SemaChecking.cpp revision ad017fa7a4df7389d245d02a49b3c79ed70bedb9
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/DeclObjC.h" 21#include "clang/AST/EvaluatedExprVisitor.h" 22#include "clang/AST/Expr.h" 23#include "clang/AST/ExprCXX.h" 24#include "clang/AST/ExprObjC.h" 25#include "clang/AST/StmtCXX.h" 26#include "clang/AST/StmtObjC.h" 27#include "clang/Analysis/Analyses/FormatString.h" 28#include "clang/Basic/ConvertUTF.h" 29#include "clang/Basic/TargetBuiltins.h" 30#include "clang/Basic/TargetInfo.h" 31#include "clang/Lex/Preprocessor.h" 32#include "clang/Sema/Initialization.h" 33#include "clang/Sema/Initialization.h" 34#include "clang/Sema/Lookup.h" 35#include "clang/Sema/ScopeInfo.h" 36#include "clang/Sema/Sema.h" 37#include "llvm/ADT/BitVector.h" 38#include "llvm/ADT/STLExtras.h" 39#include "llvm/ADT/SmallString.h" 40#include "llvm/Support/raw_ostream.h" 41#include <limits> 42using namespace clang; 43using namespace sema; 44 45SourceLocation Sema::getLocationOfStringLiteralByte(const StringLiteral *SL, 46 unsigned ByteNo) const { 47 return SL->getLocationOfByte(ByteNo, PP.getSourceManager(), 48 PP.getLangOpts(), PP.getTargetInfo()); 49} 50 51/// Checks that a call expression's argument count is the desired number. 52/// This is useful when doing custom type-checking. Returns true on error. 53static bool checkArgCount(Sema &S, CallExpr *call, unsigned desiredArgCount) { 54 unsigned argCount = call->getNumArgs(); 55 if (argCount == desiredArgCount) return false; 56 57 if (argCount < desiredArgCount) 58 return S.Diag(call->getLocEnd(), diag::err_typecheck_call_too_few_args) 59 << 0 /*function call*/ << desiredArgCount << argCount 60 << call->getSourceRange(); 61 62 // Highlight all the excess arguments. 63 SourceRange range(call->getArg(desiredArgCount)->getLocStart(), 64 call->getArg(argCount - 1)->getLocEnd()); 65 66 return S.Diag(range.getBegin(), diag::err_typecheck_call_too_many_args) 67 << 0 /*function call*/ << desiredArgCount << argCount 68 << call->getArg(1)->getSourceRange(); 69} 70 71/// Check that the first argument to __builtin_annotation is an integer 72/// and the second argument is a non-wide string literal. 73static bool SemaBuiltinAnnotation(Sema &S, CallExpr *TheCall) { 74 if (checkArgCount(S, TheCall, 2)) 75 return true; 76 77 // First argument should be an integer. 78 Expr *ValArg = TheCall->getArg(0); 79 QualType Ty = ValArg->getType(); 80 if (!Ty->isIntegerType()) { 81 S.Diag(ValArg->getLocStart(), diag::err_builtin_annotation_first_arg) 82 << ValArg->getSourceRange(); 83 return true; 84 } 85 86 // Second argument should be a constant string. 87 Expr *StrArg = TheCall->getArg(1)->IgnoreParenCasts(); 88 StringLiteral *Literal = dyn_cast<StringLiteral>(StrArg); 89 if (!Literal || !Literal->isAscii()) { 90 S.Diag(StrArg->getLocStart(), diag::err_builtin_annotation_second_arg) 91 << StrArg->getSourceRange(); 92 return true; 93 } 94 95 TheCall->setType(Ty); 96 return false; 97} 98 99ExprResult 100Sema::CheckBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) { 101 ExprResult TheCallResult(Owned(TheCall)); 102 103 // Find out if any arguments are required to be integer constant expressions. 104 unsigned ICEArguments = 0; 105 ASTContext::GetBuiltinTypeError Error; 106 Context.GetBuiltinType(BuiltinID, Error, &ICEArguments); 107 if (Error != ASTContext::GE_None) 108 ICEArguments = 0; // Don't diagnose previously diagnosed errors. 109 110 // If any arguments are required to be ICE's, check and diagnose. 111 for (unsigned ArgNo = 0; ICEArguments != 0; ++ArgNo) { 112 // Skip arguments not required to be ICE's. 113 if ((ICEArguments & (1 << ArgNo)) == 0) continue; 114 115 llvm::APSInt Result; 116 if (SemaBuiltinConstantArg(TheCall, ArgNo, Result)) 117 return true; 118 ICEArguments &= ~(1 << ArgNo); 119 } 120 121 switch (BuiltinID) { 122 case Builtin::BI__builtin___CFStringMakeConstantString: 123 assert(TheCall->getNumArgs() == 1 && 124 "Wrong # arguments to builtin CFStringMakeConstantString"); 125 if (CheckObjCString(TheCall->getArg(0))) 126 return ExprError(); 127 break; 128 case Builtin::BI__builtin_stdarg_start: 129 case Builtin::BI__builtin_va_start: 130 if (SemaBuiltinVAStart(TheCall)) 131 return ExprError(); 132 break; 133 case Builtin::BI__builtin_isgreater: 134 case Builtin::BI__builtin_isgreaterequal: 135 case Builtin::BI__builtin_isless: 136 case Builtin::BI__builtin_islessequal: 137 case Builtin::BI__builtin_islessgreater: 138 case Builtin::BI__builtin_isunordered: 139 if (SemaBuiltinUnorderedCompare(TheCall)) 140 return ExprError(); 141 break; 142 case Builtin::BI__builtin_fpclassify: 143 if (SemaBuiltinFPClassification(TheCall, 6)) 144 return ExprError(); 145 break; 146 case Builtin::BI__builtin_isfinite: 147 case Builtin::BI__builtin_isinf: 148 case Builtin::BI__builtin_isinf_sign: 149 case Builtin::BI__builtin_isnan: 150 case Builtin::BI__builtin_isnormal: 151 if (SemaBuiltinFPClassification(TheCall, 1)) 152 return ExprError(); 153 break; 154 case Builtin::BI__builtin_shufflevector: 155 return SemaBuiltinShuffleVector(TheCall); 156 // TheCall will be freed by the smart pointer here, but that's fine, since 157 // SemaBuiltinShuffleVector guts it, but then doesn't release it. 158 case Builtin::BI__builtin_prefetch: 159 if (SemaBuiltinPrefetch(TheCall)) 160 return ExprError(); 161 break; 162 case Builtin::BI__builtin_object_size: 163 if (SemaBuiltinObjectSize(TheCall)) 164 return ExprError(); 165 break; 166 case Builtin::BI__builtin_longjmp: 167 if (SemaBuiltinLongjmp(TheCall)) 168 return ExprError(); 169 break; 170 171 case Builtin::BI__builtin_classify_type: 172 if (checkArgCount(*this, TheCall, 1)) return true; 173 TheCall->setType(Context.IntTy); 174 break; 175 case Builtin::BI__builtin_constant_p: 176 if (checkArgCount(*this, TheCall, 1)) return true; 177 TheCall->setType(Context.IntTy); 178 break; 179 case Builtin::BI__sync_fetch_and_add: 180 case Builtin::BI__sync_fetch_and_add_1: 181 case Builtin::BI__sync_fetch_and_add_2: 182 case Builtin::BI__sync_fetch_and_add_4: 183 case Builtin::BI__sync_fetch_and_add_8: 184 case Builtin::BI__sync_fetch_and_add_16: 185 case Builtin::BI__sync_fetch_and_sub: 186 case Builtin::BI__sync_fetch_and_sub_1: 187 case Builtin::BI__sync_fetch_and_sub_2: 188 case Builtin::BI__sync_fetch_and_sub_4: 189 case Builtin::BI__sync_fetch_and_sub_8: 190 case Builtin::BI__sync_fetch_and_sub_16: 191 case Builtin::BI__sync_fetch_and_or: 192 case Builtin::BI__sync_fetch_and_or_1: 193 case Builtin::BI__sync_fetch_and_or_2: 194 case Builtin::BI__sync_fetch_and_or_4: 195 case Builtin::BI__sync_fetch_and_or_8: 196 case Builtin::BI__sync_fetch_and_or_16: 197 case Builtin::BI__sync_fetch_and_and: 198 case Builtin::BI__sync_fetch_and_and_1: 199 case Builtin::BI__sync_fetch_and_and_2: 200 case Builtin::BI__sync_fetch_and_and_4: 201 case Builtin::BI__sync_fetch_and_and_8: 202 case Builtin::BI__sync_fetch_and_and_16: 203 case Builtin::BI__sync_fetch_and_xor: 204 case Builtin::BI__sync_fetch_and_xor_1: 205 case Builtin::BI__sync_fetch_and_xor_2: 206 case Builtin::BI__sync_fetch_and_xor_4: 207 case Builtin::BI__sync_fetch_and_xor_8: 208 case Builtin::BI__sync_fetch_and_xor_16: 209 case Builtin::BI__sync_add_and_fetch: 210 case Builtin::BI__sync_add_and_fetch_1: 211 case Builtin::BI__sync_add_and_fetch_2: 212 case Builtin::BI__sync_add_and_fetch_4: 213 case Builtin::BI__sync_add_and_fetch_8: 214 case Builtin::BI__sync_add_and_fetch_16: 215 case Builtin::BI__sync_sub_and_fetch: 216 case Builtin::BI__sync_sub_and_fetch_1: 217 case Builtin::BI__sync_sub_and_fetch_2: 218 case Builtin::BI__sync_sub_and_fetch_4: 219 case Builtin::BI__sync_sub_and_fetch_8: 220 case Builtin::BI__sync_sub_and_fetch_16: 221 case Builtin::BI__sync_and_and_fetch: 222 case Builtin::BI__sync_and_and_fetch_1: 223 case Builtin::BI__sync_and_and_fetch_2: 224 case Builtin::BI__sync_and_and_fetch_4: 225 case Builtin::BI__sync_and_and_fetch_8: 226 case Builtin::BI__sync_and_and_fetch_16: 227 case Builtin::BI__sync_or_and_fetch: 228 case Builtin::BI__sync_or_and_fetch_1: 229 case Builtin::BI__sync_or_and_fetch_2: 230 case Builtin::BI__sync_or_and_fetch_4: 231 case Builtin::BI__sync_or_and_fetch_8: 232 case Builtin::BI__sync_or_and_fetch_16: 233 case Builtin::BI__sync_xor_and_fetch: 234 case Builtin::BI__sync_xor_and_fetch_1: 235 case Builtin::BI__sync_xor_and_fetch_2: 236 case Builtin::BI__sync_xor_and_fetch_4: 237 case Builtin::BI__sync_xor_and_fetch_8: 238 case Builtin::BI__sync_xor_and_fetch_16: 239 case Builtin::BI__sync_val_compare_and_swap: 240 case Builtin::BI__sync_val_compare_and_swap_1: 241 case Builtin::BI__sync_val_compare_and_swap_2: 242 case Builtin::BI__sync_val_compare_and_swap_4: 243 case Builtin::BI__sync_val_compare_and_swap_8: 244 case Builtin::BI__sync_val_compare_and_swap_16: 245 case Builtin::BI__sync_bool_compare_and_swap: 246 case Builtin::BI__sync_bool_compare_and_swap_1: 247 case Builtin::BI__sync_bool_compare_and_swap_2: 248 case Builtin::BI__sync_bool_compare_and_swap_4: 249 case Builtin::BI__sync_bool_compare_and_swap_8: 250 case Builtin::BI__sync_bool_compare_and_swap_16: 251 case Builtin::BI__sync_lock_test_and_set: 252 case Builtin::BI__sync_lock_test_and_set_1: 253 case Builtin::BI__sync_lock_test_and_set_2: 254 case Builtin::BI__sync_lock_test_and_set_4: 255 case Builtin::BI__sync_lock_test_and_set_8: 256 case Builtin::BI__sync_lock_test_and_set_16: 257 case Builtin::BI__sync_lock_release: 258 case Builtin::BI__sync_lock_release_1: 259 case Builtin::BI__sync_lock_release_2: 260 case Builtin::BI__sync_lock_release_4: 261 case Builtin::BI__sync_lock_release_8: 262 case Builtin::BI__sync_lock_release_16: 263 case Builtin::BI__sync_swap: 264 case Builtin::BI__sync_swap_1: 265 case Builtin::BI__sync_swap_2: 266 case Builtin::BI__sync_swap_4: 267 case Builtin::BI__sync_swap_8: 268 case Builtin::BI__sync_swap_16: 269 return SemaBuiltinAtomicOverloaded(TheCallResult); 270#define BUILTIN(ID, TYPE, ATTRS) 271#define ATOMIC_BUILTIN(ID, TYPE, ATTRS) \ 272 case Builtin::BI##ID: \ 273 return SemaAtomicOpsOverloaded(TheCallResult, AtomicExpr::AO##ID); 274#include "clang/Basic/Builtins.def" 275 case Builtin::BI__builtin_annotation: 276 if (SemaBuiltinAnnotation(*this, TheCall)) 277 return ExprError(); 278 break; 279 } 280 281 // Since the target specific builtins for each arch overlap, only check those 282 // of the arch we are compiling for. 283 if (BuiltinID >= Builtin::FirstTSBuiltin) { 284 switch (Context.getTargetInfo().getTriple().getArch()) { 285 case llvm::Triple::arm: 286 case llvm::Triple::thumb: 287 if (CheckARMBuiltinFunctionCall(BuiltinID, TheCall)) 288 return ExprError(); 289 break; 290 case llvm::Triple::mips: 291 case llvm::Triple::mipsel: 292 case llvm::Triple::mips64: 293 case llvm::Triple::mips64el: 294 if (CheckMipsBuiltinFunctionCall(BuiltinID, TheCall)) 295 return ExprError(); 296 break; 297 default: 298 break; 299 } 300 } 301 302 return TheCallResult; 303} 304 305// Get the valid immediate range for the specified NEON type code. 306static unsigned RFT(unsigned t, bool shift = false) { 307 NeonTypeFlags Type(t); 308 int IsQuad = Type.isQuad(); 309 switch (Type.getEltType()) { 310 case NeonTypeFlags::Int8: 311 case NeonTypeFlags::Poly8: 312 return shift ? 7 : (8 << IsQuad) - 1; 313 case NeonTypeFlags::Int16: 314 case NeonTypeFlags::Poly16: 315 return shift ? 15 : (4 << IsQuad) - 1; 316 case NeonTypeFlags::Int32: 317 return shift ? 31 : (2 << IsQuad) - 1; 318 case NeonTypeFlags::Int64: 319 return shift ? 63 : (1 << IsQuad) - 1; 320 case NeonTypeFlags::Float16: 321 assert(!shift && "cannot shift float types!"); 322 return (4 << IsQuad) - 1; 323 case NeonTypeFlags::Float32: 324 assert(!shift && "cannot shift float types!"); 325 return (2 << IsQuad) - 1; 326 } 327 llvm_unreachable("Invalid NeonTypeFlag!"); 328} 329 330/// getNeonEltType - Return the QualType corresponding to the elements of 331/// the vector type specified by the NeonTypeFlags. This is used to check 332/// the pointer arguments for Neon load/store intrinsics. 333static QualType getNeonEltType(NeonTypeFlags Flags, ASTContext &Context) { 334 switch (Flags.getEltType()) { 335 case NeonTypeFlags::Int8: 336 return Flags.isUnsigned() ? Context.UnsignedCharTy : Context.SignedCharTy; 337 case NeonTypeFlags::Int16: 338 return Flags.isUnsigned() ? Context.UnsignedShortTy : Context.ShortTy; 339 case NeonTypeFlags::Int32: 340 return Flags.isUnsigned() ? Context.UnsignedIntTy : Context.IntTy; 341 case NeonTypeFlags::Int64: 342 return Flags.isUnsigned() ? Context.UnsignedLongLongTy : Context.LongLongTy; 343 case NeonTypeFlags::Poly8: 344 return Context.SignedCharTy; 345 case NeonTypeFlags::Poly16: 346 return Context.ShortTy; 347 case NeonTypeFlags::Float16: 348 return Context.UnsignedShortTy; 349 case NeonTypeFlags::Float32: 350 return Context.FloatTy; 351 } 352 llvm_unreachable("Invalid NeonTypeFlag!"); 353} 354 355bool Sema::CheckARMBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) { 356 llvm::APSInt Result; 357 358 uint64_t mask = 0; 359 unsigned TV = 0; 360 int PtrArgNum = -1; 361 bool HasConstPtr = false; 362 switch (BuiltinID) { 363#define GET_NEON_OVERLOAD_CHECK 364#include "clang/Basic/arm_neon.inc" 365#undef GET_NEON_OVERLOAD_CHECK 366 } 367 368 // For NEON intrinsics which are overloaded on vector element type, validate 369 // the immediate which specifies which variant to emit. 370 unsigned ImmArg = TheCall->getNumArgs()-1; 371 if (mask) { 372 if (SemaBuiltinConstantArg(TheCall, ImmArg, Result)) 373 return true; 374 375 TV = Result.getLimitedValue(64); 376 if ((TV > 63) || (mask & (1ULL << TV)) == 0) 377 return Diag(TheCall->getLocStart(), diag::err_invalid_neon_type_code) 378 << TheCall->getArg(ImmArg)->getSourceRange(); 379 } 380 381 if (PtrArgNum >= 0) { 382 // Check that pointer arguments have the specified type. 383 Expr *Arg = TheCall->getArg(PtrArgNum); 384 if (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(Arg)) 385 Arg = ICE->getSubExpr(); 386 ExprResult RHS = DefaultFunctionArrayLvalueConversion(Arg); 387 QualType RHSTy = RHS.get()->getType(); 388 QualType EltTy = getNeonEltType(NeonTypeFlags(TV), Context); 389 if (HasConstPtr) 390 EltTy = EltTy.withConst(); 391 QualType LHSTy = Context.getPointerType(EltTy); 392 AssignConvertType ConvTy; 393 ConvTy = CheckSingleAssignmentConstraints(LHSTy, RHS); 394 if (RHS.isInvalid()) 395 return true; 396 if (DiagnoseAssignmentResult(ConvTy, Arg->getLocStart(), LHSTy, RHSTy, 397 RHS.get(), AA_Assigning)) 398 return true; 399 } 400 401 // For NEON intrinsics which take an immediate value as part of the 402 // instruction, range check them here. 403 unsigned i = 0, l = 0, u = 0; 404 switch (BuiltinID) { 405 default: return false; 406 case ARM::BI__builtin_arm_ssat: i = 1; l = 1; u = 31; break; 407 case ARM::BI__builtin_arm_usat: i = 1; u = 31; break; 408 case ARM::BI__builtin_arm_vcvtr_f: 409 case ARM::BI__builtin_arm_vcvtr_d: i = 1; u = 1; break; 410#define GET_NEON_IMMEDIATE_CHECK 411#include "clang/Basic/arm_neon.inc" 412#undef GET_NEON_IMMEDIATE_CHECK 413 }; 414 415 // We can't check the value of a dependent argument. 416 if (TheCall->getArg(i)->isTypeDependent() || 417 TheCall->getArg(i)->isValueDependent()) 418 return false; 419 420 // Check that the immediate argument is actually a constant. 421 if (SemaBuiltinConstantArg(TheCall, i, Result)) 422 return true; 423 424 // Range check against the upper/lower values for this isntruction. 425 unsigned Val = Result.getZExtValue(); 426 if (Val < l || Val > (u + l)) 427 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 428 << l << u+l << TheCall->getArg(i)->getSourceRange(); 429 430 // FIXME: VFP Intrinsics should error if VFP not present. 431 return false; 432} 433 434bool Sema::CheckMipsBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) { 435 unsigned i = 0, l = 0, u = 0; 436 switch (BuiltinID) { 437 default: return false; 438 case Mips::BI__builtin_mips_wrdsp: i = 1; l = 0; u = 63; break; 439 case Mips::BI__builtin_mips_rddsp: i = 0; l = 0; u = 63; break; 440 case Mips::BI__builtin_mips_append: i = 2; l = 0; u = 31; break; 441 case Mips::BI__builtin_mips_balign: i = 2; l = 0; u = 3; break; 442 case Mips::BI__builtin_mips_precr_sra_ph_w: i = 2; l = 0; u = 31; break; 443 case Mips::BI__builtin_mips_precr_sra_r_ph_w: i = 2; l = 0; u = 31; break; 444 case Mips::BI__builtin_mips_prepend: i = 2; l = 0; u = 31; break; 445 }; 446 447 // We can't check the value of a dependent argument. 448 if (TheCall->getArg(i)->isTypeDependent() || 449 TheCall->getArg(i)->isValueDependent()) 450 return false; 451 452 // Check that the immediate argument is actually a constant. 453 llvm::APSInt Result; 454 if (SemaBuiltinConstantArg(TheCall, i, Result)) 455 return true; 456 457 // Range check against the upper/lower values for this instruction. 458 unsigned Val = Result.getZExtValue(); 459 if (Val < l || Val > u) 460 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 461 << l << u << TheCall->getArg(i)->getSourceRange(); 462 463 return false; 464} 465 466/// Given a FunctionDecl's FormatAttr, attempts to populate the FomatStringInfo 467/// parameter with the FormatAttr's correct format_idx and firstDataArg. 468/// Returns true when the format fits the function and the FormatStringInfo has 469/// been populated. 470bool Sema::getFormatStringInfo(const FormatAttr *Format, bool IsCXXMember, 471 FormatStringInfo *FSI) { 472 FSI->HasVAListArg = Format->getFirstArg() == 0; 473 FSI->FormatIdx = Format->getFormatIdx() - 1; 474 FSI->FirstDataArg = FSI->HasVAListArg ? 0 : Format->getFirstArg() - 1; 475 476 // The way the format attribute works in GCC, the implicit this argument 477 // of member functions is counted. However, it doesn't appear in our own 478 // lists, so decrement format_idx in that case. 479 if (IsCXXMember) { 480 if(FSI->FormatIdx == 0) 481 return false; 482 --FSI->FormatIdx; 483 if (FSI->FirstDataArg != 0) 484 --FSI->FirstDataArg; 485 } 486 return true; 487} 488 489/// Handles the checks for format strings, non-POD arguments to vararg 490/// functions, and NULL arguments passed to non-NULL parameters. 491void Sema::checkCall(NamedDecl *FDecl, Expr **Args, 492 unsigned NumArgs, 493 unsigned NumProtoArgs, 494 bool IsMemberFunction, 495 SourceLocation Loc, 496 SourceRange Range, 497 VariadicCallType CallType) { 498 if (CurContext->isDependentContext()) 499 return; 500 501 // Printf and scanf checking. 502 bool HandledFormatString = false; 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, NumArgs, IsMemberFunction, CallType, 507 Loc, 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 < NumArgs; ++ArgIdx) { 514 // Args[ArgIdx] can be null in malformed code. 515 if (Expr *Arg = Args[ArgIdx]) 516 variadicArgumentPODCheck(Arg, CallType); 517 } 518 519 for (specific_attr_iterator<NonNullAttr> 520 I = FDecl->specific_attr_begin<NonNullAttr>(), 521 E = FDecl->specific_attr_end<NonNullAttr>(); I != E; ++I) 522 CheckNonNullArguments(*I, Args, Loc); 523 524 // Type safety checking. 525 for (specific_attr_iterator<ArgumentWithTypeTagAttr> 526 i = FDecl->specific_attr_begin<ArgumentWithTypeTagAttr>(), 527 e = FDecl->specific_attr_end<ArgumentWithTypeTagAttr>(); i != e; ++i) { 528 CheckArgumentWithTypeTag(*i, Args); 529 } 530} 531 532/// CheckConstructorCall - Check a constructor call for correctness and safety 533/// properties not enforced by the C type system. 534void Sema::CheckConstructorCall(FunctionDecl *FDecl, Expr **Args, 535 unsigned NumArgs, 536 const FunctionProtoType *Proto, 537 SourceLocation Loc) { 538 VariadicCallType CallType = 539 Proto->isVariadic() ? VariadicConstructor : VariadicDoesNotApply; 540 checkCall(FDecl, Args, NumArgs, Proto->getNumArgs(), 541 /*IsMemberFunction=*/true, Loc, SourceRange(), CallType); 542} 543 544/// CheckFunctionCall - Check a direct function call for various correctness 545/// and safety properties not strictly enforced by the C type system. 546bool Sema::CheckFunctionCall(FunctionDecl *FDecl, CallExpr *TheCall, 547 const FunctionProtoType *Proto) { 548 bool IsMemberOperatorCall = isa<CXXOperatorCallExpr>(TheCall) && 549 isa<CXXMethodDecl>(FDecl); 550 bool IsMemberFunction = isa<CXXMemberCallExpr>(TheCall) || 551 IsMemberOperatorCall; 552 VariadicCallType CallType = getVariadicCallType(FDecl, Proto, 553 TheCall->getCallee()); 554 unsigned NumProtoArgs = Proto ? Proto->getNumArgs() : 0; 555 Expr** Args = TheCall->getArgs(); 556 unsigned NumArgs = TheCall->getNumArgs(); 557 if (IsMemberOperatorCall) { 558 // If this is a call to a member operator, hide the first argument 559 // from checkCall. 560 // FIXME: Our choice of AST representation here is less than ideal. 561 ++Args; 562 --NumArgs; 563 } 564 checkCall(FDecl, Args, NumArgs, NumProtoArgs, 565 IsMemberFunction, TheCall->getRParenLoc(), 566 TheCall->getCallee()->getSourceRange(), CallType); 567 568 IdentifierInfo *FnInfo = FDecl->getIdentifier(); 569 // None of the checks below are needed for functions that don't have 570 // simple names (e.g., C++ conversion functions). 571 if (!FnInfo) 572 return false; 573 574 unsigned CMId = FDecl->getMemoryFunctionKind(); 575 if (CMId == 0) 576 return false; 577 578 // Handle memory setting and copying functions. 579 if (CMId == Builtin::BIstrlcpy || CMId == Builtin::BIstrlcat) 580 CheckStrlcpycatArguments(TheCall, FnInfo); 581 else if (CMId == Builtin::BIstrncat) 582 CheckStrncatArguments(TheCall, FnInfo); 583 else 584 CheckMemaccessArguments(TheCall, CMId, FnInfo); 585 586 return false; 587} 588 589bool Sema::CheckObjCMethodCall(ObjCMethodDecl *Method, SourceLocation lbrac, 590 Expr **Args, unsigned NumArgs) { 591 VariadicCallType CallType = 592 Method->isVariadic() ? VariadicMethod : VariadicDoesNotApply; 593 594 checkCall(Method, Args, NumArgs, Method->param_size(), 595 /*IsMemberFunction=*/false, 596 lbrac, Method->getSourceRange(), CallType); 597 598 return false; 599} 600 601bool Sema::CheckBlockCall(NamedDecl *NDecl, CallExpr *TheCall, 602 const FunctionProtoType *Proto) { 603 const VarDecl *V = dyn_cast<VarDecl>(NDecl); 604 if (!V) 605 return false; 606 607 QualType Ty = V->getType(); 608 if (!Ty->isBlockPointerType()) 609 return false; 610 611 VariadicCallType CallType = 612 Proto && Proto->isVariadic() ? VariadicBlock : VariadicDoesNotApply ; 613 unsigned NumProtoArgs = Proto ? Proto->getNumArgs() : 0; 614 615 checkCall(NDecl, TheCall->getArgs(), TheCall->getNumArgs(), 616 NumProtoArgs, /*IsMemberFunction=*/false, 617 TheCall->getRParenLoc(), 618 TheCall->getCallee()->getSourceRange(), CallType); 619 620 return false; 621} 622 623ExprResult Sema::SemaAtomicOpsOverloaded(ExprResult TheCallResult, 624 AtomicExpr::AtomicOp Op) { 625 CallExpr *TheCall = cast<CallExpr>(TheCallResult.get()); 626 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 627 628 // All these operations take one of the following forms: 629 enum { 630 // C __c11_atomic_init(A *, C) 631 Init, 632 // C __c11_atomic_load(A *, int) 633 Load, 634 // void __atomic_load(A *, CP, int) 635 Copy, 636 // C __c11_atomic_add(A *, M, int) 637 Arithmetic, 638 // C __atomic_exchange_n(A *, CP, int) 639 Xchg, 640 // void __atomic_exchange(A *, C *, CP, int) 641 GNUXchg, 642 // bool __c11_atomic_compare_exchange_strong(A *, C *, CP, int, int) 643 C11CmpXchg, 644 // bool __atomic_compare_exchange(A *, C *, CP, bool, int, int) 645 GNUCmpXchg 646 } Form = Init; 647 const unsigned NumArgs[] = { 2, 2, 3, 3, 3, 4, 5, 6 }; 648 const unsigned NumVals[] = { 1, 0, 1, 1, 1, 2, 2, 3 }; 649 // where: 650 // C is an appropriate type, 651 // A is volatile _Atomic(C) for __c11 builtins and is C for GNU builtins, 652 // CP is C for __c11 builtins and GNU _n builtins and is C * otherwise, 653 // M is C if C is an integer, and ptrdiff_t if C is a pointer, and 654 // the int parameters are for orderings. 655 656 assert(AtomicExpr::AO__c11_atomic_init == 0 && 657 AtomicExpr::AO__c11_atomic_fetch_xor + 1 == AtomicExpr::AO__atomic_load 658 && "need to update code for modified C11 atomics"); 659 bool IsC11 = Op >= AtomicExpr::AO__c11_atomic_init && 660 Op <= AtomicExpr::AO__c11_atomic_fetch_xor; 661 bool IsN = Op == AtomicExpr::AO__atomic_load_n || 662 Op == AtomicExpr::AO__atomic_store_n || 663 Op == AtomicExpr::AO__atomic_exchange_n || 664 Op == AtomicExpr::AO__atomic_compare_exchange_n; 665 bool IsAddSub = false; 666 667 switch (Op) { 668 case AtomicExpr::AO__c11_atomic_init: 669 Form = Init; 670 break; 671 672 case AtomicExpr::AO__c11_atomic_load: 673 case AtomicExpr::AO__atomic_load_n: 674 Form = Load; 675 break; 676 677 case AtomicExpr::AO__c11_atomic_store: 678 case AtomicExpr::AO__atomic_load: 679 case AtomicExpr::AO__atomic_store: 680 case AtomicExpr::AO__atomic_store_n: 681 Form = Copy; 682 break; 683 684 case AtomicExpr::AO__c11_atomic_fetch_add: 685 case AtomicExpr::AO__c11_atomic_fetch_sub: 686 case AtomicExpr::AO__atomic_fetch_add: 687 case AtomicExpr::AO__atomic_fetch_sub: 688 case AtomicExpr::AO__atomic_add_fetch: 689 case AtomicExpr::AO__atomic_sub_fetch: 690 IsAddSub = true; 691 // Fall through. 692 case AtomicExpr::AO__c11_atomic_fetch_and: 693 case AtomicExpr::AO__c11_atomic_fetch_or: 694 case AtomicExpr::AO__c11_atomic_fetch_xor: 695 case AtomicExpr::AO__atomic_fetch_and: 696 case AtomicExpr::AO__atomic_fetch_or: 697 case AtomicExpr::AO__atomic_fetch_xor: 698 case AtomicExpr::AO__atomic_fetch_nand: 699 case AtomicExpr::AO__atomic_and_fetch: 700 case AtomicExpr::AO__atomic_or_fetch: 701 case AtomicExpr::AO__atomic_xor_fetch: 702 case AtomicExpr::AO__atomic_nand_fetch: 703 Form = Arithmetic; 704 break; 705 706 case AtomicExpr::AO__c11_atomic_exchange: 707 case AtomicExpr::AO__atomic_exchange_n: 708 Form = Xchg; 709 break; 710 711 case AtomicExpr::AO__atomic_exchange: 712 Form = GNUXchg; 713 break; 714 715 case AtomicExpr::AO__c11_atomic_compare_exchange_strong: 716 case AtomicExpr::AO__c11_atomic_compare_exchange_weak: 717 Form = C11CmpXchg; 718 break; 719 720 case AtomicExpr::AO__atomic_compare_exchange: 721 case AtomicExpr::AO__atomic_compare_exchange_n: 722 Form = GNUCmpXchg; 723 break; 724 } 725 726 // Check we have the right number of arguments. 727 if (TheCall->getNumArgs() < NumArgs[Form]) { 728 Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 729 << 0 << NumArgs[Form] << TheCall->getNumArgs() 730 << TheCall->getCallee()->getSourceRange(); 731 return ExprError(); 732 } else if (TheCall->getNumArgs() > NumArgs[Form]) { 733 Diag(TheCall->getArg(NumArgs[Form])->getLocStart(), 734 diag::err_typecheck_call_too_many_args) 735 << 0 << NumArgs[Form] << TheCall->getNumArgs() 736 << TheCall->getCallee()->getSourceRange(); 737 return ExprError(); 738 } 739 740 // Inspect the first argument of the atomic operation. 