. 6628 return ED1->isComplete() && ED2->isComplete() && 6629 C.hasSameType(ED1->getIntegerType(), ED2->getIntegerType()); 6630} 6631 6632/// \brief Check if two fields are layout-compatible. 6633bool isLayoutCompatible(ASTContext &C, FieldDecl *Field1, FieldDecl *Field2) { 6634 if (!isLayoutCompatible(C, Field1->getType(), Field2->getType())) 6635 return false; 6636 6637 if (Field1->isBitField() != Field2->isBitField()) 6638 return false; 6639 6640 if (Field1->isBitField()) { 6641 // Make sure that the bit-fields are the same length. 6642 unsigned Bits1 = Field1->getBitWidthValue(C); 6643 unsigned Bits2 = Field2->getBitWidthValue(C); 6644 6645 if (Bits1 != Bits2) 6646 return false; 6647 } 6648 6649 return true; 6650} 6651 6652/// \brief Check if two standard-layout structs are layout-compatible. 6653/// (C++11 [class.mem] p17) 6654bool isLayoutCompatibleStruct(ASTContext &C, 6655 RecordDecl *RD1, 6656 RecordDecl *RD2) { 6657 // If both records are C++ classes, check that base classes match. 6658 if (const CXXRecordDecl *D1CXX = dyn_cast<CXXRecordDecl>(RD1)) { 6659 // If one of records is a CXXRecordDecl we are in C++ mode, 6660 // thus the other one is a CXXRecordDecl, too. 6661 const CXXRecordDecl *D2CXX = cast<CXXRecordDecl>(RD2); 6662 // Check number of base classes. 6663 if (D1CXX->getNumBases() != D2CXX->getNumBases()) 6664 return false; 6665 6666 // Check the base classes. 6667 for (CXXRecordDecl::base_class_const_iterator 6668 Base1 = D1CXX->bases_begin(), 6669 BaseEnd1 = D1CXX->bases_end(), 6670 Base2 = D2CXX->bases_begin(); 6671 Base1 != BaseEnd1; 6672 ++Base1, ++Base2) { 6673 if (!isLayoutCompatible(C, Base1->getType(), Base2->getType())) 6674 return false; 6675 } 6676 } else if (const CXXRecordDecl *D2CXX = dyn_cast<CXXRecordDecl>(RD2)) { 6677 // If only RD2 is a C++ class, it should have zero base classes. 6678 if (D2CXX->getNumBases() > 0) 6679 return false; 6680 } 6681 6682 // Check the fields. 6683 RecordDecl::field_iterator Field2 = RD2->field_begin(), 6684 Field2End = RD2->field_end(), 6685 Field1 = RD1->field_begin(), 6686 Field1End = RD1->field_end(); 6687 for ( ; Field1 != Field1End && Field2 != Field2End; ++Field1, ++Field2) { 6688 if (!isLayoutCompatible(C, *Field1, *Field2)) 6689 return false; 6690 } 6691 if (Field1 != Field1End || Field2 != Field2End) 6692 return false; 6693 6694 return true; 6695} 6696 6697/// \brief Check if two standard-layout unions are layout-compatible. 6698/// (C++11 [class.mem] p18) 6699bool isLayoutCompatibleUnion(ASTContext &C, 6700 RecordDecl *RD1, 6701 RecordDecl *RD2) { 6702 llvm::SmallPtrSet<FieldDecl *, 8> UnmatchedFields; 6703 for (RecordDecl::field_iterator Field2 = RD2->field_begin(), 6704 Field2End = RD2->field_end(); 6705 Field2 != Field2End; ++Field2) { 6706 UnmatchedFields.insert(*Field2); 6707 } 6708 6709 for (RecordDecl::field_iterator Field1 = RD1->field_begin(), 6710 Field1End = RD1->field_end(); 6711 Field1 != Field1End; ++Field1) { 6712 llvm::SmallPtrSet<FieldDecl *, 8>::iterator 6713 I = UnmatchedFields.begin(), 6714 E = UnmatchedFields.end(); 6715 6716 for ( ; I != E; ++I) { 6717 if (isLayoutCompatible(C, *Field1, *I)) { 6718 bool Result = UnmatchedFields.erase(*I); 6719 (void) Result; 