741 Expr *Ptr = TheCall->getArg(0); 742 Ptr = DefaultFunctionArrayLvalueConversion(Ptr).get(); 743 const PointerType *pointerType = Ptr->getType()->getAs<PointerType>(); 744 if (!pointerType) { 745 Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer) 746 << Ptr->getType() << Ptr->getSourceRange(); 747 return ExprError(); 748 } 749 750 // For a __c11 builtin, this should be a pointer to an _Atomic type. 751 QualType AtomTy = pointerType->getPointeeType(); // 'A' 752 QualType ValType = AtomTy; // 'C' 753 if (IsC11) { 754 if (!AtomTy->isAtomicType()) { 755 Diag(DRE->getLocStart(), diag::err_atomic_op_needs_atomic) 756 << Ptr->getType() << Ptr->getSourceRange(); 757 return ExprError(); 758 } 759 if (AtomTy.isConstQualified()) { 760 Diag(DRE->getLocStart(), diag::err_atomic_op_needs_non_const_atomic) 761 << Ptr->getType() << Ptr->getSourceRange(); 762 return ExprError(); 763 } 764 ValType = AtomTy->getAs<AtomicType>()->getValueType(); 765 } 766 767 // For an arithmetic operation, the implied arithmetic must be well-formed. 768 if (Form == Arithmetic) { 769 // gcc does not enforce these rules for GNU atomics, but we do so for sanity. 770 if (IsAddSub && !ValType->isIntegerType() && !ValType->isPointerType()) { 771 Diag(DRE->getLocStart(), diag::err_atomic_op_needs_atomic_int_or_ptr) 772 << IsC11 << Ptr->getType() << Ptr->getSourceRange(); 773 return ExprError(); 774 } 775 if (!IsAddSub && !ValType->isIntegerType()) { 776 Diag(DRE->getLocStart(), diag::err_atomic_op_bitwise_needs_atomic_int) 777 << IsC11 << Ptr->getType() << Ptr->getSourceRange(); 778 return ExprError(); 779 } 780 } else if (IsN && !ValType->isIntegerType() && !ValType->isPointerType()) { 781 // For __atomic_*_n operations, the value type must be a scalar integral or 782 // pointer type which is 1, 2, 4, 8 or 16 bytes in length. 783 Diag(DRE->getLocStart(), diag::err_atomic_op_needs_atomic_int_or_ptr) 784 << IsC11 << Ptr->getType() << Ptr->getSourceRange(); 785 return ExprError(); 786 } 787 788 if (!IsC11 && !AtomTy.isTriviallyCopyableType(Context)) { 789 // For GNU atomics, require a trivially-copyable type. This is not part of 790 // the GNU atomics specification, but we enforce it for sanity. 791 Diag(DRE->getLocStart(), diag::err_atomic_op_needs_trivial_copy) 792 << Ptr->getType() << Ptr->getSourceRange(); 793 return ExprError(); 794 } 795 796 // FIXME: For any builtin other than a load, the ValType must not be 797 // const-qualified. 798 799 switch (ValType.getObjCLifetime()) { 800 case Qualifiers::OCL_None: 801 case Qualifiers::OCL_ExplicitNone: 802 // okay 803 break; 804 805 case Qualifiers::OCL_Weak: 806 case Qualifiers::OCL_Strong: 807 case Qualifiers::OCL_Autoreleasing: 808 // FIXME: Can this happen? By this point, ValType should be known 809 // to be trivially copyable. 810 Diag(DRE->getLocStart(), diag::err_arc_atomic_ownership) 811 << ValType << Ptr->getSourceRange(); 812 return ExprError(); 813 } 814 815 QualType ResultType = ValType; 816 if (Form == Copy || Form == GNUXchg || Form == Init) 817 ResultType = Context.VoidTy; 818 else if (Form == C11CmpXchg || Form == GNUCmpXchg) 819 ResultType = Context.BoolTy; 820 821 // The type of a parameter passed 'by value'. In the GNU atomics, such 822 // arguments are actually passed as pointers. 823 QualType ByValType = ValType; // 'CP' 824 if (!IsC11 && !IsN) 825 ByValType = Ptr->getType(); 826 827 // The first argument --- the pointer --- has a fixed type; we 828 // deduce the types of the rest of the arguments accordingly. Walk 829 // the remaining arguments, converting them to the deduced value type. 830 for (unsigned i = 1; i != NumArgs[Form]; ++i) { 831 QualType Ty; 832 if (i < NumVals[Form] + 1) { 833 switch (i) { 834 case 1: 835 // The second argument is the non-atomic operand. For arithmetic, this 836 // is always passed by value, and for a compare_exchange it is always 837 // passed by address. For the rest, GNU uses by-address and C11 uses 838 // by-value. 839 assert(Form != Load); 840 if (Form == Init || (Form == Arithmetic && ValType->isIntegerType())) 841 Ty = ValType; 842 else if (Form == Copy || Form == Xchg) 843 Ty = ByValType; 844 else if (Form == Arithmetic) 845 Ty = Context.getPointerDiffType(); 846 else 847 Ty = Context.getPointerType(ValType.getUnqualifiedType()); 848 break; 849 case 2: 850 // The third argument to compare_exchange / GNU exchange is a 851 // (pointer to a) desired value. 852 Ty = ByValType; 853 break; 854 case 3: 855 // The fourth argument to GNU compare_exchange is a 'weak' flag. 856 Ty = Context.BoolTy; 857 break; 858 } 859 } else { 860 // The order(s) are always converted to int. 861 Ty = Context.IntTy; 862 } 863 864 InitializedEntity Entity = 865 InitializedEntity::InitializeParameter(Context, Ty, false); 866 ExprResult Arg = TheCall->getArg(i); 867 Arg = PerformCopyInitialization(Entity, SourceLocation(), Arg); 868 if (Arg.isInvalid()) 869 return true; 870 TheCall->setArg(i, Arg.get()); 871 } 872 873 // Permute the arguments into a 'consistent' order. 874 SmallVector<Expr*, 5> SubExprs; 875 SubExprs.push_back(Ptr); 876 switch (Form) { 877 case Init: 878 // Note, AtomicExpr::getVal1() has a special case for this atomic. 879 SubExprs.push_back(TheCall->getArg(1)); // Val1 880 break; 881 case Load: 882 SubExprs.push_back(TheCall->getArg(1)); // Order 883 break; 884 case Copy: 885 case Arithmetic: 886 case Xchg: 887 SubExprs.push_back(TheCall->getArg(2)); // Order 888 SubExprs.push_back(TheCall->getArg(1)); // Val1 889 break; 890 case GNUXchg: 891 // Note, AtomicExpr::getVal2() has a special case for this atomic. 892 SubExprs.push_back(TheCall->getArg(3)); // Order 893 SubExprs.push_back(TheCall->getArg(1)); // Val1 894 SubExprs.push_back(TheCall->getArg(2)); // Val2 895 break; 896 case C11CmpXchg: 897 SubExprs.push_back(TheCall->getArg(3)); // Order 898 SubExprs.push_back(TheCall->getArg(1)); // Val1 899 SubExprs.push_back(TheCall->getArg(4)); // OrderFail 900 SubExprs.push_back(TheCall->getArg(2)); // Val2 901 break; 902 case GNUCmpXchg: 903 SubExprs.push_back(TheCall->getArg(4)); // Order 904 SubExprs.push_back(TheCall->getArg(1)); // Val1 905 SubExprs.push_back(TheCall->getArg(5)); // OrderFail 906 SubExprs.push_back(TheCall->getArg(2)); // Val2 907 SubExprs.push_back(TheCall->getArg(3)); // Weak 908 break; 909 } 910 911 return Owned(new (Context) AtomicExpr(TheCall->getCallee()->getLocStart(), 912 SubExprs, ResultType, Op, 913 TheCall->getRParenLoc())); 914} 915 916 917/// checkBuiltinArgument - Given a call to a builtin function, perform 918/// normal type-checking on the given argument, updating the call in 919/// place. This is useful when a builtin function requires custom 920/// type-checking for some of its arguments but not necessarily all of 921/// them. 922/// 923/// Returns true on error. 924static bool checkBuiltinArgument(Sema &S, CallExpr *E, unsigned ArgIndex) { 925 FunctionDecl *Fn = E->getDirectCallee(); 926 assert(Fn && "builtin call without direct callee!"); 927 928 ParmVarDecl *Param = Fn->getParamDecl(ArgIndex); 929 InitializedEntity Entity = 930 InitializedEntity::InitializeParameter(S.Context, Param); 931 932 ExprResult Arg = E->getArg(0); 933 Arg = S.PerformCopyInitialization(Entity, SourceLocation(), Arg); 934 if (Arg.isInvalid()) 935 return true; 936 937 E->setArg(ArgIndex, Arg.take()); 938 return false; 939} 940 941/// SemaBuiltinAtomicOverloaded - We have a call to a function like 942/// __sync_fetch_and_add, which is an overloaded function based on the pointer 943/// type of its first argument. The main ActOnCallExpr routines have already 944/// promoted the types of arguments because all of these calls are prototyped as 945/// void(...). 946/// 947/// This function goes through and does final semantic checking for these 948/// builtins, 949ExprResult 950Sema::SemaBuiltinAtomicOverloaded(ExprResult TheCallResult) { 951 CallExpr *TheCall = (CallExpr *)TheCallResult.get(); 952 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 953 FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl()); 954 955 // Ensure that we have at least one argument to do type inference from. 956 if (TheCall->getNumArgs() < 1) { 957 Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args_at_least) 958 << 0 << 1 << TheCall->getNumArgs() 959 << TheCall->getCallee()->getSourceRange(); 960 return ExprError(); 961 } 962 963 // Inspect the first argument of the atomic builtin. This should always be 964 // a pointer type, whose element is an integral scalar or pointer type. 965 // Because it is a pointer type, we don't have to worry about any implicit 966 // casts here. 967 // FIXME: We don't allow floating point scalars as input. 968 Expr *FirstArg = TheCall->getArg(0); 969 ExprResult FirstArgResult = DefaultFunctionArrayLvalueConversion(FirstArg); 970 if (FirstArgResult.isInvalid()) 971 return ExprError(); 972 FirstArg = FirstArgResult.take(); 973 TheCall->setArg(0, FirstArg); 974 975 const PointerType *pointerType = FirstArg->getType()->getAs<PointerType>(); 976 if (!pointerType) { 977 Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer) 978 << FirstArg->getType() << FirstArg->getSourceRange(); 979 return ExprError(); 980 } 981 982 QualType ValType = pointerType->getPointeeType(); 983 if (!ValType->isIntegerType() && !ValType->isAnyPointerType() && 984 !ValType->isBlockPointerType()) { 985 Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer_intptr) 986 << FirstArg->getType() << FirstArg->getSourceRange(); 987 return ExprError(); 988 } 989 990 switch (ValType.getObjCLifetime()) { 991 case Qualifiers::OCL_None: 992 case Qualifiers::OCL_ExplicitNone: 993 // okay 994 break; 995 996 case Qualifiers::OCL_Weak: 997 case Qualifiers::OCL_Strong: 998 case Qualifiers::OCL_Autoreleasing: 999 Diag(DRE->getLocStart(), diag::err_arc_atomic_ownership) 1000 << ValType << FirstArg->getSourceRange(); 1001 return ExprError(); 1002 } 1003 1004 // Strip any qualifiers off ValType. 1005 ValType = ValType.getUnqualifiedType(); 1006 1007 // The majority of builtins return a value, but a few have special return 1008 // types, so allow them to override appropriately below. 1009 QualType ResultType = ValType; 1010 1011 // We need to figure out which concrete builtin this maps onto. For example, 1012 // __sync_fetch_and_add with a 2 byte object turns into 1013 // __sync_fetch_and_add_2. 1014#define BUILTIN_ROW(x) \ 1015 { Builtin::BI##x##_1, Builtin::BI##x##_2, Builtin::BI##x##_4, \ 1016 Builtin::BI##x##_8, Builtin::BI##x##_16 } 1017 1018 static const unsigned BuiltinIndices[][5] = { 1019 BUILTIN_ROW(__sync_fetch_and_add), 1020 BUILTIN_ROW(__sync_fetch_and_sub), 1021 BUILTIN_ROW(__sync_fetch_and_or), 1022 BUILTIN_ROW(__sync_fetch_and_and), 1023 BUILTIN_ROW(__sync_fetch_and_xor), 1024 1025 BUILTIN_ROW(__sync_add_and_fetch), 1026 BUILTIN_ROW(__sync_sub_and_fetch), 1027 BUILTIN_ROW(__sync_and_and_fetch), 1028 BUILTIN_ROW(__sync_or_and_fetch), 1029 BUILTIN_ROW(__sync_xor_and_fetch), 1030 1031 BUILTIN_ROW(__sync_val_compare_and_swap), 1032 BUILTIN_ROW(__sync_bool_compare_and_swap), 1033 BUILTIN_ROW(__sync_lock_test_and_set), 1034 BUILTIN_ROW(__sync_lock_release), 1035 BUILTIN_ROW(__sync_swap) 1036 }; 1037#undef BUILTIN_ROW 1038 1039 // Determine the index of the size. 1040 unsigned SizeIndex; 1041 switch (Context.getTypeSizeInChars(ValType).getQuantity()) { 1042 case 1: SizeIndex = 0; break; 1043 case 2: SizeIndex = 1; break; 1044 case 4: SizeIndex = 2; break; 1045 case 8: SizeIndex = 3; break; 1046 case 16: SizeIndex = 4; break; 1047 default: 1048 Diag(DRE->getLocStart(), diag::err_atomic_builtin_pointer_size) 1049 << FirstArg->getType() << FirstArg->getSourceRange(); 1050 return ExprError(); 1051 } 1052 1053 // Each of these builtins has one pointer argument, followed by some number of 1054 // values (0, 1 or 2) followed by a potentially empty varags list of stuff 1055 // that we ignore. Find out which row of BuiltinIndices to read from as well 1056 // as the number of fixed args. 1057 unsigned BuiltinID = FDecl->getBuiltinID(); 1058 unsigned BuiltinIndex, NumFixed = 1; 1059 switch (BuiltinID) { 1060 default: llvm_unreachable("Unknown overloaded atomic builtin!"); 1061 case Builtin::BI__sync_fetch_and_add: 1062 case Builtin::BI__sync_fetch_and_add_1: 1063 case Builtin::BI__sync_fetch_and_add_2: 1064 case Builtin::BI__sync_fetch_and_add_4: 1065 case Builtin::BI__sync_fetch_and_add_8: 1066 case Builtin::BI__sync_fetch_and_add_16: 1067 BuiltinIndex = 0; 1068 break; 1069 1070 case Builtin::BI__sync_fetch_and_sub: 1071 case Builtin::BI__sync_fetch_and_sub_1: 1072 case Builtin::BI__sync_fetch_and_sub_2: 1073 case Builtin::BI__sync_fetch_and_sub_4: 1074 case Builtin::BI__sync_fetch_and_sub_8: 1075 case Builtin::BI__sync_fetch_and_sub_16: 1076 BuiltinIndex = 1; 1077 break; 1078 1079 case Builtin::BI__sync_fetch_and_or: 1080 case Builtin::BI__sync_fetch_and_or_1: 1081 case Builtin::BI__sync_fetch_and_or_2: 1082 case Builtin::BI__sync_fetch_and_or_4: 1083 case Builtin::BI__sync_fetch_and_or_8: 1084 case Builtin::BI__sync_fetch_and_or_16: 1085 BuiltinIndex = 2; 1086 break; 1087 1088 case Builtin::BI__sync_fetch_and_and: 1089 case Builtin::BI__sync_fetch_and_and_1: 1090 case Builtin::BI__sync_fetch_and_and_2: 1091 case Builtin::BI__sync_fetch_and_and_4: 1092 case Builtin::BI__sync_fetch_and_and_8: 1093 case Builtin::BI__sync_fetch_and_and_16: 1094 BuiltinIndex = 3; 1095 break; 1096 1097 case Builtin::BI__sync_fetch_and_xor: 1098 case Builtin::BI__sync_fetch_and_xor_1: 1099 case Builtin::BI__sync_fetch_and_xor_2: 1100 case Builtin::BI__sync_fetch_and_xor_4: 1101 case Builtin::BI__sync_fetch_and_xor_8: 1102 case Builtin::BI__sync_fetch_and_xor_16: 1103 BuiltinIndex = 4; 1104 break; 1105 1106 case Builtin::BI__sync_add_and_fetch: 1107 case Builtin::BI__sync_add_and_fetch_1: 1108 case Builtin::BI__sync_add_and_fetch_2: 1109 case Builtin::BI__sync_add_and_fetch_4: 1110 case Builtin::BI__sync_add_and_fetch_8: 1111 case Builtin::BI__sync_add_and_fetch_16: 1112 BuiltinIndex = 5; 1113 break; 1114 1115 case Builtin::BI__sync_sub_and_fetch: 1116 case Builtin::BI__sync_sub_and_fetch_1: 1117 case Builtin::BI__sync_sub_and_fetch_2: 1118 case Builtin::BI__sync_sub_and_fetch_4: 1119 case Builtin::BI__sync_sub_and_fetch_8: 1120 case Builtin::BI__sync_sub_and_fetch_16: 1121 BuiltinIndex = 6; 1122 break; 1123 1124 case Builtin::BI__sync_and_and_fetch: 1125 case Builtin::BI__sync_and_and_fetch_1: 1126 case Builtin::BI__sync_and_and_fetch_2: 1127 case Builtin::BI__sync_and_and_fetch_4: 1128 case Builtin::BI__sync_and_and_fetch_8: 1129 case Builtin::BI__sync_and_and_fetch_16: 1130 BuiltinIndex = 7; 1131 break; 1132 1133 case Builtin::BI__sync_or_and_fetch: 1134 case Builtin::BI__sync_or_and_fetch_1: 1135 case Builtin::BI__sync_or_and_fetch_2: 1136 case Builtin::BI__sync_or_and_fetch_4: 1137 case Builtin::BI__sync_or_and_fetch_8: 1138 case Builtin::BI__sync_or_and_fetch_16: 1139 BuiltinIndex = 8; 1140 break; 1141 1142 case Builtin::BI__sync_xor_and_fetch: 1143 case Builtin::BI__sync_xor_and_fetch_1: 1144 case Builtin::BI__sync_xor_and_fetch_2: 1145 case Builtin::BI__sync_xor_and_fetch_4: 1146 case Builtin::BI__sync_xor_and_fetch_8: 1147 case Builtin::BI__sync_xor_and_fetch_16: 1148 BuiltinIndex = 9; 1149 break; 1150 1151 case Builtin::BI__sync_val_compare_and_swap: 1152 case Builtin::BI__sync_val_compare_and_swap_1: 1153 case Builtin::BI__sync_val_compare_and_swap_2: 1154 case Builtin::BI__sync_val_compare_and_swap_4: 1155 case Builtin::BI__sync_val_compare_and_swap_8: 1156 case Builtin::BI__sync_val_compare_and_swap_16: 1157 BuiltinIndex = 10; 1158 NumFixed = 2; 1159 break; 1160 1161 case Builtin::BI__sync_bool_compare_and_swap: 1162 case Builtin::BI__sync_bool_compare_and_swap_1: 1163 case Builtin::BI__sync_bool_compare_and_swap_2: 1164 case Builtin::BI__sync_bool_compare_and_swap_4: 1165 case Builtin::BI__sync_bool_compare_and_swap_8: 1166 case Builtin::BI__sync_bool_compare_and_swap_16: 1167 BuiltinIndex = 11; 1168 NumFixed = 2; 1169 ResultType = Context.BoolTy; 1170 break; 1171 1172 case Builtin::BI__sync_lock_test_and_set: 1173 case Builtin::BI__sync_lock_test_and_set_1: 1174 case Builtin::BI__sync_lock_test_and_set_2: 1175 case Builtin::BI__sync_lock_test_and_set_4: 1176 case Builtin::BI__sync_lock_test_and_set_8: 1177 case Builtin::BI__sync_lock_test_and_set_16: 1178 BuiltinIndex = 12; 1179 break; 1180 1181 case Builtin::BI__sync_lock_release: 1182 case Builtin::BI__sync_lock_release_1: 1183 case Builtin::BI__sync_lock_release_2: 1184 case Builtin::BI__sync_lock_release_4: 1185 case Builtin::BI__sync_lock_release_8: 1186 case Builtin::BI__sync_lock_release_16: 1187 BuiltinIndex = 13; 1188 NumFixed = 0; 1189 ResultType = Context.VoidTy; 1190 break; 1191 1192 case Builtin::BI__sync_swap: 1193 case Builtin::BI__sync_swap_1: 1194 case Builtin::BI__sync_swap_2: 1195 case Builtin::BI__sync_swap_4: 1196 case Builtin::BI__sync_swap_8: 1197 case Builtin::BI__sync_swap_16: 1198 BuiltinIndex = 14; 1199 break; 1200 } 1201 1202 // Now that we know how many fixed arguments we expect, first check that we 1203 // have at least that many. 1204 if (TheCall->getNumArgs() < 1+NumFixed) { 1205 Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args_at_least) 1206 << 0 << 1+NumFixed << TheCall->getNumArgs() 1207 << TheCall->getCallee()->getSourceRange(); 1208 return ExprError(); 1209 } 1210 1211 // Get the decl for the concrete builtin from this, we can tell what the 1212 // concrete integer type we should convert to is. 1213 unsigned NewBuiltinID = BuiltinIndices[BuiltinIndex][SizeIndex]; 1214 const char *NewBuiltinName = Context.BuiltinInfo.GetName(NewBuiltinID); 1215 FunctionDecl *NewBuiltinDecl; 1216 if (NewBuiltinID == BuiltinID) 1217 NewBuiltinDecl = FDecl; 1218 else { 1219 // Perform builtin lookup to avoid redeclaring it. 1220 DeclarationName DN(&Context.Idents.get(NewBuiltinName)); 1221 LookupResult Res(*this, DN, DRE->getLocStart(), LookupOrdinaryName); 1222 LookupName(Res, TUScope, /*AllowBuiltinCreation=*/true); 1223 assert(Res.getFoundDecl()); 1224 NewBuiltinDecl = dyn_cast<FunctionDecl>(Res.getFoundDecl()); 1225 if (NewBuiltinDecl == 0) 1226 return ExprError(); 1227 } 1228 1229 // The first argument --- the pointer --- has a fixed type; we 1230 // deduce the types of the rest of the arguments accordingly. Walk 1231 // the remaining arguments, converting them to the deduced value type. 1232 for (unsigned i = 0; i != NumFixed; ++i) { 1233 ExprResult Arg = TheCall->getArg(i+1); 1234 1235 // GCC does an implicit conversion to the pointer or integer ValType. This 1236 // can fail in some cases (1i -> int**), check for this error case now. 1237 // Initialize the argument. 1238 InitializedEntity Entity = InitializedEntity::InitializeParameter(Context, 1239 ValType, /*consume*/ false); 1240 Arg = PerformCopyInitialization(Entity, SourceLocation(), Arg); 1241 if (Arg.isInvalid()) 1242 return ExprError(); 1243 1244 // Okay, we have something that *can* be converted to the right type. Check 1245 // to see if there is a potentially weird extension going on here. This can 1246 // happen when you do an atomic operation on something like an char* and 1247 // pass in 42. The 42 gets converted to char. This is even more strange 1248 // for things like 45.123 -> char, etc. 1249 // FIXME: Do this check. 1250 TheCall->setArg(i+1, Arg.take()); 1251 } 1252 1253 ASTContext& Context = this->getASTContext(); 1254 1255 // Create a new DeclRefExpr to refer to the new decl. 1256 DeclRefExpr* NewDRE = DeclRefExpr::Create( 1257 Context, 1258 DRE->getQualifierLoc(), 1259 SourceLocation(), 1260 NewBuiltinDecl, 1261 /*enclosing*/ false, 1262 DRE->getLocation(), 1263 Context.BuiltinFnTy, 1264 DRE->getValueKind()); 1265 1266 // Set the callee in the CallExpr. 1267 // FIXME: This loses syntactic information. 1268 QualType CalleePtrTy = Context.getPointerType(NewBuiltinDecl->getType()); 1269 ExprResult PromotedCall = ImpCastExprToType(NewDRE, CalleePtrTy, 1270 CK_BuiltinFnToFnPtr); 1271 TheCall->setCallee(PromotedCall.take()); 1272 1273 // Change the result type of the call to match the original value type. This 1274 // is arbitrary, but the codegen for these builtins ins design to handle it 1275 // gracefully. 1276 TheCall->setType(ResultType); 1277 1278 return TheCallResult; 1279} 1280 1281/// CheckObjCString - Checks that the argument to the builtin 1282/// CFString constructor is correct 1283/// Note: It might also make sense to do the UTF-16 conversion here (would 1284/// simplify the backend). 1285bool Sema::CheckObjCString(Expr *Arg) { 1286 Arg = Arg->IgnoreParenCasts(); 1287 StringLiteral *Literal = dyn_cast<StringLiteral>(Arg); 1288 1289 if (!Literal || !Literal->isAscii()) { 1290 Diag(Arg->getLocStart(), diag::err_cfstring_literal_not_string_constant) 1291 << Arg->getSourceRange(); 1292 return true; 1293 } 1294 1295 if (Literal->containsNonAsciiOrNull()) { 1296 StringRef String = Literal->getString(); 1297 unsigned NumBytes = String.size(); 1298 SmallVector<UTF16, 128> ToBuf(NumBytes); 1299 const UTF8 *FromPtr = (const UTF8 *)String.data(); 1300 UTF16 *ToPtr = &ToBuf[0]; 1301 1302 ConversionResult Result = ConvertUTF8toUTF16(&FromPtr, FromPtr + NumBytes, 1303 &ToPtr, ToPtr + NumBytes, 1304 strictConversion); 1305 // Check for conversion failure. 1306 if (Result != conversionOK) 1307 Diag(Arg->getLocStart(), 1308 diag::warn_cfstring_truncated) << Arg->getSourceRange(); 1309 } 1310 return false; 1311} 1312 1313/// SemaBuiltinVAStart - Check the arguments to __builtin_va_start for validity. 1314/// Emit an error and return true on failure, return false on success. 1315bool Sema::SemaBuiltinVAStart(CallExpr *TheCall) { 1316 Expr *Fn = TheCall->getCallee(); 1317 if (TheCall->getNumArgs() > 2) { 1318 Diag(TheCall->getArg(2)->getLocStart(), 1319 diag::err_typecheck_call_too_many_args) 1320 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 1321 << Fn->getSourceRange() 1322 << SourceRange(TheCall->getArg(2)->getLocStart(), 1323 (*(TheCall->arg_end()-1))->getLocEnd()); 1324 return true; 1325 } 1326 1327 if (TheCall->getNumArgs() < 2) { 1328 return Diag(TheCall->getLocEnd(), 1329 diag::err_typecheck_call_too_few_args_at_least) 1330 << 0 /*function call*/ << 2 << TheCall->getNumArgs(); 1331 } 1332 1333 // Type-check the first argument normally. 1334 if (checkBuiltinArgument(*this, TheCall, 0)) 1335 return true; 1336 1337 // Determine whether the current function is variadic or not. 1338 BlockScopeInfo *CurBlock = getCurBlock(); 1339 bool isVariadic; 1340 if (CurBlock) 1341 isVariadic = CurBlock->TheDecl->isVariadic(); 1342 else if (FunctionDecl *FD = getCurFunctionDecl()) 1343 isVariadic = FD->isVariadic(); 1344 else 1345 isVariadic = getCurMethodDecl()->isVariadic(); 1346 1347 if (!isVariadic) { 1348 Diag(Fn->getLocStart(), diag::err_va_start_used_in_non_variadic_function); 1349 return true; 1350 } 1351 1352 // Verify that the second argument to the builtin is the last argument of the 1353 // current function or method. 1354 bool SecondArgIsLastNamedArgument = false; 1355 const Expr *Arg = TheCall->getArg(1)->IgnoreParenCasts(); 1356 1357 if (const DeclRefExpr *DR = dyn_cast<DeclRefExpr>(Arg)) { 1358 if (const ParmVarDecl *PV = dyn_cast<ParmVarDecl>(DR->getDecl())) { 1359 // FIXME: This isn't correct for methods (results in bogus warning). 1360 // Get the last formal in the current function. 1361 const ParmVarDecl *LastArg; 1362 if (CurBlock) 1363 LastArg = *(CurBlock->TheDecl->param_end()-1); 1364 else if (FunctionDecl *FD = getCurFunctionDecl()) 1365 LastArg = *(FD->param_end()-1); 1366 else 1367 LastArg = *(getCurMethodDecl()->param_end()-1); 1368 SecondArgIsLastNamedArgument = PV == LastArg; 1369 } 1370 } 1371 1372 if (!SecondArgIsLastNamedArgument) 1373 Diag(TheCall->getArg(1)->getLocStart(), 1374 diag::warn_second_parameter_of_va_start_not_last_named_argument); 1375 return false; 1376} 1377 1378/// SemaBuiltinUnorderedCompare - Handle functions like __builtin_isgreater and 1379/// friends. This is declared to take (...), so we have to check everything. 1380bool Sema::SemaBuiltinUnorderedCompare(CallExpr *TheCall) { 1381 if (TheCall->getNumArgs() < 2) 1382 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 1383 << 0 << 2 << TheCall->getNumArgs()/*function call*/; 1384 if (TheCall->getNumArgs() > 2) 1385 return Diag(TheCall->getArg(2)->getLocStart(), 1386 diag::err_typecheck_call_too_many_args) 1387 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 1388 << SourceRange(TheCall->getArg(2)->getLocStart(), 1389 (*(TheCall->arg_end()-1))->getLocEnd()); 1390 1391 ExprResult OrigArg0 = TheCall->getArg(0); 1392 ExprResult OrigArg1 = TheCall->getArg(1); 1393 1394 // Do standard promotions between the two arguments, returning their common 1395 // type. 1396 QualType Res = UsualArithmeticConversions(OrigArg0, OrigArg1, false); 1397 if (OrigArg0.isInvalid() || OrigArg1.isInvalid()) 1398 return true; 1399 1400 // Make sure any conversions are pushed back into the call; this is 1401 // type safe since unordered compare builtins are declared as "_Bool 1402 // foo(...)". 1403 TheCall->setArg(0, OrigArg0.get()); 1404 TheCall->setArg(1, OrigArg1.get()); 1405 1406 if (OrigArg0.get()->isTypeDependent() || OrigArg1.get()->isTypeDependent()) 1407 return false; 1408 1409 // If the common type isn't a real floating type, then the arguments were 1410 // invalid for this operation. 1411 if (Res.isNull() || !Res->isRealFloatingType()) 1412 return Diag(OrigArg0.get()->getLocStart(), 1413 diag::err_typecheck_call_invalid_ordered_compare) 1414 << OrigArg0.get()->getType() << OrigArg1.get()->getType() 1415 << SourceRange(OrigArg0.get()->getLocStart(), OrigArg1.get()->getLocEnd()); 1416 1417 return false; 1418} 1419 1420/// SemaBuiltinSemaBuiltinFPClassification - Handle functions like 1421/// __builtin_isnan and friends. This is declared to take (...), so we have 1422/// to check everything. We expect the last argument to be a floating point 1423/// value. 1424bool Sema::SemaBuiltinFPClassification(CallExpr *TheCall, unsigned NumArgs) { 1425 if (TheCall->getNumArgs() < NumArgs) 1426 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 1427 << 0 << NumArgs << TheCall->getNumArgs()/*function call*/; 1428 if (TheCall->getNumArgs() > NumArgs) 1429 return Diag(TheCall->getArg(NumArgs)->getLocStart(), 1430 diag::err_typecheck_call_too_many_args) 1431 << 0 /*function call*/ << NumArgs << TheCall->getNumArgs() 1432 << SourceRange(TheCall->getArg(NumArgs)->getLocStart(), 1433 (*(TheCall->arg_end()-1))->getLocEnd()); 1434 1435 Expr *OrigArg = TheCall->getArg(NumArgs-1); 1436 1437 if (OrigArg->isTypeDependent()) 1438 return false; 1439 1440 // This operation requires a non-_Complex floating-point number. 1441 if (!OrigArg->getType()->isRealFloatingType()) 1442 return Diag(OrigArg->getLocStart(), 1443 diag::err_typecheck_call_invalid_unary_fp) 1444 << OrigArg->getType() << OrigArg->getSourceRange(); 1445 1446 // If this is an implicit conversion from float -> double, remove it. 1447 if (ImplicitCastExpr *Cast = dyn_cast<ImplicitCastExpr>(OrigArg)) { 1448 Expr *CastArg = Cast->getSubExpr(); 1449 if (CastArg->getType()->isSpecificBuiltinType(BuiltinType::Float)) { 1450 assert(Cast->getType()->isSpecificBuiltinType(BuiltinType::Double) && 1451 "promotion from float to double is the only expected cast here"); 1452 Cast->setSubExpr(0); 1453 TheCall->setArg(NumArgs-1, CastArg); 1454 } 1455 } 1456 1457 return false; 1458} 1459 1460/// SemaBuiltinShuffleVector - Handle __builtin_shufflevector. 1461// This is declared to take (...), so we have to check everything. 1462ExprResult Sema::SemaBuiltinShuffleVector(CallExpr *TheCall) { 1463 if (TheCall->getNumArgs() < 2) 1464 return ExprError(Diag(TheCall->getLocEnd(), 1465 diag::err_typecheck_call_too_few_args_at_least) 1466 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 1467 << TheCall->getSourceRange()); 1468 1469 // Determine which of the following types of shufflevector we're checking: 1470 // 1) unary, vector mask: (lhs, mask) 1471 // 2) binary, vector mask: (lhs, rhs, mask) 1472 // 3) binary, scalar mask: (lhs, rhs, index, ..., index) 1473 QualType resType = TheCall->getArg(0)->getType(); 1474 unsigned numElements = 0; 1475 1476 if (!TheCall->getArg(0)->isTypeDependent() && 1477 !TheCall->getArg(1)->isTypeDependent()) { 1478 QualType LHSType = TheCall->getArg(0)->getType(); 1479 QualType RHSType = TheCall->getArg(1)->getType(); 1480 1481 if (!LHSType->isVectorType() || !RHSType->isVectorType()) { 1482 Diag(TheCall->getLocStart(), diag::err_shufflevector_non_vector) 1483 << SourceRange(TheCall->getArg(0)->getLocStart(), 1484 TheCall->getArg(1)->getLocEnd()); 1485 return ExprError(); 1486 } 1487 1488 numElements = LHSType->getAs<VectorType>()->getNumElements(); 1489 unsigned numResElements = TheCall->getNumArgs() - 2; 1490 1491 // Check to see if we have a call with 2 vector arguments, the unary shuffle 1492 // with mask. If so, verify that RHS is an integer vector type with the 1493 // same number of elts as lhs. 1494 if (TheCall->getNumArgs() == 2) { 1495 if (!RHSType->hasIntegerRepresentation() || 1496 RHSType->getAs<VectorType>()->getNumElements() != numElements) 1497 Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector) 1498 << SourceRange(TheCall->getArg(1)->getLocStart(), 1499 TheCall->getArg(1)->getLocEnd()); 1500 numResElements = numElements; 1501 } 1502 else if (!Context.hasSameUnqualifiedType(LHSType, RHSType)) { 1503 Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector) 1504 << SourceRange(TheCall->getArg(0)->getLocStart(), 1505 TheCall->getArg(1)->getLocEnd()); 1506 return ExprError(); 1507 } else if (numElements != numResElements) { 1508 QualType eltType = LHSType->getAs<VectorType>()->getElementType(); 1509 resType = Context.getVectorType(eltType, numResElements, 1510 VectorType::GenericVector); 1511 } 1512 } 1513 1514 for (unsigned i = 2; i < TheCall->getNumArgs(); i++) { 1515 if (TheCall->getArg(i)->isTypeDependent() || 1516 TheCall->getArg(i)->isValueDependent()) 1517 continue; 1518 1519 llvm::APSInt Result(32); 1520 if (!TheCall->getArg(i)->isIntegerConstantExpr(Result, Context)) 1521 return ExprError(Diag(TheCall->getLocStart(), 1522 diag::err_shufflevector_nonconstant_argument) 1523 << TheCall->getArg(i)->getSourceRange()); 1524 1525 if (Result.getActiveBits() > 64 || Result.getZExtValue() >= numElements*2) 1526 return ExprError(Diag(TheCall->getLocStart(), 1527 diag::err_shufflevector_argument_too_large) 1528 << TheCall->getArg(i)->getSourceRange()); 1529 } 1530 1531 SmallVector<Expr*, 32> exprs; 1532 1533 for (unsigned i = 0, e = TheCall->getNumArgs(); i != e; i++) { 1534 exprs.push_back(TheCall->getArg(i)); 1535 TheCall->setArg(i, 0); 1536 } 1537 1538 return Owned(new (Context) ShuffleVectorExpr(Context, exprs, resType, 1539 TheCall->getCallee()->getLocStart(), 1540 TheCall->getRParenLoc())); 1541} 1542 1543/// SemaBuiltinPrefetch - Handle __builtin_prefetch. 1544// This is declared to take (const void*, ...) and can take two 1545// optional constant int args. 1546bool Sema::SemaBuiltinPrefetch(CallExpr *TheCall) { 1547 unsigned NumArgs = TheCall->getNumArgs(); 1548 1549 if (NumArgs > 3) 1550 return Diag(TheCall->getLocEnd(), 1551 diag::err_typecheck_call_too_many_args_at_most) 1552 << 0 /*function call*/ << 3 << NumArgs 1553 << TheCall->getSourceRange(); 1554 1555 // Argument 0 is checked for us and the remaining arguments must be 1556 // constant integers. 1557 for (unsigned i = 1; i != NumArgs; ++i) { 1558 Expr *Arg = TheCall->getArg(i); 1559 1560 // We can't check the value of a dependent argument. 1561 if (Arg->isTypeDependent() || Arg->isValueDependent()) 1562 continue; 1563 1564 llvm::APSInt Result; 1565 if (SemaBuiltinConstantArg(TheCall, i, Result)) 1566 return true; 1567 1568 // FIXME: gcc issues a warning and rewrites these to 0. These 1569 // seems especially odd for the third argument since the default 1570 // is 3. 1571 if (i == 1) { 1572 if (Result.getLimitedValue() > 1) 1573 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 1574 << "0" << "1" << Arg->getSourceRange(); 1575 } else { 1576 if (Result.getLimitedValue() > 3) 1577 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 1578 << "0" << "3" << Arg->getSourceRange(); 1579 } 1580 } 1581 1582 return false; 1583} 1584 1585/// SemaBuiltinConstantArg - Handle a check if argument ArgNum of CallExpr 1586/// TheCall is a constant expression. 1587bool Sema::SemaBuiltinConstantArg(CallExpr *TheCall, int ArgNum, 1588 llvm::APSInt &Result) { 1589 Expr *Arg = TheCall->getArg(ArgNum); 1590 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 1591 FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl()); 1592 1593 if (Arg->isTypeDependent() || Arg->isValueDependent()) return false; 1594 1595 if (!Arg->isIntegerConstantExpr(Result, Context)) 1596 return Diag(TheCall->getLocStart(), diag::err_constant_integer_arg_type) 1597 << FDecl->getDeclName() << Arg->getSourceRange(); 1598 1599 return false; 1600} 1601 1602/// SemaBuiltinObjectSize - Handle __builtin_object_size(void *ptr, 1603/// int type). This simply type checks that type is one of the defined 1604/// constants (0-3). 1605// For compatibility check 0-3, llvm only handles 0 and 2. 1606bool Sema::SemaBuiltinObjectSize(CallExpr *TheCall) { 1607 llvm::APSInt Result; 1608 1609 // We can't check the value of a dependent argument. 1610 if (TheCall->getArg(1)->isTypeDependent() || 1611 TheCall->getArg(1)->isValueDependent()) 1612 return false; 1613 1614 // Check constant-ness first. 1615 if (SemaBuiltinConstantArg(TheCall, 1, Result)) 1616 return true; 1617 1618 Expr *Arg = TheCall->getArg(1); 1619 if (Result.getSExtValue() < 0 || Result.getSExtValue() > 3) { 1620 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 1621 << "0" << "3" << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 1622 } 1623 1624 return false; 1625} 1626 1627/// SemaBuiltinLongjmp - Handle __builtin_longjmp(void *env[5], int val). 1628/// This checks that val is a constant 1. 1629bool Sema::SemaBuiltinLongjmp(CallExpr *TheCall) { 1630 Expr *Arg = TheCall->getArg(1); 1631 llvm::APSInt Result; 1632 1633 // TODO: This is less than ideal. Overload this to take a value. 1634 if (SemaBuiltinConstantArg(TheCall, 1, Result)) 1635 return true; 1636 1637 if (Result != 1) 1638 return Diag(TheCall->getLocStart(), diag::err_builtin_longjmp_invalid_val) 1639 << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 1640 1641 return false; 1642} 1643 1644// Determine if an expression is a string literal or constant string. 1645// If this function returns false on the arguments to a function expecting a 1646// format string, we will usually need to emit a warning. 1647// True string literals are then checked by CheckFormatString. 1648Sema::StringLiteralCheckType 1649Sema::checkFormatStringExpr(const Expr *E, Expr **Args, 1650 unsigned NumArgs, bool HasVAListArg, 1651 unsigned format_idx, unsigned firstDataArg, 1652 FormatStringType Type, VariadicCallType CallType, 1653 bool inFunctionCall) { 1654 tryAgain: 1655 if (E->isTypeDependent() || E->isValueDependent()) 1656 return SLCT_NotALiteral; 1657 1658 E = E->IgnoreParenCasts(); 1659 1660 if (E->isNullPointerConstant(Context, Expr::NPC_ValueDependentIsNotNull)) 1661 // Technically -Wformat-nonliteral does not warn about this case. 1662 // The behavior of printf and friends in this case is implementation 1663 // dependent. Ideally if the format string cannot be null then 1664 // it should have a 'nonnull' attribute in the function prototype. 1665 return SLCT_CheckedLiteral; 1666 1667 switch (E->getStmtClass()) { 1668 case Stmt::BinaryConditionalOperatorClass: 1669 case Stmt::ConditionalOperatorClass: { 1670 // The expression is a literal if both sub-expressions were, and it was 1671 // completely checked only if both sub-expressions were checked. 1672 const AbstractConditionalOperator *C = 1673 cast<AbstractConditionalOperator>(E); 1674 StringLiteralCheckType Left = 1675 checkFormatStringExpr(C->getTrueExpr(), Args, NumArgs, 1676 HasVAListArg, format_idx, firstDataArg, 1677 Type, CallType, inFunctionCall); 1678 if (Left == SLCT_NotALiteral) 1679 return SLCT_NotALiteral; 1680 StringLiteralCheckType Right = 1681 checkFormatStringExpr(C->getFalseExpr(), Args, NumArgs, 1682 HasVAListArg, format_idx, firstDataArg, 1683 Type, CallType, inFunctionCall); 1684 return Left < Right ? Left : Right; 1685 } 1686 1687 case Stmt::ImplicitCastExprClass: { 1688 E = cast<ImplicitCastExpr>(E)->getSubExpr(); 1689 goto tryAgain; 1690 } 1691 1692 case Stmt::OpaqueValueExprClass: 1693 if (const Expr *src = cast<OpaqueValueExpr>(E)->getSourceExpr()) { 1694 E = src; 1695 goto tryAgain; 1696 } 1697 return SLCT_NotALiteral; 1698 1699 case Stmt::PredefinedExprClass: 1700 // While __func__, etc., are technically not string literals, they 1701 // cannot contain format specifiers and thus are not a security 1702 // liability. 1703 return SLCT_UncheckedLiteral; 1704 1705 case Stmt::DeclRefExprClass: { 1706 const DeclRefExpr *DR = cast<DeclRefExpr>(E); 1707 1708 // As an exception, do not flag errors for variables binding to 1709 // const string literals. 1710 if (const VarDecl *VD = dyn_cast<VarDecl>(DR->getDecl())) { 1711 bool isConstant = false; 1712 QualType T = DR->getType(); 1713 1714 if (const ArrayType *AT = Context.getAsArrayType(T)) { 1715 isConstant = AT->getElementType().isConstant(Context); 1716 } else if (const PointerType *PT = T->getAs<PointerType>()) { 1717 isConstant = T.isConstant(Context) && 1718 PT->getPointeeType().isConstant(Context); 1719 } else if (T->isObjCObjectPointerType()) { 1720 // In ObjC, there is usually no "const ObjectPointer" type, 1721 // so don't check if the pointee type is constant. 1722 isConstant = T.isConstant(Context); 1723 } 1724 1725 if (isConstant) { 1726 if (const Expr *Init = VD->getAnyInitializer()) { 1727 // Look through initializers like const char c[] = { "foo" } 1728 if (const InitListExpr *InitList = dyn_cast<InitListExpr>(Init)) { 1729 if (InitList->isStringLiteralInit()) 1730 Init = InitList->getInit(0)->IgnoreParenImpCasts(); 1731 } 1732 return checkFormatStringExpr(Init, Args, NumArgs, 1733 HasVAListArg, format_idx, 1734 firstDataArg, Type, CallType, 1735 /*inFunctionCall*/false); 1736 } 1737 } 1738 1739 // For vprintf* functions (i.e., HasVAListArg==true), we add a 1740 // special check to see if the format string is a function parameter 1741 // of the function calling the printf function. If the function 1742 // has an attribute indicating it is a printf-like function, then we 1743 // should suppress warnings concerning non-literals being used in a call 1744 // to a vprintf function. For example: 1745 // 1746 // void 1747 // logmessage(char const *fmt __attribute__ (format (printf, 1, 2)), ...){ 1748 // va_list ap; 1749 // va_start(ap, fmt); 1750 // vprintf(fmt, ap); // Do NOT emit a warning about "fmt". 1751 // ... 1752 // 1753 if (HasVAListArg) { 1754 if (const ParmVarDecl *PV = dyn_cast<ParmVarDecl>(VD)) { 1755 if (const NamedDecl *ND = dyn_cast<NamedDecl>(PV->getDeclContext())) { 1756 int PVIndex = PV->getFunctionScopeIndex() + 1; 1757 for (specific_attr_iterator<FormatAttr> 1758 i = ND->specific_attr_begin<FormatAttr>(), 1759 e = ND->specific_attr_end<FormatAttr>(); i != e ; ++i) { 1760 FormatAttr *PVFormat = *i; 1761 // adjust for implicit parameter 1762 if (const CXXMethodDecl *MD = dyn_cast<CXXMethodDecl>(ND)) 1763 if (MD->isInstance()) 1764 ++PVIndex; 1765 // We also check if the formats are compatible. 1766 // We can't pass a 'scanf' string to a 'printf' function. 1767 if (PVIndex == PVFormat->getFormatIdx() && 1768 Type == GetFormatStringType(PVFormat)) 1769 return SLCT_UncheckedLiteral; 1770 } 1771 } 1772 } 1773 } 1774 } 1775 1776 return SLCT_NotALiteral; 1777 } 1778 1779 case Stmt::CallExprClass: 1780 case Stmt::CXXMemberCallExprClass: { 1781 const CallExpr *CE = cast<CallExpr>(E); 1782 if (const NamedDecl *ND = dyn_cast_or_null<NamedDecl>(CE->getCalleeDecl())) { 1783 if (const FormatArgAttr *FA = ND->getAttr<FormatArgAttr>()) { 1784 unsigned ArgIndex = FA->getFormatIdx(); 1785 if (const CXXMethodDecl *MD = dyn_cast<CXXMethodDecl>(ND)) 1786 if (MD->isInstance()) 1787 --ArgIndex; 1788 const Expr *Arg = CE->getArg(ArgIndex - 1); 1789 1790 return checkFormatStringExpr(Arg, Args, NumArgs, 1791 HasVAListArg, format_idx, firstDataArg, 1792 Type, CallType, inFunctionCall); 1793 } else if (const FunctionDecl *FD = dyn_cast<FunctionDecl>(ND)) { 1794 unsigned BuiltinID = FD->getBuiltinID(); 1795 if (BuiltinID == Builtin::BI__builtin___CFStringMakeConstantString || 1796 BuiltinID == Builtin::BI__builtin___NSStringMakeConstantString) { 1797 const Expr *Arg = CE->getArg(0); 1798 return checkFormatStringExpr(Arg, Args, NumArgs, 1799 HasVAListArg, format_idx, 1800 firstDataArg, Type, CallType, 1801 inFunctionCall); 1802 } 1803 } 1804 } 1805 1806 return SLCT_NotALiteral; 1807 } 1808 case Stmt::ObjCStringLiteralClass: 1809 case Stmt::StringLiteralClass: { 1810 const StringLiteral *StrE = NULL; 1811 1812 if (const ObjCStringLiteral *ObjCFExpr = dyn_cast<ObjCStringLiteral>(E)) 1813 StrE = ObjCFExpr->getString(); 1814 else 1815 StrE = cast<StringLiteral>(E); 1816 1817 if (StrE) { 1818 CheckFormatString(StrE, E, Args, NumArgs, HasVAListArg, format_idx, 1819 firstDataArg, Type, inFunctionCall, CallType); 1820 return SLCT_CheckedLiteral; 1821 } 1822 1823 return SLCT_NotALiteral; 1824 } 1825 1826 default: 1827 return SLCT_NotALiteral; 1828 } 1829} 1830 1831void 1832Sema::CheckNonNullArguments(const NonNullAttr *NonNull, 1833 const Expr * const *ExprArgs, 1834 SourceLocation CallSiteLoc) { 1835 for (NonNullAttr::args_iterator i = NonNull->args_begin(), 1836 e = NonNull->args_end(); 1837 i != e; ++i) { 1838 const Expr *ArgExpr = ExprArgs[*i]; 1839 if (ArgExpr->isNullPointerConstant(Context, 1840 Expr::NPC_ValueDependentIsNotNull)) 1841 Diag(CallSiteLoc, diag::warn_null_arg) << ArgExpr->getSourceRange(); 1842 } 1843} 1844 1845Sema::FormatStringType Sema::GetFormatStringType(const FormatAttr *Format) { 1846 return llvm::StringSwitch<FormatStringType>(Format->getType()) 1847 .Case("scanf", FST_Scanf) 1848 .Cases("printf", "printf0", FST_Printf) 1849 .Cases("NSString", "CFString", FST_NSString) 1850 .Case("strftime", FST_Strftime) 1851 .Case("strfmon", FST_Strfmon) 1852 .Cases("kprintf", "cmn_err", "vcmn_err", "zcmn_err", FST_Kprintf) 1853 .Default(FST_Unknown); 1854} 1855 1856/// CheckFormatArguments - Check calls to printf and scanf (and similar 1857/// functions) for correct use of format strings. 1858/// Returns true if a format string has been fully checked. 1859bool Sema::CheckFormatArguments(const FormatAttr *Format, Expr **Args, 1860 unsigned NumArgs, bool IsCXXMember, 1861 VariadicCallType CallType, 1862 SourceLocation Loc, SourceRange Range) { 1863 FormatStringInfo FSI; 1864 if (getFormatStringInfo(Format, IsCXXMember, &FSI)) 1865 return CheckFormatArguments(Args, NumArgs, FSI.HasVAListArg, FSI.FormatIdx, 1866 FSI.FirstDataArg, GetFormatStringType(Format), 1867 CallType, Loc, Range); 1868 return false; 1869} 1870 1871bool Sema::CheckFormatArguments(Expr **Args, unsigned NumArgs, 1872 bool HasVAListArg, unsigned format_idx, 1873 unsigned firstDataArg, FormatStringType Type, 1874 VariadicCallType CallType, 1875 SourceLocation Loc, SourceRange Range) { 1876 // CHECK: printf/scanf-like function is called with no format string. 1877 if (format_idx >= NumArgs) { 1878 Diag(Loc, diag::warn_missing_format_string) << Range; 1879 return false; 1880 } 1881 1882 const Expr *OrigFormatExpr = Args[format_idx]->IgnoreParenCasts(); 1883 1884 // CHECK: format string is not a string literal. 1885 // 1886 // Dynamically generated format strings are difficult to 1887 // automatically vet at compile time. Requiring that format strings 1888 // are string literals: (1) permits the checking of format strings by 1889 // the compiler and thereby (2) can practically remove the source of 1890 // many format string exploits. 1891 1892 // Format string can be either ObjC string (e.g. @"%d") or 1893 // C string (e.g. "%d") 1894 // ObjC string uses the same format specifiers as C string, so we can use 1895 // the same format string checking logic for both ObjC and C strings. 1896 StringLiteralCheckType CT = 1897 checkFormatStringExpr(OrigFormatExpr, Args, NumArgs, HasVAListArg, 1898 format_idx, firstDataArg, Type, CallType); 1899 if (CT != SLCT_NotALiteral) 1900 // Literal format string found, check done! 1901 return CT == SLCT_CheckedLiteral; 1902 1903 // Strftime is particular as it always uses a single 'time' argument, 1904 // so it is safe to pass a non-literal string. 1905 if (Type == FST_Strftime) 1906 return false; 1907 1908 // Do not emit diag when the string param is a macro expansion and the 1909 // format is either NSString or CFString. This is a hack to prevent 1910 // diag when using the NSLocalizedString and CFCopyLocalizedString macros 1911 // which are usually used in place of NS and CF string literals. 1912 if (Type == FST_NSString && 1913 SourceMgr.isInSystemMacro(Args[format_idx]->getLocStart())) 1914 return false; 1915 1916 // If there are no arguments specified, warn with -Wformat-security, otherwise 1917 // warn only with -Wformat-nonliteral. 1918 if (NumArgs == format_idx+1) 1919 Diag(Args[format_idx]->getLocStart(), 1920 diag::warn_format_nonliteral_noargs) 1921 << OrigFormatExpr->getSourceRange(); 1922 else 1923 Diag(Args[format_idx]->getLocStart(), 1924 diag::warn_format_nonliteral) 1925 << OrigFormatExpr->getSourceRange(); 1926 return false; 1927} 1928 1929namespace { 1930class CheckFormatHandler : public analyze_format_string::FormatStringHandler { 1931protected: 1932 Sema &S; 1933 const StringLiteral *FExpr; 1934 const Expr *OrigFormatExpr; 1935 const unsigned FirstDataArg; 1936 const unsigned NumDataArgs; 1937 const char *Beg; // Start of format string. 1938 const bool HasVAListArg; 1939 const Expr * const *Args; 1940 const unsigned NumArgs; 1941 unsigned FormatIdx; 1942 llvm::BitVector CoveredArgs; 1943 bool usesPositionalArgs; 1944 bool atFirstArg; 1945 bool inFunctionCall; 1946 Sema::VariadicCallType CallType; 1947public: 1948 CheckFormatHandler(Sema &s, const StringLiteral *fexpr, 1949 const Expr *origFormatExpr, unsigned firstDataArg, 1950 unsigned numDataArgs, const char *beg, bool hasVAListArg, 1951 Expr **args, unsigned numArgs, 1952 unsigned formatIdx, bool inFunctionCall, 1953 Sema::VariadicCallType callType) 1954 : S(s), FExpr(fexpr), OrigFormatExpr(origFormatExpr), 1955 FirstDataArg(firstDataArg), NumDataArgs(numDataArgs), 1956 Beg(beg), HasVAListArg(hasVAListArg), 1957 Args(args), NumArgs(numArgs), FormatIdx(formatIdx), 1958 usesPositionalArgs(false), atFirstArg(true), 1959 inFunctionCall(inFunctionCall), CallType(callType) { 1960 CoveredArgs.resize(numDataArgs); 1961 CoveredArgs.reset(); 1962 } 1963 1964 void DoneProcessing(); 1965 1966 void HandleIncompleteSpecifier(const char *startSpecifier, 1967 unsigned specifierLen); 1968 1969 void HandleInvalidLengthModifier( 1970 const analyze_format_string::FormatSpecifier &FS, 1971 const analyze_format_string::ConversionSpecifier &CS, 1972 const char *startSpecifier, unsigned specifierLen, unsigned DiagID); 1973 1974 void HandleNonStandardLengthModifier( 1975 const analyze_format_string::FormatSpecifier &FS, 1976 const char *startSpecifier, unsigned specifierLen); 1977 1978 void HandleNonStandardConversionSpecifier( 1979 const analyze_format_string::ConversionSpecifier &CS, 1980 const char *startSpecifier, unsigned specifierLen); 1981 1982 virtual void HandlePosition(const char *startPos, unsigned posLen); 1983 1984 virtual void HandleInvalidPosition(const char *startSpecifier, 1985 unsigned specifierLen, 1986 analyze_format_string::PositionContext p); 1987 1988 virtual void HandleZeroPosition(const char *startPos, unsigned posLen); 1989 1990 void HandleNullChar(const char *nullCharacter); 1991 1992 template <typename Range> 1993 static void EmitFormatDiagnostic(Sema &S, bool inFunctionCall, 1994 const Expr *ArgumentExpr, 1995 PartialDiagnostic PDiag, 1996 SourceLocation StringLoc, 1997 bool IsStringLocation, Range StringRange, 1998 ArrayRef<FixItHint> Fixit = ArrayRef<FixItHint>()); 1999 2000protected: 2001 bool HandleInvalidConversionSpecifier(unsigned argIndex, SourceLocation Loc, 2002 const char *startSpec, 2003 unsigned specifierLen, 2004 const char *csStart, unsigned csLen); 2005 2006 void HandlePositionalNonpositionalArgs(SourceLocation Loc, 2007 const char *startSpec, 2008 unsigned specifierLen); 2009 2010 SourceRange getFormatStringRange(); 2011 CharSourceRange getSpecifierRange(const char *startSpecifier, 2012 unsigned specifierLen); 2013 SourceLocation getLocationOfByte(const char *x); 2014 2015 const Expr *getDataArg(unsigned i) const; 2016 2017 bool CheckNumArgs(const analyze_format_string::FormatSpecifier &FS, 2018 const analyze_format_string::ConversionSpecifier &CS, 2019 const char *startSpecifier, unsigned specifierLen, 2020 unsigned argIndex); 2021 2022 template <typename Range> 2023 void EmitFormatDiagnostic(PartialDiagnostic PDiag, SourceLocation StringLoc, 2024 bool IsStringLocation, Range StringRange, 2025 ArrayRef<FixItHint> Fixit = ArrayRef<FixItHint>()); 2026 2027 void CheckPositionalAndNonpositionalArgs( 2028 const analyze_format_string::FormatSpecifier *FS); 2029}; 2030} 2031 2032SourceRange CheckFormatHandler::getFormatStringRange() { 2033 return OrigFormatExpr->getSourceRange(); 2034} 2035 2036CharSourceRange CheckFormatHandler:: 2037getSpecifierRange(const char *startSpecifier, unsigned specifierLen) { 2038 SourceLocation Start = getLocationOfByte(startSpecifier); 2039 SourceLocation End = getLocationOfByte(startSpecifier + specifierLen - 1); 2040 2041 // Advance the end SourceLocation by one due to half-open ranges. 2042 End = End.getLocWithOffset(1); 2043 2044 return CharSourceRange::getCharRange(Start, End); 2045} 2046 2047SourceLocation CheckFormatHandler::getLocationOfByte(const char *x) { 2048 return S.getLocationOfStringLiteralByte(FExpr, x - Beg); 2049} 2050 2051void CheckFormatHandler::HandleIncompleteSpecifier(const char *startSpecifier, 2052 unsigned specifierLen){ 2053 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_incomplete_specifier), 2054 getLocationOfByte(startSpecifier), 2055 /*IsStringLocation*/true, 2056 getSpecifierRange(startSpecifier, specifierLen)); 2057} 2058 2059void CheckFormatHandler::HandleInvalidLengthModifier( 2060 const analyze_format_string::FormatSpecifier &FS, 2061 const analyze_format_string::ConversionSpecifier &CS, 2062 const char *startSpecifier, unsigned specifierLen, unsigned DiagID) { 2063 using namespace analyze_format_string; 2064 2065 const LengthModifier &LM = FS.getLengthModifier(); 2066 CharSourceRange LMRange = getSpecifierRange(LM.getStart(), LM.getLength()); 2067 2068 // See if we know how to fix this length modifier. 2069 llvm::Optional<LengthModifier> FixedLM = FS.getCorrectedLengthModifier(); 2070 if (FixedLM) { 2071 EmitFormatDiagnostic(S.PDiag(DiagID) << LM.toString() << CS.toString(), 2072 getLocationOfByte(LM.getStart()), 2073 /*IsStringLocation*/true, 2074 getSpecifierRange(startSpecifier, specifierLen)); 2075 2076 S.Diag(getLocationOfByte(LM.getStart()), diag::note_format_fix_specifier) 2077 << FixedLM->toString() 2078 << FixItHint::CreateReplacement(LMRange, FixedLM->toString()); 2079 2080 } else { 2081 FixItHint Hint; 2082 if (DiagID == diag::warn_format_nonsensical_length) 2083 Hint = FixItHint::CreateRemoval(LMRange); 2084 2085 EmitFormatDiagnostic(S.PDiag(DiagID) << LM.toString() << CS.toString(), 2086 getLocationOfByte(LM.getStart()), 2087 /*IsStringLocation*/true, 2088 getSpecifierRange(startSpecifier, specifierLen), 2089 Hint); 2090 } 2091} 2092 2093void CheckFormatHandler::HandleNonStandardLengthModifier( 2094 const analyze_format_string::FormatSpecifier &FS, 2095 const char *startSpecifier, unsigned specifierLen) { 2096 using namespace analyze_format_string; 2097 2098 const LengthModifier &LM = FS.getLengthModifier(); 2099 CharSourceRange LMRange = getSpecifierRange(LM.getStart(), LM.getLength()); 2100 2101 // See if we know how to fix this length modifier. 