6720 assert(Result); 6721 break; 6722 } 6723 } 6724 if (I == E) 6725 return false; 6726 } 6727 6728 return UnmatchedFields.empty(); 6729} 6730 6731bool isLayoutCompatible(ASTContext &C, RecordDecl *RD1, RecordDecl *RD2) { 6732 if (RD1->isUnion() != RD2->isUnion()) 6733 return false; 6734 6735 if (RD1->isUnion()) 6736 return isLayoutCompatibleUnion(C, RD1, RD2); 6737 else 6738 return isLayoutCompatibleStruct(C, RD1, RD2); 6739} 6740 6741/// \brief Check if two types are layout-compatible in C++11 sense. 6742bool isLayoutCompatible(ASTContext &C, QualType T1, QualType T2) { 6743 if (T1.isNull() || T2.isNull()) 6744 return false; 6745 6746 // C++11 [basic.types] p11: 6747 // If two types T1 and T2 are the same type, then T1 and T2 are 6748 // layout-compatible types. 6749 if (C.hasSameType(T1, T2)) 6750 return true; 6751 6752 T1 = T1.getCanonicalType().getUnqualifiedType(); 6753 T2 = T2.getCanonicalType().getUnqualifiedType(); 6754 6755 const Type::TypeClass TC1 = T1->getTypeClass(); 6756 const Type::TypeClass TC2 = T2->getTypeClass(); 6757 6758 if (TC1 != TC2) 6759 return false; 6760 6761 if (TC1 == Type::Enum) { 6762 return isLayoutCompatible(C, 6763 cast<EnumType>(T1)->getDecl(), 6764 cast<EnumType>(T2)->getDecl()); 6765 } else if (TC1 == Type::Record) { 6766 if (!T1->isStandardLayoutType() || !T2->isStandardLayoutType()) 6767 return false; 6768 6769 return isLayoutCompatible(C, 6770 cast<RecordType>(T1)->getDecl(), 6771 cast<RecordType>(T2)->getDecl()); 6772 } 6773 6774 return false; 6775} 6776} 6777 6778//===--- CHECK: pointer_with_type_tag attribute: datatypes should match ----// 6779 6780namespace { 6781/// \brief Given a type tag expression find the type tag itself. 6782/// 6783/// \param TypeExpr Type tag expression, as it appears in user's code. 6784/// 6785/// \param VD Declaration of an identifier that appears in a type tag. 6786/// 6787/// \param MagicValue Type tag magic value. 6788bool FindTypeTagExpr(const Expr *TypeExpr, const ASTContext &Ctx, 6789 const ValueDecl **VD, uint64_t *MagicValue) { 6790 while(true) { 6791 if (!TypeExpr) 6792 return false; 6793 6794 TypeExpr = TypeExpr->IgnoreParenImpCasts()->IgnoreParenCasts(); 6795 6796 switch (TypeExpr->getStmtClass()) { 6797 case Stmt::UnaryOperatorClass: { 6798 const UnaryOperator *UO = cast<UnaryOperator>(TypeExpr); 6799 if (UO->getOpcode() == UO_AddrOf || UO->getOpcode() == UO_Deref) { 6800 TypeExpr = UO->getSubExpr(); 6801 continue; 6802 } 6803 return false; 6804 } 6805 6806 case Stmt::DeclRefExprClass: { 6807 const DeclRefExpr *DRE = cast<DeclRefExpr>(TypeExpr); 6808 *VD = DRE->getDecl(); 6809 return true; 6810 } 6811 6812 case Stmt::IntegerLiteralClass: { 6813 const IntegerLiteral *IL = cast<IntegerLiteral>(TypeExpr); 6814 llvm::APInt MagicValueAPInt = IL->getValue(); 6815 if (MagicValueAPInt.getActiveBits() <= 64) { 6816 *MagicValue = MagicValueAPInt.getZExtValue(); 6817 return true; 6818 } else 6819 return false; 6820 } 6821 6822 case Stmt::BinaryConditionalOperatorClass: 6823 case Stmt::ConditionalOperatorClass: { 6824 const AbstractConditionalOperator *ACO = 6825 cast<AbstractConditionalOperator>(TypeExpr); 6826 bool Result; 6827 if (ACO->getCond()->EvaluateAsBooleanCondition(Result, Ctx)) { 6828 if (Result) 6829 TypeExpr = ACO->getTrueExpr(); 6830 else 6831 TypeExpr = ACO->getFalseExpr(); 6832 continue; 6833 } 6834 return false; 6835 } 6836 6837 case Stmt::BinaryOperatorClass: { 6838 const BinaryOperator *BO = cast<BinaryOperator>(TypeExpr); 6839 if (BO->getOpcode() == BO_Comma) { 6840 TypeExpr = BO->getRHS(); 6841 continue; 6842 } 6843 return false; 6844 } 6845 6846 default: 6847 return false; 6848 } 6849 } 6850} 6851 6852/// \brief Retrieve the C type corresponding to type tag TypeExpr. 6853/// 6854/// \param TypeExpr Expression that specifies a type tag. 6855/// 6856/// \param MagicValues Registered magic values. 6857/// 6858/// \param FoundWrongKind Set to true if a type tag was found, but of a wrong 6859/// kind. 6860/// 6861/// \param TypeInfo Information about the corresponding C type. 6862/// 6863/// \returns true if the corresponding C type was found. 6864bool GetMatchingCType( 6865 const IdentifierInfo *ArgumentKind, 6866 const Expr *TypeExpr, const ASTContext &Ctx, 6867 const llvm::DenseMap<Sema::TypeTagMagicValue, 6868 Sema::TypeTagData> *MagicValues, 6869 bool &FoundWrongKind, 6870 Sema::TypeTagData &TypeInfo) { 6871 FoundWrongKind = false; 6872 6873 // Variable declaration that has type_tag_for_datatype attribute. 6874 const ValueDecl *VD = NULL; 6875 6876 uint64_t MagicValue; 6877 6878 if (!FindTypeTagExpr(TypeExpr, Ctx, &VD, &MagicValue)) 6879 return false; 6880 6881 if (VD) { 6882 for (specific_attr_iterator<TypeTagForDatatypeAttr> 6883 I = VD->specific_attr_begin<TypeTagForDatatypeAttr>(), 6884 E = VD->specific_attr_end<TypeTagForDatatypeAttr>(); 6885 I != E; ++I) { 6886 if (I->getArgumentKind() != ArgumentKind) { 6887 FoundWrongKind = true; 6888 return false; 6889 } 6890 TypeInfo.Type = I->getMatchingCType(); 6891 TypeInfo.LayoutCompatible = I->getLayoutCompatible(); 6892 TypeInfo.MustBeNull = I->getMustBeNull(); 6893 return true; 6894 } 6895 return false; 6896 } 6897 6898 if (!MagicValues) 6899 return false; 6900 6901 llvm::DenseMap<Sema::TypeTagMagicValue, 6902 Sema::TypeTagData>::const_iterator I = 6903 MagicValues->find(std::make_pair(ArgumentKind, MagicValue)); 6904 if (I == MagicValues->end()) 6905 return false; 6906 6907 TypeInfo = I->second; 6908 return true; 6909} 6910} // unnamed namespace 6911 6912void Sema::RegisterTypeTagForDatatype(const IdentifierInfo *ArgumentKind, 6913 uint64_t MagicValue, QualType Type, 6914 bool LayoutCompatible, 6915 bool MustBeNull) { 6916 if (!TypeTagForDatatypeMagicValues) 6917 TypeTagForDatatypeMagicValues.reset( 6918 new llvm::DenseMap<TypeTagMagicValue, TypeTagData>); 6919 6920 TypeTagMagicValue Magic(ArgumentKind, MagicValue); 6921 (*TypeTagForDatatypeMagicValues)[Magic] = 6922 TypeTagData(Type, LayoutCompatible, MustBeNull); 6923} 6924 6925namespace { 6926bool IsSameCharType(QualType T1, QualType T2) { 6927 const BuiltinType *BT1 = T1->getAs<BuiltinType>(); 6928 if (!BT1) 6929 return false; 6930 6931 const BuiltinType *BT2 = T2->getAs<BuiltinType>(); 6932 if (!BT2) 6933 return false; 6934 6935 BuiltinType::Kind T1Kind = BT1->getKind(); 6936 BuiltinType::Kind T2Kind = BT2->getKind(); 6937 6938 return (T1Kind == BuiltinType::SChar && T2Kind == BuiltinType::Char_S) || 