2102 llvm::Optional<LengthModifier> FixedLM = FS.getCorrectedLengthModifier(); 2103 if (FixedLM) { 2104 EmitFormatDiagnostic(S.PDiag(diag::warn_format_non_standard) 2105 << LM.toString() << 0, 2106 getLocationOfByte(LM.getStart()), 2107 /*IsStringLocation*/true, 2108 getSpecifierRange(startSpecifier, specifierLen)); 2109 2110 S.Diag(getLocationOfByte(LM.getStart()), diag::note_format_fix_specifier) 2111 << FixedLM->toString() 2112 << FixItHint::CreateReplacement(LMRange, FixedLM->toString()); 2113 2114 } else { 2115 EmitFormatDiagnostic(S.PDiag(diag::warn_format_non_standard) 2116 << LM.toString() << 0, 2117 getLocationOfByte(LM.getStart()), 2118 /*IsStringLocation*/true, 2119 getSpecifierRange(startSpecifier, specifierLen)); 2120 } 2121} 2122 2123void CheckFormatHandler::HandleNonStandardConversionSpecifier( 2124 const analyze_format_string::ConversionSpecifier &CS, 2125 const char *startSpecifier, unsigned specifierLen) { 2126 using namespace analyze_format_string; 2127 2128 // See if we know how to fix this conversion specifier. 2129 llvm::Optional<ConversionSpecifier> FixedCS = CS.getStandardSpecifier(); 2130 if (FixedCS) { 2131 EmitFormatDiagnostic(S.PDiag(diag::warn_format_non_standard) 2132 << CS.toString() << /*conversion specifier*/1, 2133 getLocationOfByte(CS.getStart()), 2134 /*IsStringLocation*/true, 2135 getSpecifierRange(startSpecifier, specifierLen)); 2136 2137 CharSourceRange CSRange = getSpecifierRange(CS.getStart(), CS.getLength()); 2138 S.Diag(getLocationOfByte(CS.getStart()), diag::note_format_fix_specifier) 2139 << FixedCS->toString() 2140 << FixItHint::CreateReplacement(CSRange, FixedCS->toString()); 2141 } else { 2142 EmitFormatDiagnostic(S.PDiag(diag::warn_format_non_standard) 2143 << CS.toString() << /*conversion specifier*/1, 2144 getLocationOfByte(CS.getStart()), 2145 /*IsStringLocation*/true, 2146 getSpecifierRange(startSpecifier, specifierLen)); 2147 } 2148} 2149 2150void CheckFormatHandler::HandlePosition(const char *startPos, 2151 unsigned posLen) { 2152 EmitFormatDiagnostic(S.PDiag(diag::warn_format_non_standard_positional_arg), 2153 getLocationOfByte(startPos), 2154 /*IsStringLocation*/true, 2155 getSpecifierRange(startPos, posLen)); 2156} 2157 2158void 2159CheckFormatHandler::HandleInvalidPosition(const char *startPos, unsigned posLen, 2160 analyze_format_string::PositionContext p) { 2161 EmitFormatDiagnostic(S.PDiag(diag::warn_format_invalid_positional_specifier) 2162 << (unsigned) p, 2163 getLocationOfByte(startPos), /*IsStringLocation*/true, 2164 getSpecifierRange(startPos, posLen)); 2165} 2166 2167void CheckFormatHandler::HandleZeroPosition(const char *startPos, 2168 unsigned posLen) { 2169 EmitFormatDiagnostic(S.PDiag(diag::warn_format_zero_positional_specifier), 2170 getLocationOfByte(startPos), 2171 /*IsStringLocation*/true, 2172 getSpecifierRange(startPos, posLen)); 2173} 2174 2175void CheckFormatHandler::HandleNullChar(const char *nullCharacter) { 2176 if (!isa<ObjCStringLiteral>(OrigFormatExpr)) { 2177 // The presence of a null character is likely an error. 2178 EmitFormatDiagnostic( 2179 S.PDiag(diag::warn_printf_format_string_contains_null_char), 2180 getLocationOfByte(nullCharacter), /*IsStringLocation*/true, 2181 getFormatStringRange()); 2182 } 2183} 2184 2185// Note that this may return NULL if there was an error parsing or building 2186// one of the argument expressions. 2187const Expr *CheckFormatHandler::getDataArg(unsigned i) const { 2188 return Args[FirstDataArg + i]; 2189} 2190 2191void CheckFormatHandler::DoneProcessing() { 2192 // Does the number of data arguments exceed the number of 2193 // format conversions in the format string? 2194 if (!HasVAListArg) { 2195 // Find any arguments that weren't covered. 2196 CoveredArgs.flip(); 2197 signed notCoveredArg = CoveredArgs.find_first(); 2198 if (notCoveredArg >= 0) { 2199 assert((unsigned)notCoveredArg < NumDataArgs); 2200 if (const Expr *E = getDataArg((unsigned) notCoveredArg)) { 2201 SourceLocation Loc = E->getLocStart(); 2202 if (!S.getSourceManager().isInSystemMacro(Loc)) { 2203 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_data_arg_not_used), 2204 Loc, /*IsStringLocation*/false, 2205 getFormatStringRange()); 2206 } 2207 } 2208 } 2209 } 2210} 2211 2212bool 2213CheckFormatHandler::HandleInvalidConversionSpecifier(unsigned argIndex, 2214 SourceLocation Loc, 2215 const char *startSpec, 2216 unsigned specifierLen, 2217 const char *csStart, 2218 unsigned csLen) { 2219 2220 bool keepGoing = true; 2221 if (argIndex < NumDataArgs) { 2222 // Consider the argument coverered, even though the specifier doesn't 2223 // make sense. 2224 CoveredArgs.set(argIndex); 2225 } 2226 else { 2227 // If argIndex exceeds the number of data arguments we 2228 // don't issue a warning because that is just a cascade of warnings (and 2229 // they may have intended '%%' anyway). We don't want to continue processing 2230 // the format string after this point, however, as we will like just get 2231 // gibberish when trying to match arguments. 2232 keepGoing = false; 2233 } 2234 2235 EmitFormatDiagnostic(S.PDiag(diag::warn_format_invalid_conversion) 2236 << StringRef(csStart, csLen), 2237 Loc, /*IsStringLocation*/true, 2238 getSpecifierRange(startSpec, specifierLen)); 2239 2240 return keepGoing; 2241} 2242 2243void 2244CheckFormatHandler::HandlePositionalNonpositionalArgs(SourceLocation Loc, 2245 const char *startSpec, 2246 unsigned specifierLen) { 2247 EmitFormatDiagnostic( 2248 S.PDiag(diag::warn_format_mix_positional_nonpositional_args), 2249 Loc, /*isStringLoc*/true, getSpecifierRange(startSpec, specifierLen)); 2250} 2251 2252bool 2253CheckFormatHandler::CheckNumArgs( 2254 const analyze_format_string::FormatSpecifier &FS, 2255 const analyze_format_string::ConversionSpecifier &CS, 2256 const char *startSpecifier, unsigned specifierLen, unsigned argIndex) { 2257 2258 if (argIndex >= NumDataArgs) { 2259 PartialDiagnostic PDiag = FS.usesPositionalArg() 2260 ? (S.PDiag(diag::warn_printf_positional_arg_exceeds_data_args) 2261 << (argIndex+1) << NumDataArgs) 2262 : S.PDiag(diag::warn_printf_insufficient_data_args); 2263 EmitFormatDiagnostic( 2264 PDiag, getLocationOfByte(CS.getStart()), /*IsStringLocation*/true, 2265 getSpecifierRange(startSpecifier, specifierLen)); 2266 return false; 2267 } 2268 return true; 2269} 2270 2271template<typename Range> 2272void CheckFormatHandler::EmitFormatDiagnostic(PartialDiagnostic PDiag, 2273 SourceLocation Loc, 2274 bool IsStringLocation, 2275 Range StringRange, 2276 ArrayRef<FixItHint> FixIt) { 2277 EmitFormatDiagnostic(S, inFunctionCall, Args[FormatIdx], PDiag, 2278 Loc, IsStringLocation, StringRange, FixIt); 2279} 2280 2281/// \brief If the format string is not within the funcion call, emit a note 2282/// so that the function call and string are in diagnostic messages. 2283/// 2284/// \param InFunctionCall if true, the format string is within the function 2285/// call and only one diagnostic message will be produced. Otherwise, an 2286/// extra note will be emitted pointing to location of the format string. 2287/// 2288/// \param ArgumentExpr the expression that is passed as the format string 2289/// argument in the function call. Used for getting locations when two 2290/// diagnostics are emitted. 2291/// 2292/// \param PDiag the callee should already have provided any strings for the 2293/// diagnostic message. This function only adds locations and fixits 2294/// to diagnostics. 2295/// 2296/// \param Loc primary location for diagnostic. If two diagnostics are 2297/// required, one will be at Loc and a new SourceLocation will be created for 2298/// the other one. 2299/// 2300/// \param IsStringLocation if true, Loc points to the format string should be 2301/// used for the note. Otherwise, Loc points to the argument list and will 2302/// be used with PDiag. 2303/// 2304/// \param StringRange some or all of the string to highlight. This is 2305/// templated so it can accept either a CharSourceRange or a SourceRange. 2306/// 2307/// \param FixIt optional fix it hint for the format string. 2308template<typename Range> 2309void CheckFormatHandler::EmitFormatDiagnostic(Sema &S, bool InFunctionCall, 2310 const Expr *ArgumentExpr, 2311 PartialDiagnostic PDiag, 2312 SourceLocation Loc, 2313 bool IsStringLocation, 2314 Range StringRange, 2315 ArrayRef<FixItHint> FixIt) { 2316 if (InFunctionCall) { 2317 const Sema::SemaDiagnosticBuilder &D = S.Diag(Loc, PDiag); 2318 D << StringRange; 2319 for (ArrayRef<FixItHint>::iterator I = FixIt.begin(), E = FixIt.end(); 2320 I != E; ++I) { 2321 D << *I; 2322 } 2323 } else { 2324 S.Diag(IsStringLocation ? ArgumentExpr->getExprLoc() : Loc, PDiag) 2325 << ArgumentExpr->getSourceRange(); 2326 2327 const Sema::SemaDiagnosticBuilder &Note = 2328 S.Diag(IsStringLocation ? Loc : StringRange.getBegin(), 2329 diag::note_format_string_defined); 2330 2331 Note << StringRange; 2332 for (ArrayRef<FixItHint>::iterator I = FixIt.begin(), E = FixIt.end(); 2333 I != E; ++I) { 2334 Note << *I; 2335 } 2336 } 2337} 2338 2339//===--- CHECK: Printf format string checking ------------------------------===// 2340 2341namespace { 2342class CheckPrintfHandler : public CheckFormatHandler { 2343 bool ObjCContext; 2344public: 2345 CheckPrintfHandler(Sema &s, const StringLiteral *fexpr, 2346 const Expr *origFormatExpr, unsigned firstDataArg, 2347 unsigned numDataArgs, bool isObjC, 2348 const char *beg, bool hasVAListArg, 2349 Expr **Args, unsigned NumArgs, 2350 unsigned formatIdx, bool inFunctionCall, 2351 Sema::VariadicCallType CallType) 2352 : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg, 2353 numDataArgs, beg, hasVAListArg, Args, NumArgs, 2354 formatIdx, inFunctionCall, CallType), ObjCContext(isObjC) 2355 {} 2356 2357 2358 bool HandleInvalidPrintfConversionSpecifier( 2359 const analyze_printf::PrintfSpecifier &FS, 2360 const char *startSpecifier, 2361 unsigned specifierLen); 2362 2363 bool HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier &FS, 2364 const char *startSpecifier, 2365 unsigned specifierLen); 2366 bool checkFormatExpr(const analyze_printf::PrintfSpecifier &FS, 2367 const char *StartSpecifier, 2368 unsigned SpecifierLen, 2369 const Expr *E); 2370 2371 bool HandleAmount(const analyze_format_string::OptionalAmount &Amt, unsigned k, 2372 const char *startSpecifier, unsigned specifierLen); 2373 void HandleInvalidAmount(const analyze_printf::PrintfSpecifier &FS, 2374 const analyze_printf::OptionalAmount &Amt, 2375 unsigned type, 2376 const char *startSpecifier, unsigned specifierLen); 2377 void HandleFlag(const analyze_printf::PrintfSpecifier &FS, 2378 const analyze_printf::OptionalFlag &flag, 2379 const char *startSpecifier, unsigned specifierLen); 2380 void HandleIgnoredFlag(const analyze_printf::PrintfSpecifier &FS, 2381 const analyze_printf::OptionalFlag &ignoredFlag, 2382 const analyze_printf::OptionalFlag &flag, 2383 const char *startSpecifier, unsigned specifierLen); 2384 bool checkForCStrMembers(const analyze_printf::ArgType &AT, 2385 const Expr *E, const CharSourceRange &CSR); 2386 2387}; 2388} 2389 2390bool CheckPrintfHandler::HandleInvalidPrintfConversionSpecifier( 2391 const analyze_printf::PrintfSpecifier &FS, 2392 const char *startSpecifier, 2393 unsigned specifierLen) { 2394 const analyze_printf::PrintfConversionSpecifier &CS = 2395 FS.getConversionSpecifier(); 2396 2397 return HandleInvalidConversionSpecifier(FS.getArgIndex(), 2398 getLocationOfByte(CS.getStart()), 2399 startSpecifier, specifierLen, 2400 CS.getStart(), CS.getLength()); 2401} 2402 2403bool CheckPrintfHandler::HandleAmount( 2404 const analyze_format_string::OptionalAmount &Amt, 2405 unsigned k, const char *startSpecifier, 2406 unsigned specifierLen) { 2407 2408 if (Amt.hasDataArgument()) { 2409 if (!HasVAListArg) { 2410 unsigned argIndex = Amt.getArgIndex(); 2411 if (argIndex >= NumDataArgs) { 2412 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_asterisk_missing_arg) 2413 << k, 2414 getLocationOfByte(Amt.getStart()), 2415 /*IsStringLocation*/true, 2416 getSpecifierRange(startSpecifier, specifierLen)); 2417 // Don't do any more checking. We will just emit 2418 // spurious errors. 2419 return false; 2420 } 2421 2422 // Type check the data argument. It should be an 'int'. 2423 // Although not in conformance with C99, we also allow the argument to be 2424 // an 'unsigned int' as that is a reasonably safe case. GCC also 2425 // doesn't emit a warning for that case. 2426 CoveredArgs.set(argIndex); 2427 const Expr *Arg = getDataArg(argIndex); 2428 if (!Arg) 2429 return false; 2430 2431 QualType T = Arg->getType(); 2432 2433 const analyze_printf::ArgType &AT = Amt.getArgType(S.Context); 2434 assert(AT.isValid()); 2435 2436 if (!AT.matchesType(S.Context, T)) { 2437 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_asterisk_wrong_type) 2438 << k << AT.getRepresentativeTypeName(S.Context) 2439 << T << Arg->getSourceRange(), 2440 getLocationOfByte(Amt.getStart()), 2441 /*IsStringLocation*/true, 2442 getSpecifierRange(startSpecifier, specifierLen)); 2443 // Don't do any more checking. We will just emit 2444 // spurious errors. 2445 return false; 2446 } 2447 } 2448 } 2449 return true; 2450} 2451 2452void CheckPrintfHandler::HandleInvalidAmount( 2453 const analyze_printf::PrintfSpecifier &FS, 2454 const analyze_printf::OptionalAmount &Amt, 2455 unsigned type, 2456 const char *startSpecifier, 2457 unsigned specifierLen) { 2458 const analyze_printf::PrintfConversionSpecifier &CS = 2459 FS.getConversionSpecifier(); 2460 2461 FixItHint fixit = 2462 Amt.getHowSpecified() == analyze_printf::OptionalAmount::Constant 2463 ? FixItHint::CreateRemoval(getSpecifierRange(Amt.getStart(), 2464 Amt.getConstantLength())) 2465 : FixItHint(); 2466 2467 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_nonsensical_optional_amount) 2468 << type << CS.toString(), 2469 getLocationOfByte(Amt.getStart()), 2470 /*IsStringLocation*/true, 2471 getSpecifierRange(startSpecifier, specifierLen), 2472 fixit); 2473} 2474 2475void CheckPrintfHandler::HandleFlag(const analyze_printf::PrintfSpecifier &FS, 2476 const analyze_printf::OptionalFlag &flag, 2477 const char *startSpecifier, 2478 unsigned specifierLen) { 2479 // Warn about pointless flag with a fixit removal. 2480 const analyze_printf::PrintfConversionSpecifier &CS = 2481 FS.getConversionSpecifier(); 2482 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_nonsensical_flag) 2483 << flag.toString() << CS.toString(), 2484 getLocationOfByte(flag.getPosition()), 2485 /*IsStringLocation*/true, 2486 getSpecifierRange(startSpecifier, specifierLen), 2487 FixItHint::CreateRemoval( 2488 getSpecifierRange(flag.getPosition(), 1))); 2489} 2490 2491void CheckPrintfHandler::HandleIgnoredFlag( 2492 const analyze_printf::PrintfSpecifier &FS, 2493 const analyze_printf::OptionalFlag &ignoredFlag, 2494 const analyze_printf::OptionalFlag &flag, 2495 const char *startSpecifier, 2496 unsigned specifierLen) { 2497 // Warn about ignored flag with a fixit removal. 2498 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_ignored_flag) 2499 << ignoredFlag.toString() << flag.toString(), 2500 getLocationOfByte(ignoredFlag.getPosition()), 2501 /*IsStringLocation*/true, 2502 getSpecifierRange(startSpecifier, specifierLen), 2503 FixItHint::CreateRemoval( 2504 getSpecifierRange(ignoredFlag.getPosition(), 1))); 2505} 2506 2507// Determines if the specified is a C++ class or struct containing 2508// a member with the specified name and kind (e.g. a CXXMethodDecl named 2509// "c_str()"). 2510template<typename MemberKind> 2511static llvm::SmallPtrSet<MemberKind*, 1> 2512CXXRecordMembersNamed(StringRef Name, Sema &S, QualType Ty) { 2513 const RecordType *RT = Ty->getAs<RecordType>(); 2514 llvm::SmallPtrSet<MemberKind*, 1> Results; 2515 2516 if (!RT) 2517 return Results; 2518 const CXXRecordDecl *RD = dyn_cast<CXXRecordDecl>(RT->getDecl()); 2519 if (!RD) 2520 return Results; 2521 2522 LookupResult R(S, &S.PP.getIdentifierTable().get(Name), SourceLocation(), 2523 Sema::LookupMemberName); 2524 2525 // We just need to include all members of the right kind turned up by the 2526 // filter, at this point. 2527 if (S.LookupQualifiedName(R, RT->getDecl())) 2528 for (LookupResult::iterator I = R.begin(), E = R.end(); I != E; ++I) { 2529 NamedDecl *decl = (*I)->getUnderlyingDecl(); 2530 if (MemberKind *FK = dyn_cast<MemberKind>(decl)) 2531 Results.insert(FK); 2532 } 2533 return Results; 2534} 2535 2536// Check if a (w)string was passed when a (w)char* was needed, and offer a 2537// better diagnostic if so. AT is assumed to be valid. 2538// Returns true when a c_str() conversion method is found. 2539bool CheckPrintfHandler::checkForCStrMembers( 2540 const analyze_printf::ArgType &AT, const Expr *E, 2541 const CharSourceRange &CSR) { 2542 typedef llvm::SmallPtrSet<CXXMethodDecl*, 1> MethodSet; 2543 2544 MethodSet Results = 2545 CXXRecordMembersNamed<CXXMethodDecl>("c_str", S, E->getType()); 2546 2547 for (MethodSet::iterator MI = Results.begin(), ME = Results.end(); 2548 MI != ME; ++MI) { 2549 const CXXMethodDecl *Method = *MI; 2550 if (Method->getNumParams() == 0 && 2551 AT.matchesType(S.Context, Method->getResultType())) { 2552 // FIXME: Suggest parens if the expression needs them. 2553 SourceLocation EndLoc = 2554 S.getPreprocessor().getLocForEndOfToken(E->getLocEnd()); 2555 S.Diag(E->getLocStart(), diag::note_printf_c_str) 2556 << "c_str()" 2557 << FixItHint::CreateInsertion(EndLoc, ".c_str()"); 2558 return true; 2559 } 2560 } 2561 2562 return false; 2563} 2564 2565bool 2566CheckPrintfHandler::HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier 2567 &FS, 2568 const char *startSpecifier, 2569 unsigned specifierLen) { 2570 2571 using namespace analyze_format_string; 2572 using namespace analyze_printf; 2573 const PrintfConversionSpecifier &CS = FS.getConversionSpecifier(); 2574 2575 if (FS.consumesDataArgument()) { 2576 if (atFirstArg) { 2577 atFirstArg = false; 2578 usesPositionalArgs = FS.usesPositionalArg(); 2579 } 2580 else if (usesPositionalArgs != FS.usesPositionalArg()) { 2581 HandlePositionalNonpositionalArgs(getLocationOfByte(CS.getStart()), 2582 startSpecifier, specifierLen); 2583 return false; 2584 } 2585 } 2586 2587 // First check if the field width, precision, and conversion specifier 2588 // have matching data arguments. 2589 if (!HandleAmount(FS.getFieldWidth(), /* field width */ 0, 2590 startSpecifier, specifierLen)) { 2591 return false; 2592 } 2593 2594 if (!HandleAmount(FS.getPrecision(), /* precision */ 1, 2595 startSpecifier, specifierLen)) { 2596 return false; 2597 } 2598 2599 if (!CS.consumesDataArgument()) { 2600 // FIXME: Technically specifying a precision or field width here 2601 // makes no sense. Worth issuing a warning at some point. 2602 return true; 2603 } 2604 2605 // Consume the argument. 2606 unsigned argIndex = FS.getArgIndex(); 2607 if (argIndex < NumDataArgs) { 2608 // The check to see if the argIndex is valid will come later. 2609 // We set the bit here because we may exit early from this 2610 // function if we encounter some other error. 2611 CoveredArgs.set(argIndex); 2612 } 2613 2614 // Check for using an Objective-C specific conversion specifier 2615 // in a non-ObjC literal. 2616 if (!ObjCContext && CS.isObjCArg()) { 2617 return HandleInvalidPrintfConversionSpecifier(FS, startSpecifier, 2618 specifierLen); 2619 } 2620 2621 // Check for invalid use of field width 2622 if (!FS.hasValidFieldWidth()) { 2623 HandleInvalidAmount(FS, FS.getFieldWidth(), /* field width */ 0, 2624 startSpecifier, specifierLen); 2625 } 2626 2627 // Check for invalid use of precision 2628 if (!FS.hasValidPrecision()) { 2629 HandleInvalidAmount(FS, FS.getPrecision(), /* precision */ 1, 2630 startSpecifier, specifierLen); 2631 } 2632 2633 // Check each flag does not conflict with any other component. 2634 if (!FS.hasValidThousandsGroupingPrefix()) 2635 HandleFlag(FS, FS.hasThousandsGrouping(), startSpecifier, specifierLen); 2636 if (!FS.hasValidLeadingZeros()) 2637 HandleFlag(FS, FS.hasLeadingZeros(), startSpecifier, specifierLen); 2638 if (!FS.hasValidPlusPrefix()) 2639 HandleFlag(FS, FS.hasPlusPrefix(), startSpecifier, specifierLen); 2640 if (!FS.hasValidSpacePrefix()) 2641 HandleFlag(FS, FS.hasSpacePrefix(), startSpecifier, specifierLen); 2642 if (!FS.hasValidAlternativeForm()) 2643 HandleFlag(FS, FS.hasAlternativeForm(), startSpecifier, specifierLen); 2644 if (!FS.hasValidLeftJustified()) 2645 HandleFlag(FS, FS.isLeftJustified(), startSpecifier, specifierLen); 2646 2647 // Check that flags are not ignored by another flag 2648 if (FS.hasSpacePrefix() && FS.hasPlusPrefix()) // ' ' ignored by '+' 2649 HandleIgnoredFlag(FS, FS.hasSpacePrefix(), FS.hasPlusPrefix(), 2650 startSpecifier, specifierLen); 2651 if (FS.hasLeadingZeros() && FS.isLeftJustified()) // '0' ignored by '-' 2652 HandleIgnoredFlag(FS, FS.hasLeadingZeros(), FS.isLeftJustified(), 2653 startSpecifier, specifierLen); 2654 2655 // Check the length modifier is valid with the given conversion specifier. 2656 if (!FS.hasValidLengthModifier(S.getASTContext().getTargetInfo())) 2657 HandleInvalidLengthModifier(FS, CS, startSpecifier, specifierLen, 2658 diag::warn_format_nonsensical_length); 2659 else if (!FS.hasStandardLengthModifier()) 2660 HandleNonStandardLengthModifier(FS, startSpecifier, specifierLen); 2661 else if (!FS.hasStandardLengthConversionCombination()) 2662 HandleInvalidLengthModifier(FS, CS, startSpecifier, specifierLen, 2663 diag::warn_format_non_standard_conversion_spec); 2664 2665 if (!FS.hasStandardConversionSpecifier(S.getLangOpts())) 2666 HandleNonStandardConversionSpecifier(CS, startSpecifier, specifierLen); 2667 2668 // The remaining checks depend on the data arguments. 2669 if (HasVAListArg) 2670 return true; 2671 2672 if (!CheckNumArgs(FS, CS, startSpecifier, specifierLen, argIndex)) 2673 return false; 2674 2675 const Expr *Arg = getDataArg(argIndex); 2676 if (!Arg) 2677 return true; 2678 2679 return checkFormatExpr(FS, startSpecifier, specifierLen, Arg); 2680} 2681 2682static bool requiresParensToAddCast(const Expr *E) { 2683 // FIXME: We should have a general way to reason about operator 2684 // precedence and whether parens are actually needed here. 2685 // Take care of a few common cases where they aren't. 2686 const Expr *Inside = E->IgnoreImpCasts(); 2687 if (const PseudoObjectExpr *POE = dyn_cast<PseudoObjectExpr>(Inside)) 2688 Inside = POE->getSyntacticForm()->IgnoreImpCasts(); 2689 2690 switch (Inside->getStmtClass()) { 2691 case Stmt::ArraySubscriptExprClass: 2692 case Stmt::CallExprClass: 2693 case Stmt::CharacterLiteralClass: 2694 case Stmt::CXXBoolLiteralExprClass: 2695 case Stmt::DeclRefExprClass: 2696 case Stmt::FloatingLiteralClass: 2697 case Stmt::IntegerLiteralClass: 2698 case Stmt::MemberExprClass: 2699 case Stmt::ObjCArrayLiteralClass: 2700 case Stmt::ObjCBoolLiteralExprClass: 2701 case Stmt::ObjCBoxedExprClass: 2702 case Stmt::ObjCDictionaryLiteralClass: 2703 case Stmt::ObjCEncodeExprClass: 2704 case Stmt::ObjCIvarRefExprClass: 2705 case Stmt::ObjCMessageExprClass: 2706 case Stmt::ObjCPropertyRefExprClass: 2707 case Stmt::ObjCStringLiteralClass: 2708 case Stmt::ObjCSubscriptRefExprClass: 2709 case Stmt::ParenExprClass: 2710 case Stmt::StringLiteralClass: 2711 case Stmt::UnaryOperatorClass: 2712 return false; 2713 default: 2714 return true; 2715 } 2716} 2717 2718bool 2719CheckPrintfHandler::checkFormatExpr(const analyze_printf::PrintfSpecifier &FS, 2720 const char *StartSpecifier, 2721 unsigned SpecifierLen, 2722 const Expr *E) { 2723 using namespace analyze_format_string; 2724 using namespace analyze_printf; 2725 // Now type check the data expression that matches the 2726 // format specifier. 2727 const analyze_printf::ArgType &AT = FS.getArgType(S.Context, 2728 ObjCContext); 2729 if (!AT.isValid()) 2730 return true; 2731 2732 QualType ExprTy = E->getType(); 2733 if (AT.matchesType(S.Context, ExprTy)) 2734 return true; 2735 2736 // Look through argument promotions for our error message's reported type. 2737 // This includes the integral and floating promotions, but excludes array 2738 // and function pointer decay; seeing that an argument intended to be a 2739 // string has type 'char [6]' is probably more confusing than 'char *'. 2740 if (const ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(E)) { 2741 if (ICE->getCastKind() == CK_IntegralCast || 2742 ICE->getCastKind() == CK_FloatingCast) { 2743 E = ICE->getSubExpr(); 2744 ExprTy = E->getType(); 2745 2746 // Check if we didn't match because of an implicit cast from a 'char' 2747 // or 'short' to an 'int'. This is done because printf is a varargs 2748 // function. 2749 if (ICE->getType() == S.Context.IntTy || 2750 ICE->getType() == S.Context.UnsignedIntTy) { 2751 // All further checking is done on the subexpression. 2752 if (AT.matchesType(S.Context, ExprTy)) 2753 return true; 2754 } 2755 } 2756 } else if (const CharacterLiteral *CL = dyn_cast<CharacterLiteral>(E)) { 2757 // Special case for 'a', which has type 'int' in C. 2758 // Note, however, that we do /not/ want to treat multibyte constants like 2759 // 'MooV' as characters! This form is deprecated but still exists. 2760 if (ExprTy == S.Context.IntTy) 2761 if (llvm::isUIntN(S.Context.getCharWidth(), CL->getValue())) 2762 ExprTy = S.Context.CharTy; 2763 } 2764 2765 // %C in an Objective-C context prints a unichar, not a wchar_t. 2766 // If the argument is an integer of some kind, believe the %C and suggest 2767 // a cast instead of changing the conversion specifier. 2768 QualType IntendedTy = ExprTy; 2769 if (ObjCContext && 2770 FS.getConversionSpecifier().getKind() == ConversionSpecifier::CArg) { 2771 if (ExprTy->isIntegralOrUnscopedEnumerationType() && 2772 !ExprTy->isCharType()) { 2773 // 'unichar' is defined as a typedef of unsigned short, but we should 2774 // prefer using the typedef if it is visible. 2775 IntendedTy = S.Context.UnsignedShortTy; 2776 2777 LookupResult Result(S, &S.Context.Idents.get("unichar"), E->getLocStart(), 2778 Sema::LookupOrdinaryName); 2779 if (S.LookupName(Result, S.getCurScope())) { 2780 NamedDecl *ND = Result.getFoundDecl(); 2781 if (TypedefNameDecl *TD = dyn_cast<TypedefNameDecl>(ND)) 2782 if (TD->getUnderlyingType() == IntendedTy) 2783 IntendedTy = S.Context.getTypedefType(TD); 2784 } 2785 } 2786 } 2787 2788 // Special-case some of Darwin's platform-independence types by suggesting 2789 // casts to primitive types that are known to be large enough. 2790 bool ShouldNotPrintDirectly = false; 2791 if (S.Context.getTargetInfo().getTriple().isOSDarwin()) { 2792 if (const TypedefType *UserTy = IntendedTy->getAs<TypedefType>()) { 2793 StringRef Name = UserTy->getDecl()->getName(); 2794 QualType CastTy = llvm::StringSwitch<QualType>(Name) 2795 .Case("NSInteger", S.Context.LongTy) 2796 .Case("NSUInteger", S.Context.UnsignedLongTy) 2797 .Case("SInt32", S.Context.IntTy) 2798 .Case("UInt32", S.Context.UnsignedIntTy) 2799 .Default(QualType()); 2800 2801 if (!CastTy.isNull()) { 2802 ShouldNotPrintDirectly = true; 2803 IntendedTy = CastTy; 2804 } 2805 } 2806 } 2807 2808 // We may be able to offer a FixItHint if it is a supported type. 2809 PrintfSpecifier fixedFS = FS; 2810 bool success = fixedFS.fixType(IntendedTy, S.getLangOpts(), 2811 S.Context, ObjCContext); 2812 2813 if (success) { 2814 // Get the fix string from the fixed format specifier 2815 SmallString<16> buf; 2816 llvm::raw_svector_ostream os(buf); 2817 fixedFS.toString(os); 2818 2819 CharSourceRange SpecRange = getSpecifierRange(StartSpecifier, SpecifierLen); 2820 2821 if (IntendedTy == ExprTy) { 2822 // In this case, the specifier is wrong and should be changed to match 2823 // the argument. 2824 EmitFormatDiagnostic( 2825 S.PDiag(diag::warn_printf_conversion_argument_type_mismatch) 2826 << AT.getRepresentativeTypeName(S.Context) << IntendedTy 2827 << E->getSourceRange(), 2828 E->getLocStart(), 2829 /*IsStringLocation*/false, 2830 SpecRange, 2831 FixItHint::CreateReplacement(SpecRange, os.str())); 2832 2833 } else { 2834 // The canonical type for formatting this value is different from the 2835 // actual type of the expression. (This occurs, for example, with Darwin's 2836 // NSInteger on 32-bit platforms, where it is typedef'd as 'int', but 2837 // should be printed as 'long' for 64-bit compatibility.) 2838 // Rather than emitting a normal format/argument mismatch, we want to 2839 // add a cast to the recommended type (and correct the format string 2840 // if necessary). 2841 SmallString<16> CastBuf; 2842 llvm::raw_svector_ostream CastFix(CastBuf); 2843 CastFix << "("; 2844 IntendedTy.print(CastFix, S.Context.getPrintingPolicy()); 2845 CastFix << ")"; 2846 2847 SmallVector<FixItHint,4> Hints; 2848 if (!AT.matchesType(S.Context, IntendedTy)) 2849 Hints.push_back(FixItHint::CreateReplacement(SpecRange, os.str())); 2850 2851 if (const CStyleCastExpr *CCast = dyn_cast<CStyleCastExpr>(E)) { 2852 // If there's already a cast present, just replace it. 2853 SourceRange CastRange(CCast->getLParenLoc(), CCast->getRParenLoc()); 2854 Hints.push_back(FixItHint::CreateReplacement(CastRange, CastFix.str())); 2855 2856 } else if (!requiresParensToAddCast(E)) { 2857 // If the expression has high enough precedence, 2858 // just write the C-style cast. 2859 Hints.push_back(FixItHint::CreateInsertion(E->getLocStart(), 2860 CastFix.str())); 2861 } else { 2862 // Otherwise, add parens around the expression as well as the cast. 2863 CastFix << "("; 2864 Hints.push_back(FixItHint::CreateInsertion(E->getLocStart(), 2865 CastFix.str())); 2866 2867 SourceLocation After = S.PP.getLocForEndOfToken(E->getLocEnd()); 2868 Hints.push_back(FixItHint::CreateInsertion(After, ")")); 2869 } 2870 2871 if (ShouldNotPrintDirectly) { 2872 // The expression has a type that should not be printed directly. 2873 // We extract the name from the typedef because we don't want to show 2874 // the underlying type in the diagnostic. 2875 StringRef Name = cast<TypedefType>(ExprTy)->getDecl()->getName(); 2876 2877 EmitFormatDiagnostic(S.PDiag(diag::warn_format_argument_needs_cast) 2878 << Name << IntendedTy 2879 << E->getSourceRange(), 2880 E->getLocStart(), /*IsStringLocation=*/false, 2881 SpecRange, Hints); 2882 } else { 2883 // In this case, the expression could be printed using a different 2884 // specifier, but we've decided that the specifier is probably correct 2885 // and we should cast instead. Just use the normal warning message. 2886 EmitFormatDiagnostic( 2887 S.PDiag(diag::warn_printf_conversion_argument_type_mismatch) 2888 << AT.getRepresentativeTypeName(S.Context) << ExprTy 2889 << E->getSourceRange(), 2890 E->getLocStart(), /*IsStringLocation*/false, 2891 SpecRange, Hints); 2892 } 2893 } 2894 } else { 2895 const CharSourceRange &CSR = getSpecifierRange(StartSpecifier, 2896 SpecifierLen); 2897 // Since the warning for passing non-POD types to variadic functions 2898 // was deferred until now, we emit a warning for non-POD 2899 // arguments here. 2900 if (S.isValidVarArgType(ExprTy) == Sema::VAK_Invalid) { 2901 unsigned DiagKind; 2902 if (ExprTy->isObjCObjectType()) 2903 DiagKind = diag::err_cannot_pass_objc_interface_to_vararg_format; 2904 else 2905 DiagKind = diag::warn_non_pod_vararg_with_format_string; 2906 2907 EmitFormatDiagnostic( 2908 S.PDiag(DiagKind) 2909 << S.getLangOpts().CPlusPlus0x 2910 << ExprTy 2911 << CallType 2912 << AT.getRepresentativeTypeName(S.Context) 2913 << CSR 2914 << E->getSourceRange(), 2915 E->getLocStart(), /*IsStringLocation*/false, CSR); 2916 2917 checkForCStrMembers(AT, E, CSR); 2918 } else 2919 EmitFormatDiagnostic( 2920 S.PDiag(diag::warn_printf_conversion_argument_type_mismatch) 2921 << AT.getRepresentativeTypeName(S.Context) << ExprTy 2922 << CSR 2923 << E->getSourceRange(), 2924 E->getLocStart(), /*IsStringLocation*/false, CSR); 2925 } 2926 2927 return true; 2928} 2929 2930//===--- CHECK: Scanf format string checking ------------------------------===// 2931 2932namespace { 2933class CheckScanfHandler : public CheckFormatHandler { 2934public: 2935 CheckScanfHandler(Sema &s, const StringLiteral *fexpr, 2936 const Expr *origFormatExpr, unsigned firstDataArg, 2937 unsigned numDataArgs, const char *beg, bool hasVAListArg, 2938 Expr **Args, unsigned NumArgs, 2939 unsigned formatIdx, bool inFunctionCall, 2940 Sema::VariadicCallType CallType) 2941 : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg, 2942 numDataArgs, beg, hasVAListArg, 2943 Args, NumArgs, formatIdx, inFunctionCall, CallType) 2944 {} 2945 2946 bool HandleScanfSpecifier(const analyze_scanf::ScanfSpecifier &FS, 2947 const char *startSpecifier, 2948 unsigned specifierLen); 2949 2950 bool HandleInvalidScanfConversionSpecifier( 2951 const analyze_scanf::ScanfSpecifier &FS, 2952 const char *startSpecifier, 2953 unsigned specifierLen); 2954 2955 void HandleIncompleteScanList(const char *start, const char *end); 2956}; 2957} 2958 2959void CheckScanfHandler::HandleIncompleteScanList(const char *start, 2960 const char *end) { 2961 EmitFormatDiagnostic(S.PDiag(diag::warn_scanf_scanlist_incomplete), 2962 getLocationOfByte(end), /*IsStringLocation*/true, 2963 getSpecifierRange(start, end - start)); 2964} 2965 2966bool CheckScanfHandler::HandleInvalidScanfConversionSpecifier( 2967 const analyze_scanf::ScanfSpecifier &FS, 2968 const char *startSpecifier, 2969 unsigned specifierLen) { 2970 2971 const analyze_scanf::ScanfConversionSpecifier &CS = 2972 FS.getConversionSpecifier(); 2973 2974 return HandleInvalidConversionSpecifier(FS.getArgIndex(), 2975 getLocationOfByte(CS.getStart()), 2976 startSpecifier, specifierLen, 2977 CS.getStart(), CS.getLength()); 2978} 2979 2980bool CheckScanfHandler::HandleScanfSpecifier( 2981 const analyze_scanf::ScanfSpecifier &FS, 2982 const char *startSpecifier, 2983 unsigned specifierLen) { 2984 2985 using namespace analyze_scanf; 2986 using namespace analyze_format_string; 2987 2988 const ScanfConversionSpecifier &CS = FS.getConversionSpecifier(); 2989 2990 // Handle case where '%' and '*' don't consume an argument. These shouldn't 2991 // be used to decide if we are using positional arguments consistently. 2992 if (FS.consumesDataArgument()) { 2993 if (atFirstArg) { 2994 atFirstArg = false; 2995 usesPositionalArgs = FS.usesPositionalArg(); 2996 } 2997 else if (usesPositionalArgs != FS.usesPositionalArg()) { 2998 HandlePositionalNonpositionalArgs(getLocationOfByte(CS.getStart()), 2999 startSpecifier, specifierLen); 3000 return false; 3001 } 3002 } 3003 3004 // Check if the field with is non-zero. 3005 const OptionalAmount &Amt = FS.getFieldWidth(); 3006 if (Amt.getHowSpecified() == OptionalAmount::Constant) { 3007 if (Amt.getConstantAmount() == 0) { 3008 const CharSourceRange &R = getSpecifierRange(Amt.getStart(), 3009 Amt.getConstantLength()); 3010 EmitFormatDiagnostic(S.PDiag(diag::warn_scanf_nonzero_width), 3011 getLocationOfByte(Amt.getStart()), 3012 /*IsStringLocation*/true, R, 3013 FixItHint::CreateRemoval(R)); 3014 } 3015 } 3016 3017 if (!FS.consumesDataArgument()) { 3018 // FIXME: Technically specifying a precision or field width here 3019 // makes no sense. Worth issuing a warning at some point. 3020 return true; 3021 } 3022 3023 // Consume the argument. 3024 unsigned argIndex = FS.getArgIndex(); 3025 if (argIndex < NumDataArgs) { 3026 // The check to see if the argIndex is valid will come later. 3027 // We set the bit here because we may exit early from this 3028 // function if we encounter some other error. 3029 CoveredArgs.set(argIndex); 3030 } 3031 3032 // Check the length modifier is valid with the given conversion specifier. 3033 if (!FS.hasValidLengthModifier(S.getASTContext().getTargetInfo())) 3034 HandleInvalidLengthModifier(FS, CS, startSpecifier, specifierLen, 3035 diag::warn_format_nonsensical_length); 3036 else if (!FS.hasStandardLengthModifier()) 3037 HandleNonStandardLengthModifier(FS, startSpecifier, specifierLen); 3038 else if (!FS.hasStandardLengthConversionCombination()) 3039 HandleInvalidLengthModifier(FS, CS, startSpecifier, specifierLen, 3040 diag::warn_format_non_standard_conversion_spec); 3041 3042 if (!FS.hasStandardConversionSpecifier(S.getLangOpts())) 3043 HandleNonStandardConversionSpecifier(CS, startSpecifier, specifierLen); 3044 3045 // The remaining checks depend on the data arguments. 3046 if (HasVAListArg) 3047 return true; 3048 3049 if (!CheckNumArgs(FS, CS, startSpecifier, specifierLen, argIndex)) 3050 return false; 3051 3052 // Check that the argument type matches the format specifier. 3053 const Expr *Ex = getDataArg(argIndex); 3054 if (!Ex) 3055 return true; 3056 3057 const analyze_format_string::ArgType &AT = FS.getArgType(S.Context); 3058 if (AT.isValid() && !AT.matchesType(S.Context, Ex->getType())) { 3059 ScanfSpecifier fixedFS = FS; 3060 bool success = fixedFS.fixType(Ex->getType(), S.getLangOpts(), 3061 S.Context); 3062 3063 if (success) { 3064 // Get the fix string from the fixed format specifier. 3065 SmallString<128> buf; 3066 llvm::raw_svector_ostream os(buf); 3067 fixedFS.toString(os); 3068 3069 EmitFormatDiagnostic( 3070 S.PDiag(diag::warn_printf_conversion_argument_type_mismatch) 3071 << AT.getRepresentativeTypeName(S.Context) << Ex->getType() 3072 << Ex->getSourceRange(), 3073 Ex->getLocStart(), 3074 /*IsStringLocation*/false, 3075 getSpecifierRange(startSpecifier, specifierLen), 3076 FixItHint::CreateReplacement( 3077 getSpecifierRange(startSpecifier, specifierLen), 3078 os.str())); 3079 } else { 3080 EmitFormatDiagnostic( 3081 S.PDiag(diag::warn_printf_conversion_argument_type_mismatch) 3082 << AT.getRepresentativeTypeName(S.Context) << Ex->getType() 3083 << Ex->getSourceRange(), 3084 Ex->getLocStart(), 3085 /*IsStringLocation*/false, 3086 getSpecifierRange(startSpecifier, specifierLen)); 3087 } 3088 } 3089 3090 return true; 3091} 3092 3093void Sema::CheckFormatString(const StringLiteral *FExpr, 3094 const Expr *OrigFormatExpr, 3095 Expr **Args, unsigned NumArgs, 3096 bool HasVAListArg, unsigned format_idx, 3097 unsigned firstDataArg, FormatStringType Type, 3098 bool inFunctionCall, VariadicCallType CallType) { 3099 3100 // CHECK: is the format string a wide literal? 3101 if (!FExpr->isAscii() && !FExpr->isUTF8()) { 3102 CheckFormatHandler::EmitFormatDiagnostic( 3103 *this, inFunctionCall, Args[format_idx], 3104 PDiag(diag::warn_format_string_is_wide_literal), FExpr->getLocStart(), 3105 /*IsStringLocation*/true, OrigFormatExpr->getSourceRange()); 3106 return; 3107 } 3108 3109 // Str - The format string. NOTE: this is NOT null-terminated! 3110 StringRef StrRef = FExpr->getString(); 3111 const char *Str = StrRef.data(); 3112 unsigned StrLen = StrRef.size(); 3113 const unsigned numDataArgs = NumArgs - firstDataArg; 3114 3115 // CHECK: empty format string? 3116 if (StrLen == 0 && numDataArgs > 0) { 3117 CheckFormatHandler::EmitFormatDiagnostic( 3118 *this, inFunctionCall, Args[format_idx], 3119 PDiag(diag::warn_empty_format_string), FExpr->getLocStart(), 3120 /*IsStringLocation*/true, OrigFormatExpr->getSourceRange()); 3121 return; 3122 } 3123 3124 if (Type == FST_Printf || Type == FST_NSString) { 3125 CheckPrintfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg, 3126 numDataArgs, (Type == FST_NSString), 3127 Str, HasVAListArg, Args, NumArgs, format_idx, 3128 inFunctionCall, CallType); 3129 3130 if (!analyze_format_string::ParsePrintfString(H, Str, Str + StrLen, 3131 getLangOpts(), 3132 Context.getTargetInfo())) 3133 H.DoneProcessing(); 3134 } else if (Type == FST_Scanf) { 3135 CheckScanfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg, numDataArgs, 3136 Str, HasVAListArg, Args, NumArgs, format_idx, 3137 inFunctionCall, CallType); 3138 3139 if (!analyze_format_string::ParseScanfString(H, Str, Str + StrLen, 3140 getLangOpts(), 3141 Context.getTargetInfo())) 3142 H.DoneProcessing(); 3143 } // TODO: handle other formats 3144} 3145 3146//===--- CHECK: Standard memory functions ---------------------------------===// 3147 3148/// \brief Determine whether the given type is a dynamic class type (e.g., 3149/// whether it has a vtable). 3150static bool isDynamicClassType(QualType T) { 3151 if (CXXRecordDecl *Record = T->getAsCXXRecordDecl()) 3152 if (CXXRecordDecl *Definition = Record->getDefinition()) 3153 if (Definition->isDynamicClass()) 3154 return true; 3155 3156 return false; 3157} 3158 3159/// \brief If E is a sizeof expression, returns its argument expression, 3160/// otherwise returns NULL. 3161static const Expr *getSizeOfExprArg(const Expr* E) { 3162 if (const UnaryExprOrTypeTraitExpr *SizeOf = 3163 dyn_cast<UnaryExprOrTypeTraitExpr>(E)) 3164 if (SizeOf->getKind() == clang::UETT_SizeOf && !SizeOf->isArgumentType()) 3165 return SizeOf->getArgumentExpr()->IgnoreParenImpCasts(); 3166 3167 return 0; 3168} 3169 3170/// \brief If E is a sizeof expression, returns its argument type. 3171static QualType getSizeOfArgType(const Expr* E) { 3172 if (const UnaryExprOrTypeTraitExpr *SizeOf = 3173 dyn_cast<UnaryExprOrTypeTraitExpr>(E)) 3174 if (SizeOf->getKind() == clang::UETT_SizeOf) 3175 return SizeOf->getTypeOfArgument(); 3176 3177 return QualType(); 3178} 3179 3180/// \brief Check for dangerous or invalid arguments to memset(). 3181/// 3182/// This issues warnings on known problematic, dangerous or unspecified 3183/// arguments to the standard 'memset', 'memcpy', 'memmove', and 'memcmp' 3184/// function calls. 3185/// 3186/// \param Call The call expression to diagnose. 3187void Sema::CheckMemaccessArguments(const CallExpr *Call, 3188 unsigned BId, 3189 IdentifierInfo *FnName) { 3190 assert(BId != 0); 3191 3192 // It is possible to have a non-standard definition of memset. Validate 3193 // we have enough arguments, and if not, abort further checking. 3194 unsigned ExpectedNumArgs = (BId == Builtin::BIstrndup ? 2 : 3); 3195 if (Call->getNumArgs() < ExpectedNumArgs) 3196 return; 3197 3198 unsigned LastArg = (BId == Builtin::BImemset || 3199 BId == Builtin::BIstrndup ? 1 : 2); 3200 unsigned LenArg = (BId == Builtin::BIstrndup ? 1 : 2); 3201 const Expr *LenExpr = Call->getArg(LenArg)->IgnoreParenImpCasts(); 3202 3203 // We have special checking when the length is a sizeof expression. 3204 QualType SizeOfArgTy = getSizeOfArgType(LenExpr); 3205 const Expr *SizeOfArg = getSizeOfExprArg(LenExpr); 3206 llvm::FoldingSetNodeID SizeOfArgID; 3207 3208 for (unsigned ArgIdx = 0; ArgIdx != LastArg; ++ArgIdx) { 3209 const Expr *Dest = Call->getArg(ArgIdx)->IgnoreParenImpCasts(); 3210 SourceRange ArgRange = Call->getArg(ArgIdx)->getSourceRange(); 3211 3212 QualType DestTy = Dest->getType(); 3213 if (const PointerType *DestPtrTy = DestTy->getAs<PointerType>()) { 3214 QualType PointeeTy = DestPtrTy->getPointeeType(); 3215 3216 // Never warn about void type pointers. This can be used to suppress 3217 // false positives. 3218 if (PointeeTy->isVoidType()) 3219 continue; 3220 3221 // Catch "memset(p, 0, sizeof(p))" -- needs to be sizeof(*p). Do this by 3222 // actually comparing the expressions for equality. Because computing the 3223 // expression IDs can be expensive, we only do this if the diagnostic is 3224 // enabled. 3225 if (SizeOfArg && 3226 Diags.getDiagnosticLevel(diag::warn_sizeof_pointer_expr_memaccess, 3227 SizeOfArg->getExprLoc())) { 3228 // We only compute IDs for expressions if the warning is enabled, and 3229 // cache the sizeof arg's ID. 3230 if (SizeOfArgID == llvm::FoldingSetNodeID()) 3231 SizeOfArg->Profile(SizeOfArgID, Context, true); 3232 llvm::FoldingSetNodeID DestID; 3233 Dest->Profile(DestID, Context, true); 3234 if (DestID == SizeOfArgID) { 3235 // TODO: For strncpy() and friends, this could suggest sizeof(dst) 3236 // over sizeof(src) as well. 3237 unsigned ActionIdx = 0; // Default is to suggest dereferencing. 3238 StringRef ReadableName = FnName->getName(); 3239 3240 if (const UnaryOperator *UnaryOp = dyn_cast<UnaryOperator>(Dest)) 3241 if (UnaryOp->getOpcode() == UO_AddrOf) 3242 ActionIdx = 1; // If its an address-of operator, just remove it. 3243 if (Context.getTypeSize(PointeeTy) == Context.getCharWidth()) 3244 ActionIdx = 2; // If the pointee's size is sizeof(char), 3245 // suggest an explicit length. 3246 3247 // If the function is defined as a builtin macro, do not show macro 3248 // expansion. 3249 SourceLocation SL = SizeOfArg->getExprLoc(); 3250 SourceRange DSR = Dest->getSourceRange(); 3251 SourceRange SSR = SizeOfArg->getSourceRange(); 3252 SourceManager &SM = PP.getSourceManager(); 3253 3254 if (SM.isMacroArgExpansion(SL)) { 3255 ReadableName = Lexer::getImmediateMacroName(SL, SM, LangOpts); 3256 SL = SM.getSpellingLoc(SL); 3257 DSR = SourceRange(SM.getSpellingLoc(DSR.getBegin()), 3258 SM.getSpellingLoc(DSR.getEnd())); 3259 SSR = SourceRange(SM.getSpellingLoc(SSR.getBegin()), 3260 SM.getSpellingLoc(SSR.getEnd())); 3261 } 3262 3263 DiagRuntimeBehavior(SL, SizeOfArg, 3264 PDiag(diag::warn_sizeof_pointer_expr_memaccess) 3265 << ReadableName 3266 << PointeeTy 3267 << DestTy 3268 << DSR 3269 << SSR); 3270 DiagRuntimeBehavior(SL, SizeOfArg, 3271 PDiag(diag::warn_sizeof_pointer_expr_memaccess_note) 3272 << ActionIdx 3273 << SSR); 3274 3275 break; 3276 } 3277 } 3278 3279 // Also check for cases where the sizeof argument is the exact same 3280 // type as the memory argument, and where it points to a user-defined 3281 // record type. 3282 if (SizeOfArgTy != QualType()) { 3283 if (PointeeTy->isRecordType() && 3284 Context.typesAreCompatible(SizeOfArgTy, DestTy)) { 3285 DiagRuntimeBehavior(LenExpr->getExprLoc(), Dest, 3286 PDiag(diag::warn_sizeof_pointer_type_memaccess) 3287 << FnName << SizeOfArgTy << ArgIdx 3288 << PointeeTy << Dest->getSourceRange() 3289 << LenExpr->getSourceRange()); 3290 break; 3291 } 3292 } 3293 3294 // Always complain about dynamic classes. 3295 if (isDynamicClassType(PointeeTy)) { 3296 3297 unsigned OperationType = 0; 3298 // "overwritten" if we're warning about the destination for any call 3299 // but memcmp; otherwise a verb appropriate to the call. 3300 if (ArgIdx != 0 || BId == Builtin::BImemcmp) { 3301 if (BId == Builtin::BImemcpy) 3302 OperationType = 1; 3303 else if(BId == Builtin::BImemmove) 3304 OperationType = 2; 3305 else if (BId == Builtin::BImemcmp) 3306 OperationType = 3; 3307 } 3308 3309 DiagRuntimeBehavior( 3310 Dest->getExprLoc(), Dest, 3311 PDiag(diag::warn_dyn_class_memaccess) 3312 << (BId == Builtin::BImemcmp ? ArgIdx + 2 : ArgIdx) 3313 << FnName << PointeeTy 3314 << OperationType 3315 << Call->getCallee()->getSourceRange()); 3316 } else if (PointeeTy.hasNonTrivialObjCLifetime() && 3317 BId != Builtin::BImemset) 3318 DiagRuntimeBehavior( 3319 Dest->getExprLoc(), Dest, 3320 PDiag(diag::warn_arc_object_memaccess) 3321 << ArgIdx << FnName << PointeeTy 3322 << Call->getCallee()->getSourceRange()); 3323 else 3324 continue; 3325 3326 DiagRuntimeBehavior( 3327 Dest->getExprLoc(), Dest, 3328 PDiag(diag::note_bad_memaccess_silence) 3329 << FixItHint::CreateInsertion(ArgRange.getBegin(), "(void*)")); 3330 break; 3331 } 3332 } 3333} 3334 3335// A little helper routine: ignore addition and subtraction of integer literals. 3336// This intentionally does not ignore all integer constant expressions because 3337// we don't want to remove sizeof(). 3338static const Expr *ignoreLiteralAdditions(const Expr *Ex, ASTContext &Ctx) { 3339 Ex = Ex->IgnoreParenCasts(); 3340 3341 for (;;) { 3342 const BinaryOperator * BO = dyn_cast<BinaryOperator>(Ex); 3343 if (!BO || !BO->isAdditiveOp()) 3344 break; 3345 3346 const Expr *RHS = BO->getRHS()->IgnoreParenCasts(); 3347 const Expr *LHS = BO->getLHS()->IgnoreParenCasts(); 3348 3349 if (isa<IntegerLiteral>(RHS)) 3350 Ex = LHS; 3351 else if (isa<IntegerLiteral>(LHS)) 3352 Ex = RHS; 3353 else 3354 break; 3355 } 3356 3357 return Ex; 3358} 3359 3360static bool isConstantSizeArrayWithMoreThanOneElement(QualType Ty, 3361 ASTContext &Context) { 3362 // Only handle constant-sized or VLAs, but not flexible members. 3363 if (const ConstantArrayType *CAT = Context.getAsConstantArrayType(Ty)) { 3364 // Only issue the FIXIT for arrays of size > 1. 3365 if (CAT->getSize().getSExtValue() <= 1) 3366 return false; 3367 } else if (!Ty->isVariableArrayType()) { 3368 return false; 3369 } 3370 return true; 3371} 3372 3373// Warn if the user has made the 'size' argument to strlcpy or strlcat 3374// be the size of the source, instead of the destination. 3375void Sema::CheckStrlcpycatArguments(const CallExpr *Call, 3376 IdentifierInfo *FnName) { 3377 3378 // Don't crash if the user has the wrong number of arguments 3379 if (Call->getNumArgs() != 3) 3380 return; 3381 3382 const Expr *SrcArg = ignoreLiteralAdditions(Call->getArg(1), Context); 3383 const Expr *SizeArg = ignoreLiteralAdditions(Call->getArg(2), Context); 3384 const Expr *CompareWithSrc = NULL; 3385 3386 // Look for 'strlcpy(dst, x, sizeof(x))' 3387 if (const Expr *Ex = getSizeOfExprArg(SizeArg)) 3388 CompareWithSrc = Ex; 3389 else { 3390 // Look for 'strlcpy(dst, x, strlen(x))' 3391 if (const CallExpr *SizeCall = dyn_cast<CallExpr>(SizeArg)) { 3392 if (SizeCall->isBuiltinCall() == Builtin::BIstrlen 3393 && SizeCall->getNumArgs() == 1) 3394 CompareWithSrc = ignoreLiteralAdditions(SizeCall->getArg(0), Context); 3395 } 3396 } 3397 3398 if (!CompareWithSrc) 3399 return; 3400 3401 // Determine if the argument to sizeof/strlen is equal to the source 3402 // argument. In principle there's all kinds of things you could do 3403 // here, for instance creating an == expression and evaluating it with 3404 // EvaluateAsBooleanCondition, but this uses a more direct technique: 3405 const DeclRefExpr *SrcArgDRE = dyn_cast<DeclRefExpr>(SrcArg); 3406 if (!SrcArgDRE) 3407 return; 3408 3409 const DeclRefExpr *CompareWithSrcDRE = dyn_cast<DeclRefExpr>(CompareWithSrc); 3410 if (!CompareWithSrcDRE || 3411 SrcArgDRE->getDecl() != CompareWithSrcDRE->getDecl()) 3412 return; 3413 3414 const Expr *OriginalSizeArg = Call->getArg(2); 3415 Diag(CompareWithSrcDRE->getLocStart(), diag::warn_strlcpycat_wrong_size) 3416 << OriginalSizeArg->getSourceRange() << FnName; 3417 3418 // Output a FIXIT hint if the destination is an array (rather than a 3419 // pointer to an array). This could be enhanced to handle some 3420 // pointers if we know the actual size, like if DstArg is 'array+2' 3421 // we could say 'sizeof(array)-2'. 3422 const Expr *DstArg = Call->getArg(0)->IgnoreParenImpCasts(); 3423 if (!isConstantSizeArrayWithMoreThanOneElement(DstArg->getType(), Context)) 3424 return; 3425 3426 SmallString<128> sizeString; 3427 llvm::raw_svector_ostream OS(sizeString); 3428 OS << "sizeof("; 3429 DstArg->printPretty(OS, 0, getPrintingPolicy()); 3430 OS << ")"; 3431 3432 Diag(OriginalSizeArg->getLocStart(), diag::note_strlcpycat_wrong_size) 3433 << FixItHint::CreateReplacement(OriginalSizeArg->getSourceRange(), 3434 OS.str()); 3435} 3436 3437/// Check if two expressions refer to the same declaration. 3438static bool referToTheSameDecl(const Expr *E1, const Expr *E2) { 3439 if (const DeclRefExpr *D1 = dyn_cast_or_null<DeclRefExpr>(E1)) 3440 if (const DeclRefExpr *D2 = dyn_cast_or_null<DeclRefExpr>(E2)) 3441 return D1->getDecl() == D2->getDecl(); 3442 return false; 3443} 3444 3445static const Expr *getStrlenExprArg(const Expr *E) { 3446 if (const CallExpr *CE = dyn_cast<CallExpr>(E)) { 3447 const FunctionDecl *FD = CE->getDirectCallee(); 3448 if (!FD || FD->getMemoryFunctionKind() != Builtin::BIstrlen) 3449 return 0; 3450 return CE->getArg(0)->IgnoreParenCasts(); 3451 } 3452 return 0; 3453} 3454 3455// Warn on anti-patterns as the 'size' argument to strncat. 3456// The correct size argument should look like following: 3457// strncat(dst, src, sizeof(dst) - strlen(dest) - 1); 3458void Sema::CheckStrncatArguments(const CallExpr *CE, 3459 IdentifierInfo *FnName) { 3460 // Don't crash if the user has the wrong number of arguments. 3461 if (CE->getNumArgs() < 3) 3462 return; 3463 const Expr *DstArg = CE->getArg(0)->IgnoreParenCasts(); 3464 const Expr *SrcArg = CE->getArg(1)->IgnoreParenCasts(); 3465 const Expr *LenArg = CE->getArg(2)->IgnoreParenCasts(); 3466 3467 // Identify common expressions, which are wrongly used as the size argument 3468 // to strncat and may lead to buffer overflows. 3469 unsigned PatternType = 0; 3470 if (const Expr *SizeOfArg = getSizeOfExprArg(LenArg)) { 3471 // - sizeof(dst) 3472 if (referToTheSameDecl(SizeOfArg, DstArg)) 3473 PatternType = 1; 3474 // - sizeof(src) 3475 else if (referToTheSameDecl(SizeOfArg, SrcArg)) 3476 PatternType = 2; 3477 } else if (const BinaryOperator *BE = dyn_cast<BinaryOperator>(LenArg)) { 3478 if (BE->getOpcode() == BO_Sub) { 3479 const Expr *L = BE->getLHS()->IgnoreParenCasts(); 3480 const Expr *R = BE->getRHS()->IgnoreParenCasts(); 3481 // - sizeof(dst) - strlen(dst) 3482 if (referToTheSameDecl(DstArg, getSizeOfExprArg(L)) && 3483 referToTheSameDecl(DstArg, getStrlenExprArg(R))) 3484 PatternType = 1; 3485 // - sizeof(src) - (anything) 3486 else if (referToTheSameDecl(SrcArg, getSizeOfExprArg(L))) 3487 PatternType = 2; 3488 } 3489 } 3490 3491 if (PatternType == 0) 3492 return; 3493 3494 // Generate the diagnostic. 3495 SourceLocation SL = LenArg->getLocStart(); 3496 SourceRange SR = LenArg->getSourceRange(); 3497 SourceManager &SM = PP.getSourceManager(); 3498 3499 // If the function is defined as a builtin macro, do not show macro expansion. 