6939 (T1Kind == BuiltinType::UChar && T2Kind == BuiltinType::Char_U) || 6940 (T1Kind == BuiltinType::Char_U && T2Kind == BuiltinType::UChar) || 6941 (T1Kind == BuiltinType::Char_S && T2Kind == BuiltinType::SChar); 6942} 6943} // unnamed namespace 6944 6945void Sema::CheckArgumentWithTypeTag(const ArgumentWithTypeTagAttr *Attr, 6946 const Expr * const *ExprArgs) { 6947 const IdentifierInfo *ArgumentKind = Attr->getArgumentKind(); 6948 bool IsPointerAttr = Attr->getIsPointer(); 6949 6950 const Expr *TypeTagExpr = ExprArgs[Attr->getTypeTagIdx()]; 6951 bool FoundWrongKind; 6952 TypeTagData TypeInfo; 6953 if (!GetMatchingCType(ArgumentKind, TypeTagExpr, Context, 6954 TypeTagForDatatypeMagicValues.get(), 6955 FoundWrongKind, TypeInfo)) { 6956 if (FoundWrongKind) 6957 Diag(TypeTagExpr->getExprLoc(), 6958 diag::warn_type_tag_for_datatype_wrong_kind) 6959 << TypeTagExpr->getSourceRange(); 6960 return; 6961 } 6962 6963 const Expr *ArgumentExpr = ExprArgs[Attr->getArgumentIdx()]; 6964 if (IsPointerAttr) { 6965 // Skip implicit cast of pointer to `void *' (as a function argument). 6966 if (const ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(ArgumentExpr)) 6967 if (ICE->getType()->isVoidPointerType() && 6968 ICE->getCastKind() == CK_BitCast) 6969 ArgumentExpr = ICE->getSubExpr(); 6970 } 6971 QualType ArgumentType = ArgumentExpr->getType(); 6972 6973 // Passing a `void*' pointer shouldn't trigger a warning. 6974 if (IsPointerAttr && ArgumentType->isVoidPointerType()) 6975 return; 6976 6977 if (TypeInfo.MustBeNull) { 6978 // Type tag with matching void type requires a null pointer. 6979 if (!ArgumentExpr->isNullPointerConstant(Context, 6980 Expr::NPC_ValueDependentIsNotNull)) { 6981 Diag(ArgumentExpr->getExprLoc(), 6982 diag::warn_type_safety_null_pointer_required) 6983 << ArgumentKind->getName() 6984 << ArgumentExpr->getSourceRange() 6985 << TypeTagExpr->getSourceRange(); 6986 } 6987 return; 6988 } 6989 6990 QualType RequiredType = TypeInfo.Type; 6991 if (IsPointerAttr) 6992 RequiredType = Context.getPointerType(RequiredType); 6993 6994 bool mismatch = false; 6995 if (!TypeInfo.LayoutCompatible) { 6996 mismatch = !Context.hasSameType(ArgumentType, RequiredType); 6997 6998 // C++11 [basic.fundamental] p1: 6999 // Plain char, signed char, and unsigned char are three distinct types. 7000 // 7001 // But we treat plain `char' as equivalent to `signed char' or `unsigned 7002 // char' depending on the current char signedness mode. 7003 if (mismatch) 7004 if ((IsPointerAttr && IsSameCharType(ArgumentType->getPointeeType(), 7005 RequiredType->getPointeeType())) || 7006 (!IsPointerAttr && IsSameCharType(ArgumentType, RequiredType))) 7007 mismatch = false; 7008 } else 7009 if (IsPointerAttr) 7010 mismatch = !isLayoutCompatible(Context, 7011 ArgumentType->getPointeeType(), 7012 RequiredType->getPointeeType()); 7013 else 7014 mismatch = !isLayoutCompatible(Context, ArgumentType, RequiredType); 7015 7016 if (mismatch) 7017 Diag(ArgumentExpr->getExprLoc(), diag::warn_type_safety_type_mismatch) 7018 << ArgumentType << ArgumentKind->getName() 7019 << TypeInfo.LayoutCompatible << RequiredType 7020 << ArgumentExpr->getSourceRange() 7021 << TypeTagExpr->getSourceRange(); 7022} 7023