3500 if (SM.isMacroArgExpansion(SL)) { 3501 SL = SM.getSpellingLoc(SL); 3502 SR = SourceRange(SM.getSpellingLoc(SR.getBegin()), 3503 SM.getSpellingLoc(SR.getEnd())); 3504 } 3505 3506 // Check if the destination is an array (rather than a pointer to an array). 3507 QualType DstTy = DstArg->getType(); 3508 bool isKnownSizeArray = isConstantSizeArrayWithMoreThanOneElement(DstTy, 3509 Context); 3510 if (!isKnownSizeArray) { 3511 if (PatternType == 1) 3512 Diag(SL, diag::warn_strncat_wrong_size) << SR; 3513 else 3514 Diag(SL, diag::warn_strncat_src_size) << SR; 3515 return; 3516 } 3517 3518 if (PatternType == 1) 3519 Diag(SL, diag::warn_strncat_large_size) << SR; 3520 else 3521 Diag(SL, diag::warn_strncat_src_size) << SR; 3522 3523 SmallString<128> sizeString; 3524 llvm::raw_svector_ostream OS(sizeString); 3525 OS << "sizeof("; 3526 DstArg->printPretty(OS, 0, getPrintingPolicy()); 3527 OS << ") - "; 3528 OS << "strlen("; 3529 DstArg->printPretty(OS, 0, getPrintingPolicy()); 3530 OS << ") - 1"; 3531 3532 Diag(SL, diag::note_strncat_wrong_size) 3533 << FixItHint::CreateReplacement(SR, OS.str()); 3534} 3535 3536//===--- CHECK: Return Address of Stack Variable --------------------------===// 3537 3538static Expr *EvalVal(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars, 3539 Decl *ParentDecl); 3540static Expr *EvalAddr(Expr* E, SmallVectorImpl<DeclRefExpr *> &refVars, 3541 Decl *ParentDecl); 3542 3543/// CheckReturnStackAddr - Check if a return statement returns the address 3544/// of a stack variable. 3545void 3546Sema::CheckReturnStackAddr(Expr *RetValExp, QualType lhsType, 3547 SourceLocation ReturnLoc) { 3548 3549 Expr *stackE = 0; 3550 SmallVector<DeclRefExpr *, 8> refVars; 3551 3552 // Perform checking for returned stack addresses, local blocks, 3553 // label addresses or references to temporaries. 3554 if (lhsType->isPointerType() || 3555 (!getLangOpts().ObjCAutoRefCount && lhsType->isBlockPointerType())) { 3556 stackE = EvalAddr(RetValExp, refVars, /*ParentDecl=*/0); 3557 } else if (lhsType->isReferenceType()) { 3558 stackE = EvalVal(RetValExp, refVars, /*ParentDecl=*/0); 3559 } 3560 3561 if (stackE == 0) 3562 return; // Nothing suspicious was found. 3563 3564 SourceLocation diagLoc; 3565 SourceRange diagRange; 3566 if (refVars.empty()) { 3567 diagLoc = stackE->getLocStart(); 3568 diagRange = stackE->getSourceRange(); 3569 } else { 3570 // We followed through a reference variable. 'stackE' contains the 3571 // problematic expression but we will warn at the return statement pointing 3572 // at the reference variable. We will later display the "trail" of 3573 // reference variables using notes. 3574 diagLoc = refVars[0]->getLocStart(); 3575 diagRange = refVars[0]->getSourceRange(); 3576 } 3577 3578 if (DeclRefExpr *DR = dyn_cast<DeclRefExpr>(stackE)) { //address of local var. 3579 Diag(diagLoc, lhsType->isReferenceType() ? diag::warn_ret_stack_ref 3580 : diag::warn_ret_stack_addr) 3581 << DR->getDecl()->getDeclName() << diagRange; 3582 } else if (isa<BlockExpr>(stackE)) { // local block. 3583 Diag(diagLoc, diag::err_ret_local_block) << diagRange; 3584 } else if (isa<AddrLabelExpr>(stackE)) { // address of label. 3585 Diag(diagLoc, diag::warn_ret_addr_label) << diagRange; 3586 } else { // local temporary. 3587 Diag(diagLoc, lhsType->isReferenceType() ? diag::warn_ret_local_temp_ref 3588 : diag::warn_ret_local_temp_addr) 3589 << diagRange; 3590 } 3591 3592 // Display the "trail" of reference variables that we followed until we 3593 // found the problematic expression using notes. 3594 for (unsigned i = 0, e = refVars.size(); i != e; ++i) { 3595 VarDecl *VD = cast<VarDecl>(refVars[i]->getDecl()); 3596 // If this var binds to another reference var, show the range of the next 3597 // var, otherwise the var binds to the problematic expression, in which case 3598 // show the range of the expression. 3599 SourceRange range = (i < e-1) ? refVars[i+1]->getSourceRange() 3600 : stackE->getSourceRange(); 3601 Diag(VD->getLocation(), diag::note_ref_var_local_bind) 3602 << VD->getDeclName() << range; 3603 } 3604} 3605 3606/// EvalAddr - EvalAddr and EvalVal are mutually recursive functions that 3607/// check if the expression in a return statement evaluates to an address 3608/// to a location on the stack, a local block, an address of a label, or a 3609/// reference to local temporary. The recursion is used to traverse the 3610/// AST of the return expression, with recursion backtracking when we 3611/// encounter a subexpression that (1) clearly does not lead to one of the 3612/// above problematic expressions (2) is something we cannot determine leads to 3613/// a problematic expression based on such local checking. 3614/// 3615/// Both EvalAddr and EvalVal follow through reference variables to evaluate 3616/// the expression that they point to. Such variables are added to the 3617/// 'refVars' vector so that we know what the reference variable "trail" was. 3618/// 3619/// EvalAddr processes expressions that are pointers that are used as 3620/// references (and not L-values). EvalVal handles all other values. 3621/// At the base case of the recursion is a check for the above problematic 3622/// expressions. 3623/// 3624/// This implementation handles: 3625/// 3626/// * pointer-to-pointer casts 3627/// * implicit conversions from array references to pointers 3628/// * taking the address of fields 3629/// * arbitrary interplay between "&" and "*" operators 3630/// * pointer arithmetic from an address of a stack variable 3631/// * taking the address of an array element where the array is on the stack 3632static Expr *EvalAddr(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars, 3633 Decl *ParentDecl) { 3634 if (E->isTypeDependent()) 3635 return NULL; 3636 3637 // We should only be called for evaluating pointer expressions. 3638 assert((E->getType()->isAnyPointerType() || 3639 E->getType()->isBlockPointerType() || 3640 E->getType()->isObjCQualifiedIdType()) && 3641 "EvalAddr only works on pointers"); 3642 3643 E = E->IgnoreParens(); 3644 3645 // Our "symbolic interpreter" is just a dispatch off the currently 3646 // viewed AST node. We then recursively traverse the AST by calling 3647 // EvalAddr and EvalVal appropriately. 3648 switch (E->getStmtClass()) { 3649 case Stmt::DeclRefExprClass: { 3650 DeclRefExpr *DR = cast<DeclRefExpr>(E); 3651 3652 if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl())) 3653 // If this is a reference variable, follow through to the expression that 3654 // it points to. 3655 if (V->hasLocalStorage() && 3656 V->getType()->isReferenceType() && V->hasInit()) { 3657 // Add the reference variable to the "trail". 3658 refVars.push_back(DR); 3659 return EvalAddr(V->getInit(), refVars, ParentDecl); 3660 } 3661 3662 return NULL; 3663 } 3664 3665 case Stmt::UnaryOperatorClass: { 3666 // The only unary operator that make sense to handle here 3667 // is AddrOf. All others don't make sense as pointers. 3668 UnaryOperator *U = cast<UnaryOperator>(E); 3669 3670 if (U->getOpcode() == UO_AddrOf) 3671 return EvalVal(U->getSubExpr(), refVars, ParentDecl); 3672 else 3673 return NULL; 3674 } 3675 3676 case Stmt::BinaryOperatorClass: { 3677 // Handle pointer arithmetic. All other binary operators are not valid 3678 // in this context. 3679 BinaryOperator *B = cast<BinaryOperator>(E); 3680 BinaryOperatorKind op = B->getOpcode(); 3681 3682 if (op != BO_Add && op != BO_Sub) 3683 return NULL; 3684 3685 Expr *Base = B->getLHS(); 3686 3687 // Determine which argument is the real pointer base. It could be 3688 // the RHS argument instead of the LHS. 3689 if (!Base->getType()->isPointerType()) Base = B->getRHS(); 3690 3691 assert (Base->getType()->isPointerType()); 3692 return EvalAddr(Base, refVars, ParentDecl); 3693 } 3694 3695 // For conditional operators we need to see if either the LHS or RHS are 3696 // valid DeclRefExpr*s. If one of them is valid, we return it. 3697 case Stmt::ConditionalOperatorClass: { 3698 ConditionalOperator *C = cast<ConditionalOperator>(E); 3699 3700 // Handle the GNU extension for missing LHS. 3701 if (Expr *lhsExpr = C->getLHS()) { 3702 // In C++, we can have a throw-expression, which has 'void' type. 3703 if (!lhsExpr->getType()->isVoidType()) 3704 if (Expr* LHS = EvalAddr(lhsExpr, refVars, ParentDecl)) 3705 return LHS; 3706 } 3707 3708 // In C++, we can have a throw-expression, which has 'void' type. 3709 if (C->getRHS()->getType()->isVoidType()) 3710 return NULL; 3711 3712 return EvalAddr(C->getRHS(), refVars, ParentDecl); 3713 } 3714 3715 case Stmt::BlockExprClass: 3716 if (cast<BlockExpr>(E)->getBlockDecl()->hasCaptures()) 3717 return E; // local block. 3718 return NULL; 3719 3720 case Stmt::AddrLabelExprClass: 3721 return E; // address of label. 3722 3723 case Stmt::ExprWithCleanupsClass: 3724 return EvalAddr(cast<ExprWithCleanups>(E)->getSubExpr(), refVars, 3725 ParentDecl); 3726 3727 // For casts, we need to handle conversions from arrays to 3728 // pointer values, and pointer-to-pointer conversions. 3729 case Stmt::ImplicitCastExprClass: 3730 case Stmt::CStyleCastExprClass: 3731 case Stmt::CXXFunctionalCastExprClass: 3732 case Stmt::ObjCBridgedCastExprClass: 3733 case Stmt::CXXStaticCastExprClass: 3734 case Stmt::CXXDynamicCastExprClass: 3735 case Stmt::CXXConstCastExprClass: 3736 case Stmt::CXXReinterpretCastExprClass: { 3737 Expr* SubExpr = cast<CastExpr>(E)->getSubExpr(); 3738 switch (cast<CastExpr>(E)->getCastKind()) { 3739 case CK_BitCast: 3740 case CK_LValueToRValue: 3741 case CK_NoOp: 3742 case CK_BaseToDerived: 3743 case CK_DerivedToBase: 3744 case CK_UncheckedDerivedToBase: 3745 case CK_Dynamic: 3746 case CK_CPointerToObjCPointerCast: 3747 case CK_BlockPointerToObjCPointerCast: 3748 case CK_AnyPointerToBlockPointerCast: 3749 return EvalAddr(SubExpr, refVars, ParentDecl); 3750 3751 case CK_ArrayToPointerDecay: 3752 return EvalVal(SubExpr, refVars, ParentDecl); 3753 3754 default: 3755 return 0; 3756 } 3757 } 3758 3759 case Stmt::MaterializeTemporaryExprClass: 3760 if (Expr *Result = EvalAddr( 3761 cast<MaterializeTemporaryExpr>(E)->GetTemporaryExpr(), 3762 refVars, ParentDecl)) 3763 return Result; 3764 3765 return E; 3766 3767 // Everything else: we simply don't reason about them. 3768 default: 3769 return NULL; 3770 } 3771} 3772 3773 3774/// EvalVal - This function is complements EvalAddr in the mutual recursion. 3775/// See the comments for EvalAddr for more details. 3776static Expr *EvalVal(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars, 3777 Decl *ParentDecl) { 3778do { 3779 // We should only be called for evaluating non-pointer expressions, or 3780 // expressions with a pointer type that are not used as references but instead 3781 // are l-values (e.g., DeclRefExpr with a pointer type). 3782 3783 // Our "symbolic interpreter" is just a dispatch off the currently 3784 // viewed AST node. We then recursively traverse the AST by calling 3785 // EvalAddr and EvalVal appropriately. 3786 3787 E = E->IgnoreParens(); 3788 switch (E->getStmtClass()) { 3789 case Stmt::ImplicitCastExprClass: { 3790 ImplicitCastExpr *IE = cast<ImplicitCastExpr>(E); 3791 if (IE->getValueKind() == VK_LValue) { 3792 E = IE->getSubExpr(); 3793 continue; 3794 } 3795 return NULL; 3796 } 3797 3798 case Stmt::ExprWithCleanupsClass: 3799 return EvalVal(cast<ExprWithCleanups>(E)->getSubExpr(), refVars,ParentDecl); 3800 3801 case Stmt::DeclRefExprClass: { 3802 // When we hit a DeclRefExpr we are looking at code that refers to a 3803 // variable's name. If it's not a reference variable we check if it has 3804 // local storage within the function, and if so, return the expression. 3805 DeclRefExpr *DR = cast<DeclRefExpr>(E); 3806 3807 if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl())) { 3808 // Check if it refers to itself, e.g. "int& i = i;". 3809 if (V == ParentDecl) 3810 return DR; 3811 3812 if (V->hasLocalStorage()) { 3813 if (!V->getType()->isReferenceType()) 3814 return DR; 3815 3816 // Reference variable, follow through to the expression that 3817 // it points to. 3818 if (V->hasInit()) { 3819 // Add the reference variable to the "trail". 3820 refVars.push_back(DR); 3821 return EvalVal(V->getInit(), refVars, V); 3822 } 3823 } 3824 } 3825 3826 return NULL; 3827 } 3828 3829 case Stmt::UnaryOperatorClass: { 3830 // The only unary operator that make sense to handle here 3831 // is Deref. All others don't resolve to a "name." This includes 3832 // handling all sorts of rvalues passed to a unary operator. 3833 UnaryOperator *U = cast<UnaryOperator>(E); 3834 3835 if (U->getOpcode() == UO_Deref) 3836 return EvalAddr(U->getSubExpr(), refVars, ParentDecl); 3837 3838 return NULL; 3839 } 3840 3841 case Stmt::ArraySubscriptExprClass: { 3842 // Array subscripts are potential references to data on the stack. We 3843 // retrieve the DeclRefExpr* for the array variable if it indeed 3844 // has local storage. 3845 return EvalAddr(cast<ArraySubscriptExpr>(E)->getBase(), refVars,ParentDecl); 3846 } 3847 3848 case Stmt::ConditionalOperatorClass: { 3849 // For conditional operators we need to see if either the LHS or RHS are 3850 // non-NULL Expr's. If one is non-NULL, we return it. 3851 ConditionalOperator *C = cast<ConditionalOperator>(E); 3852 3853 // Handle the GNU extension for missing LHS. 3854 if (Expr *lhsExpr = C->getLHS()) 3855 if (Expr *LHS = EvalVal(lhsExpr, refVars, ParentDecl)) 3856 return LHS; 3857 3858 return EvalVal(C->getRHS(), refVars, ParentDecl); 3859 } 3860 3861 // Accesses to members are potential references to data on the stack. 3862 case Stmt::MemberExprClass: { 3863 MemberExpr *M = cast<MemberExpr>(E); 3864 3865 // Check for indirect access. We only want direct field accesses. 3866 if (M->isArrow()) 3867 return NULL; 3868 3869 // Check whether the member type is itself a reference, in which case 3870 // we're not going to refer to the member, but to what the member refers to. 3871 if (M->getMemberDecl()->getType()->isReferenceType()) 3872 return NULL; 3873 3874 return EvalVal(M->getBase(), refVars, ParentDecl); 3875 } 3876 3877 case Stmt::MaterializeTemporaryExprClass: 3878 if (Expr *Result = EvalVal( 3879 cast<MaterializeTemporaryExpr>(E)->GetTemporaryExpr(), 3880 refVars, ParentDecl)) 3881 return Result; 3882 3883 return E; 3884 3885 default: 3886 // Check that we don't return or take the address of a reference to a 3887 // temporary. This is only useful in C++. 3888 if (!E->isTypeDependent() && E->isRValue()) 3889 return E; 3890 3891 // Everything else: we simply don't reason about them. 3892 return NULL; 3893 } 3894} while (true); 3895} 3896 3897//===--- CHECK: Floating-Point comparisons (-Wfloat-equal) ---------------===// 3898 3899/// Check for comparisons of floating point operands using != and ==. 3900/// Issue a warning if these are no self-comparisons, as they are not likely 3901/// to do what the programmer intended. 3902void Sema::CheckFloatComparison(SourceLocation Loc, Expr* LHS, Expr *RHS) { 3903 Expr* LeftExprSansParen = LHS->IgnoreParenImpCasts(); 3904 Expr* RightExprSansParen = RHS->IgnoreParenImpCasts(); 3905 3906 // Special case: check for x == x (which is OK). 3907 // Do not emit warnings for such cases. 3908 if (DeclRefExpr* DRL = dyn_cast<DeclRefExpr>(LeftExprSansParen)) 3909 if (DeclRefExpr* DRR = dyn_cast<DeclRefExpr>(RightExprSansParen)) 3910 if (DRL->getDecl() == DRR->getDecl()) 3911 return; 3912 3913 3914 // Special case: check for comparisons against literals that can be exactly 3915 // represented by APFloat. In such cases, do not emit a warning. This 3916 // is a heuristic: often comparison against such literals are used to 3917 // detect if a value in a variable has not changed. This clearly can 3918 // lead to false negatives. 3919 if (FloatingLiteral* FLL = dyn_cast<FloatingLiteral>(LeftExprSansParen)) { 3920 if (FLL->isExact()) 3921 return; 3922 } else 3923 if (FloatingLiteral* FLR = dyn_cast<FloatingLiteral>(RightExprSansParen)) 3924 if (FLR->isExact()) 3925 return; 3926 3927 // Check for comparisons with builtin types. 3928 if (CallExpr* CL = dyn_cast<CallExpr>(LeftExprSansParen)) 3929 if (CL->isBuiltinCall()) 3930 return; 3931 3932 if (CallExpr* CR = dyn_cast<CallExpr>(RightExprSansParen)) 3933 if (CR->isBuiltinCall()) 3934 return; 3935 3936 // Emit the diagnostic. 3937 Diag(Loc, diag::warn_floatingpoint_eq) 3938 << LHS->getSourceRange() << RHS->getSourceRange(); 3939} 3940 3941//===--- CHECK: Integer mixed-sign comparisons (-Wsign-compare) --------===// 3942//===--- CHECK: Lossy implicit conversions (-Wconversion) --------------===// 3943 3944namespace { 3945 3946/// Structure recording the 'active' range of an integer-valued 3947/// expression. 3948struct IntRange { 3949 /// The number of bits active in the int. 3950 unsigned Width; 3951 3952 /// True if the int is known not to have negative values. 3953 bool NonNegative; 3954 3955 IntRange(unsigned Width, bool NonNegative) 3956 : Width(Width), NonNegative(NonNegative) 3957 {} 3958 3959 /// Returns the range of the bool type. 3960 static IntRange forBoolType() { 3961 return IntRange(1, true); 3962 } 3963 3964 /// Returns the range of an opaque value of the given integral type. 3965 static IntRange forValueOfType(ASTContext &C, QualType T) { 3966 return forValueOfCanonicalType(C, 3967 T->getCanonicalTypeInternal().getTypePtr()); 3968 } 3969 3970 /// Returns the range of an opaque value of a canonical integral type. 3971 static IntRange forValueOfCanonicalType(ASTContext &C, const Type *T) { 3972 assert(T->isCanonicalUnqualified()); 3973 3974 if (const VectorType *VT = dyn_cast<VectorType>(T)) 3975 T = VT->getElementType().getTypePtr(); 3976 if (const ComplexType *CT = dyn_cast<ComplexType>(T)) 3977 T = CT->getElementType().getTypePtr(); 3978 3979 // For enum types, use the known bit width of the enumerators. 3980 if (const EnumType *ET = dyn_cast<EnumType>(T)) { 3981 EnumDecl *Enum = ET->getDecl(); 3982 if (!Enum->isCompleteDefinition()) 3983 return IntRange(C.getIntWidth(QualType(T, 0)), false); 3984 3985 unsigned NumPositive = Enum->getNumPositiveBits(); 3986 unsigned NumNegative = Enum->getNumNegativeBits(); 3987 3988 if (NumNegative == 0) 3989 return IntRange(NumPositive, true/*NonNegative*/); 3990 else 3991 return IntRange(std::max(NumPositive + 1, NumNegative), 3992 false/*NonNegative*/); 3993 } 3994 3995 const BuiltinType *BT = cast<BuiltinType>(T); 3996 assert(BT->isInteger()); 3997 3998 return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger()); 3999 } 4000 4001 /// Returns the "target" range of a canonical integral type, i.e. 4002 /// the range of values expressible in the type. 4003 /// 4004 /// This matches forValueOfCanonicalType except that enums have the 4005 /// full range of their type, not the range of their enumerators. 4006 static IntRange forTargetOfCanonicalType(ASTContext &C, const Type *T) { 4007 assert(T->isCanonicalUnqualified()); 4008 4009 if (const VectorType *VT = dyn_cast<VectorType>(T)) 4010 T = VT->getElementType().getTypePtr(); 4011 if (const ComplexType *CT = dyn_cast<ComplexType>(T)) 4012 T = CT->getElementType().getTypePtr(); 4013 if (const EnumType *ET = dyn_cast<EnumType>(T)) 4014 T = C.getCanonicalType(ET->getDecl()->getIntegerType()).getTypePtr(); 4015 4016 const BuiltinType *BT = cast<BuiltinType>(T); 4017 assert(BT->isInteger()); 4018 4019 return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger()); 4020 } 4021 4022 /// Returns the supremum of two ranges: i.e. their conservative merge. 4023 static IntRange join(IntRange L, IntRange R) { 4024 return IntRange(std::max(L.Width, R.Width), 4025 L.NonNegative && R.NonNegative); 4026 } 4027 4028 /// Returns the infinum of two ranges: i.e. their aggressive merge. 4029 static IntRange meet(IntRange L, IntRange R) { 4030 return IntRange(std::min(L.Width, R.Width), 4031 L.NonNegative || R.NonNegative); 4032 } 4033}; 4034 4035static IntRange GetValueRange(ASTContext &C, llvm::APSInt &value, 4036 unsigned MaxWidth) { 4037 if (value.isSigned() && value.isNegative()) 4038 return IntRange(value.getMinSignedBits(), false); 4039 4040 if (value.getBitWidth() > MaxWidth) 4041 value = value.trunc(MaxWidth); 4042 4043 // isNonNegative() just checks the sign bit without considering 4044 // signedness. 4045 return IntRange(value.getActiveBits(), true); 4046} 4047 4048static IntRange GetValueRange(ASTContext &C, APValue &result, QualType Ty, 4049 unsigned MaxWidth) { 4050 if (result.isInt()) 4051 return GetValueRange(C, result.getInt(), MaxWidth); 4052 4053 if (result.isVector()) { 4054 IntRange R = GetValueRange(C, result.getVectorElt(0), Ty, MaxWidth); 4055 for (unsigned i = 1, e = result.getVectorLength(); i != e; ++i) { 4056 IntRange El = GetValueRange(C, result.getVectorElt(i), Ty, MaxWidth); 4057 R = IntRange::join(R, El); 4058 } 4059 return R; 4060 } 4061 4062 if (result.isComplexInt()) { 4063 IntRange R = GetValueRange(C, result.getComplexIntReal(), MaxWidth); 4064 IntRange I = GetValueRange(C, result.getComplexIntImag(), MaxWidth); 4065 return IntRange::join(R, I); 4066 } 4067 4068 // This can happen with lossless casts to intptr_t of "based" lvalues. 4069 // Assume it might use arbitrary bits. 4070 // FIXME: The only reason we need to pass the type in here is to get 4071 // the sign right on this one case. It would be nice if APValue 4072 // preserved this. 4073 assert(result.isLValue() || result.isAddrLabelDiff()); 4074 return IntRange(MaxWidth, Ty->isUnsignedIntegerOrEnumerationType()); 4075} 4076 4077/// Pseudo-evaluate the given integer expression, estimating the 4078/// range of values it might take. 4079/// 4080/// \param MaxWidth - the width to which the value will be truncated 4081static IntRange GetExprRange(ASTContext &C, Expr *E, unsigned MaxWidth) { 4082 E = E->IgnoreParens(); 4083 4084 // Try a full evaluation first. 4085 Expr::EvalResult result; 4086 if (E->EvaluateAsRValue(result, C)) 4087 return GetValueRange(C, result.Val, E->getType(), MaxWidth); 4088 4089 // I think we only want to look through implicit casts here; if the 4090 // user has an explicit widening cast, we should treat the value as 4091 // being of the new, wider type. 4092 if (ImplicitCastExpr *CE = dyn_cast<ImplicitCastExpr>(E)) { 4093 if (CE->getCastKind() == CK_NoOp || CE->getCastKind() == CK_LValueToRValue) 4094 return GetExprRange(C, CE->getSubExpr(), MaxWidth); 4095 4096 IntRange OutputTypeRange = IntRange::forValueOfType(C, CE->getType()); 4097 4098 bool isIntegerCast = (CE->getCastKind() == CK_IntegralCast); 4099 4100 // Assume that non-integer casts can span the full range of the type. 4101 if (!isIntegerCast) 4102 return OutputTypeRange; 4103 4104 IntRange SubRange 4105 = GetExprRange(C, CE->getSubExpr(), 4106 std::min(MaxWidth, OutputTypeRange.Width)); 4107 4108 // Bail out if the subexpr's range is as wide as the cast type. 4109 if (SubRange.Width >= OutputTypeRange.Width) 4110 return OutputTypeRange; 4111 4112 // Otherwise, we take the smaller width, and we're non-negative if 4113 // either the output type or the subexpr is. 4114 return IntRange(SubRange.Width, 4115 SubRange.NonNegative || OutputTypeRange.NonNegative); 4116 } 4117 4118 if (ConditionalOperator *CO = dyn_cast<ConditionalOperator>(E)) { 4119 // If we can fold the condition, just take that operand. 4120 bool CondResult; 4121 if (CO->getCond()->EvaluateAsBooleanCondition(CondResult, C)) 4122 return GetExprRange(C, CondResult ? CO->getTrueExpr() 4123 : CO->getFalseExpr(), 4124 MaxWidth); 4125 4126 // Otherwise, conservatively merge. 4127 IntRange L = GetExprRange(C, CO->getTrueExpr(), MaxWidth); 4128 IntRange R = GetExprRange(C, CO->getFalseExpr(), MaxWidth); 4129 return IntRange::join(L, R); 4130 } 4131 4132 if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) { 4133 switch (BO->getOpcode()) { 4134 4135 // Boolean-valued operations are single-bit and positive. 4136 case BO_LAnd: 4137 case BO_LOr: 4138 case BO_LT: 4139 case BO_GT: 4140 case BO_LE: 4141 case BO_GE: 4142 case BO_EQ: 4143 case BO_NE: 4144 return IntRange::forBoolType(); 4145 4146 // The type of the assignments is the type of the LHS, so the RHS 4147 // is not necessarily the same type. 4148 case BO_MulAssign: 4149 case BO_DivAssign: 4150 case BO_RemAssign: 4151 case BO_AddAssign: 4152 case BO_SubAssign: 4153 case BO_XorAssign: 4154 case BO_OrAssign: 4155 // TODO: bitfields? 4156 return IntRange::forValueOfType(C, E->getType()); 4157 4158 // Simple assignments just pass through the RHS, which will have 4159 // been coerced to the LHS type. 4160 case BO_Assign: 4161 // TODO: bitfields? 4162 return GetExprRange(C, BO->getRHS(), MaxWidth); 4163 4164 // Operations with opaque sources are black-listed. 4165 case BO_PtrMemD: 4166 case BO_PtrMemI: 4167 return IntRange::forValueOfType(C, E->getType()); 4168 4169 // Bitwise-and uses the *infinum* of the two source ranges. 4170 case BO_And: 4171 case BO_AndAssign: 4172 return IntRange::meet(GetExprRange(C, BO->getLHS(), MaxWidth), 4173 GetExprRange(C, BO->getRHS(), MaxWidth)); 4174 4175 // Left shift gets black-listed based on a judgement call. 4176 case BO_Shl: 4177 // ...except that we want to treat '1 << (blah)' as logically 4178 // positive. It's an important idiom. 4179 if (IntegerLiteral *I 4180 = dyn_cast<IntegerLiteral>(BO->getLHS()->IgnoreParenCasts())) { 4181 if (I->getValue() == 1) { 4182 IntRange R = IntRange::forValueOfType(C, E->getType()); 4183 return IntRange(R.Width, /*NonNegative*/ true); 4184 } 4185 } 4186 // fallthrough 4187 4188 case BO_ShlAssign: 4189 return IntRange::forValueOfType(C, E->getType()); 4190 4191 // Right shift by a constant can narrow its left argument. 4192 case BO_Shr: 4193 case BO_ShrAssign: { 4194 IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth); 4195 4196 // If the shift amount is a positive constant, drop the width by 4197 // that much. 4198 llvm::APSInt shift; 4199 if (BO->getRHS()->isIntegerConstantExpr(shift, C) && 4200 shift.isNonNegative()) { 4201 unsigned zext = shift.getZExtValue(); 4202 if (zext >= L.Width) 4203 L.Width = (L.NonNegative ? 0 : 1); 4204 else 4205 L.Width -= zext; 4206 } 4207 4208 return L; 4209 } 4210 4211 // Comma acts as its right operand. 4212 case BO_Comma: 4213 return GetExprRange(C, BO->getRHS(), MaxWidth); 4214 4215 // Black-list pointer subtractions. 4216 case BO_Sub: 4217 if (BO->getLHS()->getType()->isPointerType()) 4218 return IntRange::forValueOfType(C, E->getType()); 4219 break; 4220 4221 // The width of a division result is mostly determined by the size 4222 // of the LHS. 4223 case BO_Div: { 4224 // Don't 'pre-truncate' the operands. 4225 unsigned opWidth = C.getIntWidth(E->getType()); 4226 IntRange L = GetExprRange(C, BO->getLHS(), opWidth); 4227 4228 // If the divisor is constant, use that. 4229 llvm::APSInt divisor; 4230 if (BO->getRHS()->isIntegerConstantExpr(divisor, C)) { 4231 unsigned log2 = divisor.logBase2(); // floor(log_2(divisor)) 4232 if (log2 >= L.Width) 4233 L.Width = (L.NonNegative ? 0 : 1); 4234 else 4235 L.Width = std::min(L.Width - log2, MaxWidth); 4236 return L; 4237 } 4238 4239 // Otherwise, just use the LHS's width. 4240 IntRange R = GetExprRange(C, BO->getRHS(), opWidth); 4241 return IntRange(L.Width, L.NonNegative && R.NonNegative); 4242 } 4243 4244 // The result of a remainder can't be larger than the result of 4245 // either side. 4246 case BO_Rem: { 4247 // Don't 'pre-truncate' the operands. 4248 unsigned opWidth = C.getIntWidth(E->getType()); 4249 IntRange L = GetExprRange(C, BO->getLHS(), opWidth); 4250 IntRange R = GetExprRange(C, BO->getRHS(), opWidth); 4251 4252 IntRange meet = IntRange::meet(L, R); 4253 meet.Width = std::min(meet.Width, MaxWidth); 4254 return meet; 4255 } 4256 4257 // The default behavior is okay for these. 4258 case BO_Mul: 4259 case BO_Add: 4260 case BO_Xor: 4261 case BO_Or: 4262 break; 4263 } 4264 4265 // The default case is to treat the operation as if it were closed 4266 // on the narrowest type that encompasses both operands. 4267 IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth); 4268 IntRange R = GetExprRange(C, BO->getRHS(), MaxWidth); 4269 return IntRange::join(L, R); 4270 } 4271 4272 if (UnaryOperator *UO = dyn_cast<UnaryOperator>(E)) { 4273 switch (UO->getOpcode()) { 4274 // Boolean-valued operations are white-listed. 4275 case UO_LNot: 4276 return IntRange::forBoolType(); 4277 4278 // Operations with opaque sources are black-listed. 4279 case UO_Deref: 4280 case UO_AddrOf: // should be impossible 4281 return IntRange::forValueOfType(C, E->getType()); 4282 4283 default: 4284 return GetExprRange(C, UO->getSubExpr(), MaxWidth); 4285 } 4286 } 4287 4288 if (dyn_cast<OffsetOfExpr>(E)) { 4289 IntRange::forValueOfType(C, E->getType()); 4290 } 4291 4292 if (FieldDecl *BitField = E->getBitField()) 4293 return IntRange(BitField->getBitWidthValue(C), 4294 BitField->getType()->isUnsignedIntegerOrEnumerationType()); 4295 4296 return IntRange::forValueOfType(C, E->getType()); 4297} 4298 4299static IntRange GetExprRange(ASTContext &C, Expr *E) { 4300 return GetExprRange(C, E, C.getIntWidth(E->getType())); 4301} 4302 4303/// Checks whether the given value, which currently has the given 4304/// source semantics, has the same value when coerced through the 4305/// target semantics. 4306static bool IsSameFloatAfterCast(const llvm::APFloat &value, 4307 const llvm::fltSemantics &Src, 4308 const llvm::fltSemantics &Tgt) { 4309 llvm::APFloat truncated = value; 4310 4311 bool ignored; 4312 truncated.convert(Src, llvm::APFloat::rmNearestTiesToEven, &ignored); 4313 truncated.convert(Tgt, llvm::APFloat::rmNearestTiesToEven, &ignored); 4314 4315 return truncated.bitwiseIsEqual(value); 4316} 4317 4318/// Checks whether the given value, which currently has the given 4319/// source semantics, has the same value when coerced through the 4320/// target semantics. 4321/// 4322/// The value might be a vector of floats (or a complex number). 4323static bool IsSameFloatAfterCast(const APValue &value, 4324 const llvm::fltSemantics &Src, 4325 const llvm::fltSemantics &Tgt) { 4326 if (value.isFloat()) 4327 return IsSameFloatAfterCast(value.getFloat(), Src, Tgt); 4328 4329 if (value.isVector()) { 4330 for (unsigned i = 0, e = value.getVectorLength(); i != e; ++i) 4331 if (!IsSameFloatAfterCast(value.getVectorElt(i), Src, Tgt)) 4332 return false; 4333 return true; 4334 } 4335 4336 assert(value.isComplexFloat()); 4337 return (IsSameFloatAfterCast(value.getComplexFloatReal(), Src, Tgt) && 4338 IsSameFloatAfterCast(value.getComplexFloatImag(), Src, Tgt)); 4339} 4340 4341static void AnalyzeImplicitConversions(Sema &S, Expr *E, SourceLocation CC); 4342 4343static bool IsZero(Sema &S, Expr *E) { 4344 // Suppress cases where we are comparing against an enum constant. 4345 if (const DeclRefExpr *DR = 4346 dyn_cast<DeclRefExpr>(E->IgnoreParenImpCasts())) 4347 if (isa<EnumConstantDecl>(DR->getDecl())) 4348 return false; 4349 4350 // Suppress cases where the '0' value is expanded from a macro. 4351 if (E->getLocStart().isMacroID()) 4352 return false; 4353 4354 llvm::APSInt Value; 4355 return E->isIntegerConstantExpr(Value, S.Context) && Value == 0; 4356} 4357 4358static bool HasEnumType(Expr *E) { 4359 // Strip off implicit integral promotions. 4360 while (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(E)) { 4361 if (ICE->getCastKind() != CK_IntegralCast && 4362 ICE->getCastKind() != CK_NoOp) 4363 break; 4364 E = ICE->getSubExpr(); 4365 } 4366 4367 return E->getType()->isEnumeralType(); 4368} 4369 4370static void CheckTrivialUnsignedComparison(Sema &S, BinaryOperator *E) { 4371 BinaryOperatorKind op = E->getOpcode(); 4372 if (E->isValueDependent()) 4373 return; 4374 4375 if (op == BO_LT && IsZero(S, E->getRHS())) { 4376 S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison) 4377 << "< 0" << "false" << HasEnumType(E->getLHS()) 4378 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 4379 } else if (op == BO_GE && IsZero(S, E->getRHS())) { 4380 S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison) 4381 << ">= 0" << "true" << HasEnumType(E->getLHS()) 4382 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 4383 } else if (op == BO_GT && IsZero(S, E->getLHS())) { 4384 S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison) 4385 << "0 >" << "false" << HasEnumType(E->getRHS()) 4386 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 4387 } else if (op == BO_LE && IsZero(S, E->getLHS())) { 4388 S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison) 4389 << "0 <=" << "true" << HasEnumType(E->getRHS()) 4390 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 4391 } 4392} 4393 4394static void DiagnoseOutOfRangeComparison(Sema &S, BinaryOperator *E, 4395 Expr *Constant, Expr *Other, 4396 llvm::APSInt Value, 4397 bool RhsConstant) { 4398 // 0 values are handled later by CheckTrivialUnsignedComparison(). 4399 if (Value == 0) 4400 return; 4401 4402 BinaryOperatorKind op = E->getOpcode(); 4403 QualType OtherT = Other->getType(); 4404 QualType ConstantT = Constant->getType(); 4405 QualType CommonT = E->getLHS()->getType(); 4406 if (S.Context.hasSameUnqualifiedType(OtherT, ConstantT)) 4407 return; 4408 assert((OtherT->isIntegerType() && ConstantT->isIntegerType()) 4409 && "comparison with non-integer type"); 4410 4411 bool ConstantSigned = ConstantT->isSignedIntegerType(); 4412 bool CommonSigned = CommonT->isSignedIntegerType(); 4413 4414 bool EqualityOnly = false; 4415 4416 // TODO: Investigate using GetExprRange() to get tighter bounds on 4417 // on the bit ranges. 4418 IntRange OtherRange = IntRange::forValueOfType(S.Context, OtherT); 4419 unsigned OtherWidth = OtherRange.Width; 4420 4421 if (CommonSigned) { 4422 // The common type is signed, therefore no signed to unsigned conversion. 4423 if (!OtherRange.NonNegative) { 4424 // Check that the constant is representable in type OtherT. 4425 if (ConstantSigned) { 4426 if (OtherWidth >= Value.getMinSignedBits()) 4427 return; 4428 } else { // !ConstantSigned 4429 if (OtherWidth >= Value.getActiveBits() + 1) 4430 return; 4431 } 4432 } else { // !OtherSigned 4433 // Check that the constant is representable in type OtherT. 4434 // Negative values are out of range. 4435 if (ConstantSigned) { 4436 if (Value.isNonNegative() && OtherWidth >= Value.getActiveBits()) 4437 return; 4438 } else { // !ConstantSigned 4439 if (OtherWidth >= Value.getActiveBits()) 4440 return; 4441 } 4442 } 4443 } else { // !CommonSigned 4444 if (OtherRange.NonNegative) { 4445 if (OtherWidth >= Value.getActiveBits()) 4446 return; 4447 } else if (!OtherRange.NonNegative && !ConstantSigned) { 4448 // Check to see if the constant is representable in OtherT. 4449 if (OtherWidth > Value.getActiveBits()) 4450 return; 4451 // Check to see if the constant is equivalent to a negative value 4452 // cast to CommonT. 4453 if (S.Context.getIntWidth(ConstantT) == S.Context.getIntWidth(CommonT) && 4454 Value.isNegative() && Value.getMinSignedBits() <= OtherWidth) 4455 return; 4456 // The constant value rests between values that OtherT can represent after 4457 // conversion. Relational comparison still works, but equality 4458 // comparisons will be tautological. 4459 EqualityOnly = true; 4460 } else { // OtherSigned && ConstantSigned 4461 assert(0 && "Two signed types converted to unsigned types."); 4462 } 4463 } 4464 4465 bool PositiveConstant = !ConstantSigned || Value.isNonNegative(); 4466 4467 bool IsTrue = true; 4468 if (op == BO_EQ || op == BO_NE) { 4469 IsTrue = op == BO_NE; 4470 } else if (EqualityOnly) { 4471 return; 4472 } else if (RhsConstant) { 4473 if (op == BO_GT || op == BO_GE) 4474 IsTrue = !PositiveConstant; 4475 else // op == BO_LT || op == BO_LE 4476 IsTrue = PositiveConstant; 4477 } else { 4478 if (op == BO_LT || op == BO_LE) 4479 IsTrue = !PositiveConstant; 4480 else // op == BO_GT || op == BO_GE 4481 IsTrue = PositiveConstant; 4482 } 4483 SmallString<16> PrettySourceValue(Value.toString(10)); 4484 S.Diag(E->getOperatorLoc(), diag::warn_out_of_range_compare) 4485 << PrettySourceValue << OtherT << IsTrue 4486 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 4487} 4488 4489/// Analyze the operands of the given comparison. Implements the 4490/// fallback case from AnalyzeComparison. 4491static void AnalyzeImpConvsInComparison(Sema &S, BinaryOperator *E) { 4492 AnalyzeImplicitConversions(S, E->getLHS(), E->getOperatorLoc()); 4493 AnalyzeImplicitConversions(S, E->getRHS(), E->getOperatorLoc()); 4494} 4495 4496/// \brief Implements -Wsign-compare. 4497/// 4498/// \param E the binary operator to check for warnings 4499static void AnalyzeComparison(Sema &S, BinaryOperator *E) { 4500 // The type the comparison is being performed in. 4501 QualType T = E->getLHS()->getType(); 4502 assert(S.Context.hasSameUnqualifiedType(T, E->getRHS()->getType()) 4503 && "comparison with mismatched types"); 4504 if (E->isValueDependent()) 4505 return AnalyzeImpConvsInComparison(S, E); 4506 4507 Expr *LHS = E->getLHS()->IgnoreParenImpCasts(); 4508 Expr *RHS = E->getRHS()->IgnoreParenImpCasts(); 4509 4510 bool IsComparisonConstant = false; 4511 4512 // Check whether an integer constant comparison results in a value 4513 // of 'true' or 'false'. 4514 if (T->isIntegralType(S.Context)) { 4515 llvm::APSInt RHSValue; 4516 bool IsRHSIntegralLiteral = 4517 RHS->isIntegerConstantExpr(RHSValue, S.Context); 4518 llvm::APSInt LHSValue; 4519 bool IsLHSIntegralLiteral = 4520 LHS->isIntegerConstantExpr(LHSValue, S.Context); 4521 if (IsRHSIntegralLiteral && !IsLHSIntegralLiteral) 4522 DiagnoseOutOfRangeComparison(S, E, RHS, LHS, RHSValue, true); 4523 else if (!IsRHSIntegralLiteral && IsLHSIntegralLiteral) 4524 DiagnoseOutOfRangeComparison(S, E, LHS, RHS, LHSValue, false); 4525 else 4526 IsComparisonConstant = 4527 (IsRHSIntegralLiteral && IsLHSIntegralLiteral); 4528 } else if (!T->hasUnsignedIntegerRepresentation()) 4529 IsComparisonConstant = E->isIntegerConstantExpr(S.Context); 4530 4531 // We don't do anything special if this isn't an unsigned integral 4532 // comparison: we're only interested in integral comparisons, and 4533 // signed comparisons only happen in cases we don't care to warn about. 4534 // 4535 // We also don't care about value-dependent expressions or expressions 4536 // whose result is a constant. 4537 if (!T->hasUnsignedIntegerRepresentation() || IsComparisonConstant) 4538 return AnalyzeImpConvsInComparison(S, E); 4539 4540 // Check to see if one of the (unmodified) operands is of different 4541 // signedness. 4542 Expr *signedOperand, *unsignedOperand; 4543 if (LHS->getType()->hasSignedIntegerRepresentation()) { 4544 assert(!RHS->getType()->hasSignedIntegerRepresentation() && 4545 "unsigned comparison between two signed integer expressions?"); 4546 signedOperand = LHS; 4547 unsignedOperand = RHS; 4548 } else if (RHS->getType()->hasSignedIntegerRepresentation()) { 4549 signedOperand = RHS; 4550 unsignedOperand = LHS; 4551 } else { 4552 CheckTrivialUnsignedComparison(S, E); 4553 return AnalyzeImpConvsInComparison(S, E); 4554 } 4555 4556 // Otherwise, calculate the effective range of the signed operand. 4557 IntRange signedRange = GetExprRange(S.Context, signedOperand); 4558 4559 // Go ahead and analyze implicit conversions in the operands. Note 4560 // that we skip the implicit conversions on both sides. 4561 AnalyzeImplicitConversions(S, LHS, E->getOperatorLoc()); 4562 AnalyzeImplicitConversions(S, RHS, E->getOperatorLoc()); 4563 4564 // If the signed range is non-negative, -Wsign-compare won't fire, 4565 // but we should still check for comparisons which are always true 4566 // or false. 4567 if (signedRange.NonNegative) 4568 return CheckTrivialUnsignedComparison(S, E); 4569 4570 // For (in)equality comparisons, if the unsigned operand is a 4571 // constant which cannot collide with a overflowed signed operand, 4572 // then reinterpreting the signed operand as unsigned will not 4573 // change the result of the comparison. 4574 if (E->isEqualityOp()) { 4575 unsigned comparisonWidth = S.Context.getIntWidth(T); 4576 IntRange unsignedRange = GetExprRange(S.Context, unsignedOperand); 4577 4578 // We should never be unable to prove that the unsigned operand is 4579 // non-negative. 4580 assert(unsignedRange.NonNegative && "unsigned range includes negative?"); 4581 4582 if (unsignedRange.Width < comparisonWidth) 4583 return; 4584 } 4585 4586 S.DiagRuntimeBehavior(E->getOperatorLoc(), E, 4587 S.PDiag(diag::warn_mixed_sign_comparison) 4588 << LHS->getType() << RHS->getType() 4589 << LHS->getSourceRange() << RHS->getSourceRange()); 4590} 4591 4592/// Analyzes an attempt to assign the given value to a bitfield. 4593/// 4594/// Returns true if there was something fishy about the attempt. 4595static bool AnalyzeBitFieldAssignment(Sema &S, FieldDecl *Bitfield, Expr *Init, 4596 SourceLocation InitLoc) { 4597 assert(Bitfield->isBitField()); 4598 if (Bitfield->isInvalidDecl()) 4599 return false; 4600 4601 // White-list bool bitfields. 4602 if (Bitfield->getType()->isBooleanType()) 4603 return false; 4604 4605 // Ignore value- or type-dependent expressions. 4606 if (Bitfield->getBitWidth()->isValueDependent() || 4607 Bitfield->getBitWidth()->isTypeDependent() || 4608 Init->isValueDependent() || 4609 Init->isTypeDependent()) 4610 return false; 4611 4612 Expr *OriginalInit = Init->IgnoreParenImpCasts(); 4613 4614 llvm::APSInt Value; 4615 if (!OriginalInit->EvaluateAsInt(Value, S.Context, Expr::SE_AllowSideEffects)) 4616 return false; 4617 4618 unsigned OriginalWidth = Value.getBitWidth(); 4619 unsigned FieldWidth = Bitfield->getBitWidthValue(S.Context); 4620 4621 if (OriginalWidth <= FieldWidth) 4622 return false; 4623 4624 // Compute the value which the bitfield will contain. 4625 llvm::APSInt TruncatedValue = Value.trunc(FieldWidth); 4626 TruncatedValue.setIsSigned(Bitfield->getType()->isSignedIntegerType()); 4627 4628 // Check whether the stored value is equal to the original value. 4629 TruncatedValue = TruncatedValue.extend(OriginalWidth); 4630 if (llvm::APSInt::isSameValue(Value, TruncatedValue)) 4631 return false; 4632 4633 // Special-case bitfields of width 1: booleans are naturally 0/1, and 4634 // therefore don't strictly fit into a signed bitfield of width 1. 4635 if (FieldWidth == 1 && Value == 1) 4636 return false; 4637 4638 std::string PrettyValue = Value.toString(10); 4639 std::string PrettyTrunc = TruncatedValue.toString(10); 4640 4641 S.Diag(InitLoc, diag::warn_impcast_bitfield_precision_constant) 4642 << PrettyValue << PrettyTrunc << OriginalInit->getType() 4643 << Init->getSourceRange(); 4644 4645 return true; 4646} 4647 4648/// Analyze the given simple or compound assignment for warning-worthy 4649/// operations. 4650static void AnalyzeAssignment(Sema &S, BinaryOperator *E) { 4651 // Just recurse on the LHS. 4652 AnalyzeImplicitConversions(S, E->getLHS(), E->getOperatorLoc()); 4653 4654 // We want to recurse on the RHS as normal unless we're assigning to 4655 // a bitfield. 4656 if (FieldDecl *Bitfield = E->getLHS()->getBitField()) { 4657 if (AnalyzeBitFieldAssignment(S, Bitfield, E->getRHS(), 4658 E->getOperatorLoc())) { 4659 // Recurse, ignoring any implicit conversions on the RHS. 4660 return AnalyzeImplicitConversions(S, E->getRHS()->IgnoreParenImpCasts(), 4661 E->getOperatorLoc()); 4662 } 4663 } 4664 4665 AnalyzeImplicitConversions(S, E->getRHS(), E->getOperatorLoc()); 4666} 4667 4668/// Diagnose an implicit cast; purely a helper for CheckImplicitConversion. 4669static void DiagnoseImpCast(Sema &S, Expr *E, QualType SourceType, QualType T, 4670 SourceLocation CContext, unsigned diag, 4671 bool pruneControlFlow = false) { 4672 if (pruneControlFlow) { 4673 S.DiagRuntimeBehavior(E->getExprLoc(), E, 4674 S.PDiag(diag) 4675 << SourceType << T << E->getSourceRange() 4676 << SourceRange(CContext)); 4677 return; 4678 } 4679 S.Diag(E->getExprLoc(), diag) 4680 << SourceType << T << E->getSourceRange() << SourceRange(CContext); 4681} 4682 4683/// Diagnose an implicit cast; purely a helper for CheckImplicitConversion. 4684static void DiagnoseImpCast(Sema &S, Expr *E, QualType T, 4685 SourceLocation CContext, unsigned diag, 4686 bool pruneControlFlow = false) { 4687 DiagnoseImpCast(S, E, E->getType(), T, CContext, diag, pruneControlFlow); 4688} 4689 4690/// Diagnose an implicit cast from a literal expression. Does not warn when the 4691/// cast wouldn't lose information. 4692void DiagnoseFloatingLiteralImpCast(Sema &S, FloatingLiteral *FL, QualType T, 4693 SourceLocation CContext) { 4694 // Try to convert the literal exactly to an integer. If we can, don't warn. 4695 bool isExact = false; 4696 const llvm::APFloat &Value = FL->getValue(); 4697 llvm::APSInt IntegerValue(S.Context.getIntWidth(T), 4698 T->hasUnsignedIntegerRepresentation()); 4699 if (Value.convertToInteger(IntegerValue, 4700 llvm::APFloat::rmTowardZero, &isExact) 4701 == llvm::APFloat::opOK && isExact) 4702 return; 4703 4704 SmallString<16> PrettySourceValue; 4705 Value.toString(PrettySourceValue); 4706 SmallString<16> PrettyTargetValue; 4707 if (T->isSpecificBuiltinType(BuiltinType::Bool)) 4708 PrettyTargetValue = IntegerValue == 0 ? "false" : "true"; 4709 else 4710 IntegerValue.toString(PrettyTargetValue); 4711 4712 S.Diag(FL->getExprLoc(), diag::warn_impcast_literal_float_to_integer) 4713 << FL->getType() << T.getUnqualifiedType() << PrettySourceValue 4714 << PrettyTargetValue << FL->getSourceRange() << SourceRange(CContext); 4715} 4716 4717std::string PrettyPrintInRange(const llvm::APSInt &Value, IntRange Range) { 4718 if (!Range.Width) return "0"; 4719 4720 llvm::APSInt ValueInRange = Value; 4721 ValueInRange.setIsSigned(!Range.NonNegative); 4722 ValueInRange = ValueInRange.trunc(Range.Width); 4723 return ValueInRange.toString(10); 4724} 4725 4726static bool IsImplicitBoolFloatConversion(Sema &S, Expr *Ex, bool ToBool) { 4727 if (!isa<ImplicitCastExpr>(Ex)) 4728 return false; 4729 4730 Expr *InnerE = Ex->IgnoreParenImpCasts(); 4731 const Type *Target = S.Context.getCanonicalType(Ex->getType()).getTypePtr(); 4732 const Type *Source = 4733 S.Context.getCanonicalType(InnerE->getType()).getTypePtr(); 4734 if (Target->isDependentType()) 4735 return false; 4736 4737 const BuiltinType *FloatCandidateBT = 4738 dyn_cast<BuiltinType>(ToBool ? Source : Target); 4739 const Type *BoolCandidateType = ToBool ? Target : Source; 4740 4741 return (BoolCandidateType->isSpecificBuiltinType(BuiltinType::Bool) && 4742 FloatCandidateBT && (FloatCandidateBT->isFloatingPoint())); 4743} 4744 4745void CheckImplicitArgumentConversions(Sema &S, CallExpr *TheCall, 4746 SourceLocation CC) { 4747 unsigned NumArgs = TheCall->getNumArgs(); 4748 for (unsigned i = 0; i < NumArgs; ++i) { 4749 Expr *CurrA = TheCall->getArg(i); 4750 if (!IsImplicitBoolFloatConversion(S, CurrA, true)) 4751 continue; 4752 4753 bool IsSwapped = ((i > 0) && 4754 IsImplicitBoolFloatConversion(S, TheCall->getArg(i - 1), false)); 4755 IsSwapped |= ((i < (NumArgs - 1)) && 4756 IsImplicitBoolFloatConversion(S, TheCall->getArg(i + 1), false)); 4757 if (IsSwapped) { 4758 // Warn on this floating-point to bool conversion. 4759 DiagnoseImpCast(S, CurrA->IgnoreParenImpCasts(), 4760 CurrA->getType(), CC, 4761 diag::warn_impcast_floating_point_to_bool); 4762 } 4763 } 4764} 4765 4766void CheckImplicitConversion(Sema &S, Expr *E, QualType T, 4767 SourceLocation CC, bool *ICContext = 0) { 4768 if (E->isTypeDependent() || E->isValueDependent()) return; 4769 4770 const Type *Source = S.Context.getCanonicalType(E->getType()).getTypePtr(); 4771 const Type *Target = S.Context.getCanonicalType(T).getTypePtr(); 4772 if (Source == Target) return; 4773 if (Target->isDependentType()) return; 4774 4775 // If the conversion context location is invalid don't complain. We also 4776 // don't want to emit a warning if the issue occurs from the expansion of 4777 // a system macro. The problem is that 'getSpellingLoc()' is slow, so we 4778 // delay this check as long as possible. Once we detect we are in that 4779 // scenario, we just return. 4780 if (CC.isInvalid()) 4781 return; 4782 4783 // Diagnose implicit casts to bool. 4784 if (Target->isSpecificBuiltinType(BuiltinType::Bool)) { 4785 if (isa<StringLiteral>(E)) 4786 // Warn on string literal to bool. Checks for string literals in logical 4787 // expressions, for instances, assert(0 && "error here"), is prevented 4788 // by a check in AnalyzeImplicitConversions(). 4789 return DiagnoseImpCast(S, E, T, CC, 4790 diag::warn_impcast_string_literal_to_bool); 4791 if (Source->isFunctionType()) { 4792 // Warn on function to bool. Checks free functions and static member 4793 // functions. Weakly imported functions are excluded from the check, 4794 // since it's common to test their value to check whether the linker 4795 // found a definition for them. 4796 ValueDecl *D = 0; 4797 if (DeclRefExpr* R = dyn_cast<DeclRefExpr>(E)) { 4798 D = R->getDecl(); 4799 } else if (MemberExpr *M = dyn_cast<MemberExpr>(E)) { 4800 D = M->getMemberDecl(); 4801 } 4802 4803 if (D && !D->isWeak()) { 4804 if (FunctionDecl* F = dyn_cast<FunctionDecl>(D)) { 4805 S.Diag(E->getExprLoc(), diag::warn_impcast_function_to_bool) 4806 << F << E->getSourceRange() << SourceRange(CC); 4807 S.Diag(E->getExprLoc(), diag::note_function_to_bool_silence) 4808 << FixItHint::CreateInsertion(E->getExprLoc(), "&"); 4809 QualType ReturnType; 4810 UnresolvedSet<4> NonTemplateOverloads; 4811 S.isExprCallable(*E, ReturnType, NonTemplateOverloads); 4812 if (!ReturnType.isNull() 4813 && ReturnType->isSpecificBuiltinType(BuiltinType::Bool)) 4814 S.Diag(E->getExprLoc(), diag::note_function_to_bool_call) 4815 << FixItHint::CreateInsertion( 4816 S.getPreprocessor().getLocForEndOfToken(E->getLocEnd()), "()"); 4817 return; 4818 } 4819 } 4820 } 4821 } 4822 4823 // Strip vector types. 4824 if (isa<VectorType>(Source)) { 4825 if (!isa<VectorType>(Target)) { 4826 if (S.SourceMgr.isInSystemMacro(CC)) 4827 return; 4828 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_vector_scalar); 4829 } 4830 4831 // If the vector cast is cast between two vectors of the same size, it is 4832 // a bitcast, not a conversion. 4833 if (S.Context.getTypeSize(Source) == S.Context.getTypeSize(Target)) 4834 return; 4835 4836 Source = cast<VectorType>(Source)->getElementType().getTypePtr(); 4837 Target = cast<VectorType>(Target)->getElementType().getTypePtr(); 4838 } 4839 4840 // Strip complex types. 4841 if (isa<ComplexType>(Source)) { 4842 if (!isa<ComplexType>(Target)) { 4843 if (S.SourceMgr.isInSystemMacro(CC)) 4844 return; 4845 4846 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_complex_scalar); 4847 } 4848 4849 Source = cast<ComplexType>(Source)->getElementType().getTypePtr(); 4850 Target = cast<ComplexType>(Target)->getElementType().getTypePtr(); 4851 } 4852 4853 const BuiltinType *SourceBT = dyn_cast<BuiltinType>(Source); 4854 const BuiltinType *TargetBT = dyn_cast<BuiltinType>(Target); 4855 4856 // If the source is floating point... 4857 if (SourceBT && SourceBT->isFloatingPoint()) { 4858 // ...and the target is floating point... 4859 if (TargetBT && TargetBT->isFloatingPoint()) { 4860 // ...then warn if we're dropping FP rank. 4861 4862 // Builtin FP kinds are ordered by increasing FP rank. 4863 if (SourceBT->getKind() > TargetBT->getKind()) { 4864 // Don't warn about float constants that are precisely 4865 // representable in the target type. 4866 Expr::EvalResult result; 4867 if (E->EvaluateAsRValue(result, S.Context)) { 4868 // Value might be a float, a float vector, or a float complex. 4869 if (IsSameFloatAfterCast(result.Val, 4870 S.Context.getFloatTypeSemantics(QualType(TargetBT, 0)), 4871 S.Context.getFloatTypeSemantics(QualType(SourceBT, 0)))) 4872 return; 4873 } 4874 4875 if (S.SourceMgr.isInSystemMacro(CC)) 4876 return; 4877 4878 DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_float_precision); 4879 } 4880 return; 4881 } 4882 4883 // If the target is integral, always warn. 4884 if (TargetBT && TargetBT->isInteger()) { 4885 if (S.SourceMgr.isInSystemMacro(CC)) 4886 return; 4887 4888 Expr *InnerE = E->IgnoreParenImpCasts(); 4889 // We also want to warn on, e.g., "int i = -1.234" 4890 if (UnaryOperator *UOp = dyn_cast<UnaryOperator>(InnerE)) 4891 if (UOp->getOpcode() == UO_Minus || UOp->getOpcode() == UO_Plus) 4892 InnerE = UOp->getSubExpr()->IgnoreParenImpCasts(); 4893 4894 if (FloatingLiteral *FL = dyn_cast<FloatingLiteral>(InnerE)) { 4895 DiagnoseFloatingLiteralImpCast(S, FL, T, CC); 4896 } else { 4897 DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_float_integer); 4898 } 4899 } 4900 4901 // If the target is bool, warn if expr is a function or method call. 4902 if (Target->isSpecificBuiltinType(BuiltinType::Bool) && 4903 isa<CallExpr>(E)) { 4904 // Check last argument of function call to see if it is an 4905 // implicit cast from a type matching the type the result 4906 // is being cast to. 4907 CallExpr *CEx = cast<CallExpr>(E); 4908 unsigned NumArgs = CEx->getNumArgs(); 4909 if (NumArgs > 0) { 4910 Expr *LastA = CEx->getArg(NumArgs - 1); 4911 Expr *InnerE = LastA->IgnoreParenImpCasts(); 4912 const Type *InnerType = 4913 S.Context.getCanonicalType(InnerE->getType()).getTypePtr(); 4914 if (isa<ImplicitCastExpr>(LastA) && (InnerType == Target)) { 4915 // Warn on this floating-point to bool conversion 4916 DiagnoseImpCast(S, E, T, CC, 4917 diag::warn_impcast_floating_point_to_bool); 4918 } 4919 } 4920 } 4921 return; 4922 } 4923 4924 if ((E->isNullPointerConstant(S.Context, Expr::NPC_ValueDependentIsNotNull) 4925 == Expr::NPCK_GNUNull) && !Target->isAnyPointerType() 4926 && !Target->isBlockPointerType() && !Target->isMemberPointerType() 4927 && Target->isScalarType()) { 4928 SourceLocation Loc = E->getSourceRange().getBegin(); 4929 if (Loc.isMacroID()) 4930 Loc = S.SourceMgr.getImmediateExpansionRange(Loc).first; 4931 if (!Loc.isMacroID() || CC.isMacroID()) 4932 S.Diag(Loc, diag::warn_impcast_null_pointer_to_integer) 4933 << T << clang::SourceRange(CC) 4934 << FixItHint::CreateReplacement(Loc, S.getFixItZeroLiteralForType(T)); 4935 } 4936 4937 if (!Source->isIntegerType() || !Target->isIntegerType()) 4938 return; 4939 4940 // TODO: remove this early return once the false positives for constant->bool 4941 // in templates, macros, etc, are reduced or removed. 4942 if (Target->isSpecificBuiltinType(BuiltinType::Bool)) 4943 return; 4944 4945 IntRange SourceRange = GetExprRange(S.Context, E); 4946 IntRange TargetRange = IntRange::forTargetOfCanonicalType(S.Context, Target); 4947 4948 if (SourceRange.Width > TargetRange.Width) { 4949 // If the source is a constant, use a default-on diagnostic. 4950 // TODO: this should happen for bitfield stores, too. 4951 llvm::APSInt Value(32); 4952 if (E->isIntegerConstantExpr(Value, S.Context)) { 4953 if (S.SourceMgr.isInSystemMacro(CC)) 4954 return; 4955 4956 std::string PrettySourceValue = Value.toString(10); 4957 std::string PrettyTargetValue = PrettyPrintInRange(Value, TargetRange); 4958 4959 S.DiagRuntimeBehavior(E->getExprLoc(), E, 4960 S.PDiag(diag::warn_impcast_integer_precision_constant) 4961 << PrettySourceValue << PrettyTargetValue 4962 << E->getType() << T << E->getSourceRange() 4963 << clang::SourceRange(CC)); 4964 return; 4965 } 4966 4967 // People want to build with -Wshorten-64-to-32 and not -Wconversion. 4968 if (S.SourceMgr.isInSystemMacro(CC)) 4969 return; 4970 4971 if (TargetRange.Width == 32 && S.Context.getIntWidth(E->getType()) == 64) 4972 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_integer_64_32, 4973 /* pruneControlFlow */ true); 4974 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_integer_precision); 4975 } 4976 4977 if ((TargetRange.NonNegative && !SourceRange.NonNegative) || 4978 (!TargetRange.NonNegative && SourceRange.NonNegative && 4979 SourceRange.Width == TargetRange.Width)) { 4980 4981 if (S.SourceMgr.isInSystemMacro(CC)) 4982 return; 4983 4984 unsigned DiagID = diag::warn_impcast_integer_sign; 4985 4986 // Traditionally, gcc has warned about this under -Wsign-compare. 4987 // We also want to warn about it in -Wconversion. 4988 // So if -Wconversion is off, use a completely identical diagnostic 4989 // in the sign-compare group. 4990 // The conditional-checking code will 4991 if (ICContext) { 4992 DiagID = diag::warn_impcast_integer_sign_conditional; 4993 *ICContext = true; 4994 } 4995 4996 return DiagnoseImpCast(S, E, T, CC, DiagID); 4997 } 4998 4999 // Diagnose conversions between different enumeration types. 5000 // In C, we pretend that the type of an EnumConstantDecl is its enumeration 5001 // type, to give us better diagnostics. 5002 QualType SourceType = E->getType(); 5003 if (!S.getLangOpts().CPlusPlus) { 5004 if (DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(E)) 5005 if (EnumConstantDecl *ECD = dyn_cast<EnumConstantDecl>(DRE->getDecl())) { 5006 EnumDecl *Enum = cast<EnumDecl>(ECD->getDeclContext()); 5007 SourceType = S.Context.getTypeDeclType(Enum); 5008 Source = S.Context.getCanonicalType(SourceType).getTypePtr(); 5009 } 5010 } 5011 5012 if (const EnumType *SourceEnum = Source->getAs<EnumType>()) 5013 if (const EnumType *TargetEnum = Target->getAs<EnumType>()) 5014 if ((SourceEnum->getDecl()->getIdentifier() || 5015 SourceEnum->getDecl()->getTypedefNameForAnonDecl()) && 5016 (TargetEnum->getDecl()->getIdentifier() || 5017 TargetEnum->getDecl()->getTypedefNameForAnonDecl()) && 5018 SourceEnum != TargetEnum) { 5019 if (S.SourceMgr.isInSystemMacro(CC)) 5020 return; 5021 5022 return DiagnoseImpCast(S, E, SourceType, T, CC, 5023 diag::warn_impcast_different_enum_types); 5024 } 5025 5026 return; 5027} 5028 5029void CheckConditionalOperator(Sema &S, ConditionalOperator *E, 5030 SourceLocation CC, QualType T); 5031 5032void CheckConditionalOperand(Sema &S, Expr *E, QualType T, 5033 SourceLocation CC, bool &ICContext) { 5034 E = E->IgnoreParenImpCasts(); 5035 5036 if (isa<ConditionalOperator>(E)) 5037 return CheckConditionalOperator(S, cast<ConditionalOperator>(E), CC, T); 5038 5039 AnalyzeImplicitConversions(S, E, CC); 5040 if (E->getType() != T) 5041 return CheckImplicitConversion(S, E, T, CC, &ICContext); 5042 return; 5043} 5044 5045void CheckConditionalOperator(Sema &S, ConditionalOperator *E, 5046 SourceLocation CC, QualType T) { 5047 AnalyzeImplicitConversions(S, E->getCond(), CC); 5048 5049 bool Suspicious = false; 5050 CheckConditionalOperand(S, E->getTrueExpr(), T, CC, Suspicious); 5051 CheckConditionalOperand(S, E->getFalseExpr(), T, CC, Suspicious); 5052 5053 // If -Wconversion would have warned about either of the candidates 5054 // for a signedness conversion to the context type... 5055 if (!Suspicious) return; 5056 5057 // ...but it's currently ignored... 5058 if (S.Diags.getDiagnosticLevel(diag::warn_impcast_integer_sign_conditional, 5059 CC)) 5060 return; 5061 5062 // ...then check whether it would have warned about either of the 5063 // candidates for a signedness conversion to the condition type. 5064 if (E->getType() == T) return; 5065 5066 Suspicious = false; 5067 CheckImplicitConversion(S, E->getTrueExpr()->IgnoreParenImpCasts(), 5068 E->getType(), CC, &Suspicious); 5069 if (!Suspicious) 5070 CheckImplicitConversion(S, E->getFalseExpr()->IgnoreParenImpCasts(), 5071 E->getType(), CC, &Suspicious); 5072} 5073 5074/// AnalyzeImplicitConversions - Find and report any interesting 5075/// implicit conversions in the given expression. There are a couple 5076/// of competing diagnostics here, -Wconversion and -Wsign-compare. 5077void AnalyzeImplicitConversions(Sema &S, Expr *OrigE, SourceLocation CC) { 5078 QualType T = OrigE->getType(); 5079 Expr *E = OrigE->IgnoreParenImpCasts(); 5080 5081 if (E->isTypeDependent() || E->isValueDependent()) 5082 return; 5083 5084 // For conditional operators, we analyze the arguments as if they 5085 // were being fed directly into the output. 5086 if (isa<ConditionalOperator>(E)) { 5087 ConditionalOperator *CO = cast<ConditionalOperator>(E); 5088 CheckConditionalOperator(S, CO, CC, T); 5089 return; 5090 } 5091 5092 // Check implicit argument conversions for function calls. 5093 if (CallExpr *Call = dyn_cast<CallExpr>(E)) 5094 CheckImplicitArgumentConversions(S, Call, CC); 5095 5096 // Go ahead and check any implicit conversions we might have skipped. 5097 // The non-canonical typecheck is just an optimization; 5098 // CheckImplicitConversion will filter out dead implicit conversions. 5099 if (E->getType() != T) 5100 CheckImplicitConversion(S, E, T, CC); 5101 5102 // Now continue drilling into this expression. 5103 5104 // Skip past explicit casts. 5105 if (isa<ExplicitCastExpr>(E)) { 5106 E = cast<ExplicitCastExpr>(E)->getSubExpr()->IgnoreParenImpCasts(); 5107 return AnalyzeImplicitConversions(S, E, CC); 5108 } 5109 5110 if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) { 5111 // Do a somewhat different check with comparison operators. 5112 if (BO->isComparisonOp()) 5113 return AnalyzeComparison(S, BO); 5114 5115 // And with simple assignments. 5116 if (BO->getOpcode() == BO_Assign) 5117 return AnalyzeAssignment(S, BO); 5118 } 5119 5120 // These break the otherwise-useful invariant below. Fortunately, 5121 // we don't really need to recurse into them, because any internal 5122 // expressions should have been analyzed already when they were 5123 // built into statements. 5124 if (isa<StmtExpr>(E)) return; 5125 5126 // Don't descend into unevaluated contexts. 5127 if (isa<UnaryExprOrTypeTraitExpr>(E)) return; 5128 5129 // Now just recurse over the expression's children. 5130 CC = E->getExprLoc(); 5131 BinaryOperator *BO = dyn_cast<BinaryOperator>(E); 5132 bool IsLogicalOperator = BO && BO->isLogicalOp(); 5133 for (Stmt::child_range I = E->children(); I; ++I) { 5134 Expr *ChildExpr = dyn_cast_or_null<Expr>(*I); 5135 if (!ChildExpr) 5136 continue; 5137 5138 if (IsLogicalOperator && 5139 isa<StringLiteral>(ChildExpr->IgnoreParenImpCasts())) 5140 // Ignore checking string literals that are in logical operators. 5141 continue; 5142 AnalyzeImplicitConversions(S, ChildExpr, CC); 5143 } 5144} 5145 5146} // end anonymous namespace 5147 5148/// Diagnoses "dangerous" implicit conversions within the given 5149/// expression (which is a full expression). Implements -Wconversion 5150/// and -Wsign-compare. 5151/// 5152/// \param CC the "context" location of the implicit conversion, i.e. 5153/// the most location of the syntactic entity requiring the implicit 5154/// conversion 5155void Sema::CheckImplicitConversions(Expr *E, SourceLocation CC) { 5156 // Don't diagnose in unevaluated contexts. 5157 if (isUnevaluatedContext()) 5158 return; 5159 5160 // Don't diagnose for value- or type-dependent expressions. 5161 if (E->isTypeDependent() || E->isValueDependent()) 5162 return; 5163 5164 // Check for array bounds violations in cases where the check isn't triggered 5165 // elsewhere for other Expr types (like BinaryOperators), e.g. when an 5166 // ArraySubscriptExpr is on the RHS of a variable initialization. 5167 CheckArrayAccess(E); 5168 5169 // This is not the right CC for (e.g.) a variable initialization. 5170 AnalyzeImplicitConversions(*this, E, CC); 5171} 5172 5173void Sema::CheckBitFieldInitialization(SourceLocation InitLoc, 5174 FieldDecl *BitField, 5175 Expr *Init) { 5176 (void) AnalyzeBitFieldAssignment(*this, BitField, Init, InitLoc); 5177} 5178 5179/// CheckParmsForFunctionDef - Check that the parameters of the given 5180/// function are appropriate for the definition of a function. This 5181/// takes care of any checks that cannot be performed on the 5182/// declaration itself, e.g., that the types of each of the function 5183/// parameters are complete. 5184bool Sema::CheckParmsForFunctionDef(ParmVarDecl **P, ParmVarDecl **PEnd, 5185 bool CheckParameterNames) { 5186 bool HasInvalidParm = false; 5187 for (; P != PEnd; ++P) { 5188 ParmVarDecl *Param = *P; 5189 5190 // C99 6.7.5.3p4: the parameters in a parameter type list in a 5191 // function declarator that is part of a function definition of 5192 // that function shall not have incomplete type. 5193 // 5194 // This is also C++ [dcl.fct]p6. 5195 if (!Param->isInvalidDecl() && 5196 RequireCompleteType(Param->getLocation(), Param->getType(), 5197 diag::err_typecheck_decl_incomplete_type)) { 5198 Param->setInvalidDecl(); 5199 HasInvalidParm = true; 5200 } 5201 5202 // C99 6.9.1p5: If the declarator includes a parameter type list, the 5203 // declaration of each parameter shall include an identifier. 5204 if (CheckParameterNames && 5205 Param->getIdentifier() == 0 && 5206 !Param->isImplicit() && 5207 !getLangOpts().CPlusPlus) 5208 Diag(Param->getLocation(), diag::err_parameter_name_omitted); 5209 5210 // C99 6.7.5.3p12: 5211 // If the function declarator is not part of a definition of that 5212 // function, parameters may have incomplete type and may use the [*] 5213 // notation in their sequences of declarator specifiers to specify 5214 // variable length array types. 5215 QualType PType = Param->getOriginalType(); 5216 if (const ArrayType *AT = Context.getAsArrayType(PType)) { 5217 if (AT->getSizeModifier() == ArrayType::Star) { 5218 // FIXME: This diagnosic should point the '[*]' if source-location 5219 // information is added for it. 5220 Diag(Param->getLocation(), diag::err_array_star_in_function_definition); 5221 } 5222 } 5223 } 5224 5225 return HasInvalidParm; 5226} 5227 5228/// CheckCastAlign - Implements -Wcast-align, which warns when a 5229/// pointer cast increases the alignment requirements. 5230void Sema::CheckCastAlign(Expr *Op, QualType T, SourceRange TRange) { 5231 // This is actually a lot of work to potentially be doing on every 5232 // cast; don't do it if we're ignoring -Wcast_align (as is the default). 5233 if (getDiagnostics().getDiagnosticLevel(diag::warn_cast_align, 5234 TRange.getBegin()) 5235 == DiagnosticsEngine::Ignored) 5236 return; 5237 5238 // Ignore dependent types. 5239 if (T->isDependentType() || Op->getType()->isDependentType()) 5240 return; 5241 5242 // Require that the destination be a pointer type. 5243 const PointerType *DestPtr = T->getAs<PointerType>(); 5244 if (!DestPtr) return; 5245 5246 // If the destination has alignment 1, we're done. 5247 QualType DestPointee = DestPtr->getPointeeType(); 5248 if (DestPointee->isIncompleteType()) return; 5249 CharUnits DestAlign = Context.getTypeAlignInChars(DestPointee); 5250 if (DestAlign.isOne()) return; 5251 5252 // Require that the source be a pointer type. 5253 const PointerType *SrcPtr = Op->getType()->getAs<PointerType>(); 5254 if (!SrcPtr) return; 5255 QualType SrcPointee = SrcPtr->getPointeeType(); 5256 5257 // Whitelist casts from cv void*. We already implicitly 5258 // whitelisted casts to cv void*, since they have alignment 1. 5259 // Also whitelist casts involving incomplete types, which implicitly 5260 // includes 'void'. 5261 if (SrcPointee->isIncompleteType()) return; 5262 5263 CharUnits SrcAlign = Context.getTypeAlignInChars(SrcPointee); 5264 if (SrcAlign >= DestAlign) return; 5265 5266 Diag(TRange.getBegin(), diag::warn_cast_align) 5267 << Op->getType() << T 5268 << static_cast<unsigned>(SrcAlign.getQuantity()) 5269 << static_cast<unsigned>(DestAlign.getQuantity()) 5270 << TRange << Op->getSourceRange(); 5271} 5272 5273static const Type* getElementType(const Expr *BaseExpr) { 5274 const Type* EltType = BaseExpr->getType().getTypePtr(); 5275 if (EltType->isAnyPointerType()) 5276 return EltType->getPointeeType().getTypePtr(); 5277 else if (EltType->isArrayType()) 5278 return EltType->getBaseElementTypeUnsafe(); 5279 return EltType; 5280} 5281 5282/// \brief Check whether this array fits the idiom of a size-one tail padded 5283/// array member of a struct. 5284/// 5285/// We avoid emitting out-of-bounds access warnings for such arrays as they are 5286/// commonly used to emulate flexible arrays in C89 code. 5287static bool IsTailPaddedMemberArray(Sema &S, llvm::APInt Size, 5288 const NamedDecl *ND) { 5289 if (Size != 1 || !ND) return false; 5290 5291 const FieldDecl *FD = dyn_cast<FieldDecl>(ND); 5292 if (!FD) return false; 5293 5294 // Don't consider sizes resulting from macro expansions or template argument 5295 // substitution to form C89 tail-padded arrays. 5296 5297 TypeSourceInfo *TInfo = FD->getTypeSourceInfo(); 5298 while (TInfo) { 5299 TypeLoc TL = TInfo->getTypeLoc(); 5300 // Look through typedefs. 5301 const TypedefTypeLoc *TTL = dyn_cast<TypedefTypeLoc>(&TL); 5302 if (TTL) { 5303 const TypedefNameDecl *TDL = TTL->getTypedefNameDecl(); 5304 TInfo = TDL->getTypeSourceInfo(); 5305 continue; 5306 } 5307 ConstantArrayTypeLoc CTL = cast<ConstantArrayTypeLoc>(TL); 5308 const Expr *SizeExpr = dyn_cast<IntegerLiteral>(CTL.getSizeExpr()); 5309 if (!SizeExpr || SizeExpr->getExprLoc().isMacroID()) 5310 return false; 5311 break; 5312 } 5313 5314 const RecordDecl *RD = dyn_cast<RecordDecl>(FD->getDeclContext()); 5315 if (!RD) return false; 5316 if (RD->isUnion()) return false; 5317 if (const CXXRecordDecl *CRD = dyn_cast<CXXRecordDecl>(RD)) { 5318 if (!CRD->isStandardLayout()) return false; 5319 } 5320 5321 // See if this is the last field decl in the record. 5322 const Decl *D = FD; 5323 while ((D = D->getNextDeclInContext())) 5324 if (isa<FieldDecl>(D)) 5325 return false; 5326 return true; 5327} 5328 5329void Sema::CheckArrayAccess(const Expr *BaseExpr, const Expr *IndexExpr, 5330 const ArraySubscriptExpr *ASE, 5331 bool AllowOnePastEnd, bool IndexNegated) { 5332 IndexExpr = IndexExpr->IgnoreParenImpCasts(); 5333 if (IndexExpr->isValueDependent()) 5334 return; 5335 5336 const Type *EffectiveType = getElementType(BaseExpr); 5337 BaseExpr = BaseExpr->IgnoreParenCasts(); 5338 const ConstantArrayType *ArrayTy = 5339 Context.getAsConstantArrayType(BaseExpr->getType()); 5340 if (!ArrayTy) 5341 return; 5342 5343 llvm::APSInt index; 5344 if (!IndexExpr->EvaluateAsInt(index, Context)) 5345 return; 5346 if (IndexNegated) 5347 index = -index; 5348 5349 const NamedDecl *ND = NULL; 5350 if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(BaseExpr)) 5351 ND = dyn_cast<NamedDecl>(DRE->getDecl()); 5352 if (const MemberExpr *ME = dyn_cast<MemberExpr>(BaseExpr)) 5353 ND = dyn_cast<NamedDecl>(ME->getMemberDecl()); 5354 5355 if (index.isUnsigned() || !index.isNegative()) { 5356 llvm::APInt size = ArrayTy->getSize(); 5357 if (!size.isStrictlyPositive()) 5358 return; 5359 5360 const Type* BaseType = getElementType(BaseExpr); 5361 if (BaseType != EffectiveType) { 5362 // Make sure we're comparing apples to apples when comparing index to size 5363 uint64_t ptrarith_typesize = Context.getTypeSize(EffectiveType); 5364 uint64_t array_typesize = Context.getTypeSize(BaseType); 5365 // Handle ptrarith_typesize being zero, such as when casting to void* 5366 if (!ptrarith_typesize) ptrarith_typesize = 1; 5367 if (ptrarith_typesize != array_typesize) { 5368 // There's a cast to a different size type involved 5369 uint64_t ratio = array_typesize / ptrarith_typesize; 5370 // TODO: Be smarter about handling cases where array_typesize is not a 5371 // multiple of ptrarith_typesize 5372 if (ptrarith_typesize * ratio == array_typesize) 5373 size *= llvm::APInt(size.getBitWidth(), ratio); 5374 } 5375 } 5376 5377 if (size.getBitWidth() > index.getBitWidth()) 5378 index = index.zext(size.getBitWidth()); 5379 else if (size.getBitWidth() < index.getBitWidth()) 5380 size = size.zext(index.getBitWidth()); 5381 5382 // For array subscripting the index must be less than size, but for pointer 5383 // arithmetic also allow the index (offset) to be equal to size since 5384 // computing the next address after the end of the array is legal and 5385 // commonly done e.g. in C++ iterators and range-based for loops. 5386 if (AllowOnePastEnd ? index.ule(size) : index.ult(size)) 5387 return; 5388 5389 // Also don't warn for arrays of size 1 which are members of some 5390 // structure. These are often used to approximate flexible arrays in C89 5391 // code. 5392 if (IsTailPaddedMemberArray(*this, size, ND)) 5393 return; 5394 5395 // Suppress the warning if the subscript expression (as identified by the 5396 // ']' location) and the index expression are both from macro expansions 5397 // within a system header. 5398 if (ASE) { 5399 SourceLocation RBracketLoc = SourceMgr.getSpellingLoc( 5400 ASE->getRBracketLoc()); 5401 if (SourceMgr.isInSystemHeader(RBracketLoc)) { 5402 SourceLocation IndexLoc = SourceMgr.getSpellingLoc( 5403 IndexExpr->getLocStart()); 5404 if (SourceMgr.isFromSameFile(RBracketLoc, IndexLoc)) 5405 return; 5406 } 5407 } 5408 5409 unsigned DiagID = diag::warn_ptr_arith_exceeds_bounds; 5410 if (ASE) 5411 DiagID = diag::warn_array_index_exceeds_bounds; 5412 5413 DiagRuntimeBehavior(BaseExpr->getLocStart(), BaseExpr, 5414 PDiag(DiagID) << index.toString(10, true) 5415 << size.toString(10, true) 5416 << (unsigned)size.getLimitedValue(~0U) 5417 << IndexExpr->getSourceRange()); 5418 } else { 5419 unsigned DiagID = diag::warn_array_index_precedes_bounds; 5420 if (!ASE) { 5421 DiagID = diag::warn_ptr_arith_precedes_bounds; 5422 if (index.isNegative()) index = -index; 5423 } 5424 5425 DiagRuntimeBehavior(BaseExpr->getLocStart(), BaseExpr, 5426 PDiag(DiagID) << index.toString(10, true) 5427 << IndexExpr->getSourceRange()); 5428 } 5429 5430 if (!ND) { 5431 // Try harder to find a NamedDecl to point at in the note. 5432 while (const ArraySubscriptExpr *ASE = 5433 dyn_cast<ArraySubscriptExpr>(BaseExpr)) 5434 BaseExpr = ASE->getBase()->IgnoreParenCasts(); 5435 if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(BaseExpr)) 5436 ND = dyn_cast<NamedDecl>(DRE->getDecl()); 5437 if (const MemberExpr *ME = dyn_cast<MemberExpr>(BaseExpr)) 5438 ND = dyn_cast<NamedDecl>(ME->getMemberDecl()); 5439 } 5440 5441 if (ND) 5442 DiagRuntimeBehavior(ND->getLocStart(), BaseExpr, 5443 PDiag(diag::note_array_index_out_of_bounds) 5444 << ND->getDeclName()); 5445} 5446 5447void Sema::CheckArrayAccess(const Expr *expr) { 5448 int AllowOnePastEnd = 0; 5449 while (expr) { 5450 expr = expr->IgnoreParenImpCasts(); 5451 switch (expr->getStmtClass()) { 5452 case Stmt::ArraySubscriptExprClass: { 5453 const ArraySubscriptExpr *ASE = cast<ArraySubscriptExpr>(expr); 5454 CheckArrayAccess(ASE->getBase(), ASE->getIdx(), ASE, 5455 AllowOnePastEnd > 0); 5456 return; 5457 } 5458 case Stmt::UnaryOperatorClass: { 5459 // Only unwrap the * and & unary operators 5460 const UnaryOperator *UO = cast<UnaryOperator>(expr); 5461 expr = UO->getSubExpr(); 5462 switch (UO->getOpcode()) { 5463 case UO_AddrOf: 5464 AllowOnePastEnd++; 5465 break; 5466 case UO_Deref: 5467 AllowOnePastEnd--; 5468 break; 5469 default: 5470 return; 5471 } 5472 break; 5473 } 5474 case Stmt::ConditionalOperatorClass: { 5475 const ConditionalOperator *cond = cast<ConditionalOperator>(expr); 5476 if (const Expr *lhs = cond->getLHS()) 5477 CheckArrayAccess(lhs); 5478 if (const Expr *rhs = cond->getRHS()) 5479 CheckArrayAccess(rhs); 5480 return; 5481 } 5482 default: 5483 return; 5484 } 5485 } 5486} 5487 5488//===--- CHECK: Objective-C retain cycles ----------------------------------// 5489 5490namespace { 5491 struct RetainCycleOwner { 5492 RetainCycleOwner() : Variable(0), Indirect(false) {} 5493 VarDecl *Variable; 5494 SourceRange Range; 5495 SourceLocation Loc; 5496 bool Indirect; 5497 5498 void setLocsFrom(Expr *e) { 5499 Loc = e->getExprLoc(); 5500 Range = e->getSourceRange(); 5501 } 5502 }; 5503} 5504 5505/// Consider whether capturing the given variable can possibly lead to 5506/// a retain cycle. 5507static bool considerVariable(VarDecl *var, Expr *ref, RetainCycleOwner &owner) { 5508 // In ARC, it's captured strongly iff the variable has __strong 5509 // lifetime. In MRR, it's captured strongly if the variable is 5510 // __block and has an appropriate type. 5511 if (var->getType().getObjCLifetime() != Qualifiers::OCL_Strong) 5512 return false; 5513 5514 owner.Variable = var; 5515 if (ref) 5516 owner.setLocsFrom(ref); 5517 return true; 5518} 5519 5520static bool findRetainCycleOwner(Sema &S, Expr *e, RetainCycleOwner &owner) { 5521 while (true) { 5522 e = e->IgnoreParens(); 5523 if (CastExpr *cast = dyn_cast<CastExpr>(e)) { 5524 switch (cast->getCastKind()) { 5525 case CK_BitCast: 5526 case CK_LValueBitCast: 5527 case CK_LValueToRValue: 5528 case CK_ARCReclaimReturnedObject: 5529 e = cast->getSubExpr(); 5530 continue; 5531 5532 default: 5533 return false; 5534 } 5535 } 5536 5537 if (ObjCIvarRefExpr *ref = dyn_cast<ObjCIvarRefExpr>(e)) { 5538 ObjCIvarDecl *ivar = ref->getDecl(); 5539 if (ivar->getType().getObjCLifetime() != Qualifiers::OCL_Strong) 5540 return false; 5541 5542 // Try to find a retain cycle in the base. 5543 if (!findRetainCycleOwner(S, ref->getBase(), owner)) 5544 return false; 5545 5546 if (ref->isFreeIvar()) owner.setLocsFrom(ref); 5547 owner.Indirect = true; 5548 return true; 5549 } 5550 5551 if (DeclRefExpr *ref = dyn_cast<DeclRefExpr>(e)) { 5552 VarDecl *var = dyn_cast<VarDecl>(ref->getDecl()); 5553 if (!var) return false; 5554 return considerVariable(var, ref, owner); 5555 } 5556 5557 if (MemberExpr *member = dyn_cast<MemberExpr>(e)) { 5558 if (member->isArrow()) return false; 5559 5560 // Don't count this as an indirect ownership. 5561 e = member->getBase(); 5562 continue; 5563 } 5564 5565 if (PseudoObjectExpr *pseudo = dyn_cast<PseudoObjectExpr>(e)) { 5566 // Only pay attention to pseudo-objects on property references. 5567 ObjCPropertyRefExpr *pre 5568 = dyn_cast<ObjCPropertyRefExpr>(pseudo->getSyntacticForm() 5569 ->IgnoreParens()); 5570 if (!pre) return false; 5571 if (pre->isImplicitProperty()) return false; 5572 ObjCPropertyDecl *property = pre->getExplicitProperty(); 5573 if (!property->isRetaining() && 5574 !(property->getPropertyIvarDecl() && 5575 property->getPropertyIvarDecl()->getType() 5576 .getObjCLifetime() == Qualifiers::OCL_Strong)) 5577 return false; 5578 5579 owner.Indirect = true; 5580 if (pre->isSuperReceiver()) { 5581 owner.Variable = S.getCurMethodDecl()->getSelfDecl(); 5582 if (!owner.Variable) 5583 return false; 5584 owner.Loc = pre->getLocation(); 5585 owner.Range = pre->getSourceRange(); 5586 return true; 5587 } 5588 e = const_cast<Expr*>(cast<OpaqueValueExpr>(pre->getBase()) 5589 ->getSourceExpr()); 5590 continue; 5591 } 5592 5593 // Array ivars? 5594 5595 return false; 5596 } 5597} 5598 5599namespace { 5600 struct FindCaptureVisitor : EvaluatedExprVisitor<FindCaptureVisitor> { 5601 FindCaptureVisitor(ASTContext &Context, VarDecl *variable) 5602 : EvaluatedExprVisitor<FindCaptureVisitor>(Context), 5603 Variable(variable), Capturer(0) {} 5604 5605 VarDecl *Variable; 5606 Expr *Capturer; 5607 5608 void VisitDeclRefExpr(DeclRefExpr *ref) { 5609 if (ref->getDecl() == Variable && !Capturer) 5610 Capturer = ref; 5611 } 5612 5613 void VisitObjCIvarRefExpr(ObjCIvarRefExpr *ref) { 5614 if (Capturer) return; 5615 Visit(ref->getBase()); 5616 if (Capturer && ref->isFreeIvar()) 5617 Capturer = ref; 5618 } 5619 5620 void VisitBlockExpr(BlockExpr *block) { 5621 // Look inside nested blocks 5622 if (block->getBlockDecl()->capturesVariable(Variable)) 5623 Visit(block->getBlockDecl()->getBody()); 5624 } 5625 5626 void VisitOpaqueValueExpr(OpaqueValueExpr *OVE) { 5627 if (Capturer) return; 5628 if (OVE->getSourceExpr()) 5629 Visit(OVE->getSourceExpr()); 5630 } 5631 }; 5632} 5633 5634/// Check whether the given argument is a block which captures a 5635/// variable. 5636static Expr *findCapturingExpr(Sema &S, Expr *e, RetainCycleOwner &owner) { 5637 assert(owner.Variable && owner.Loc.isValid()); 5638 5639 e = e->IgnoreParenCasts(); 5640 5641 // Look through [^{...} copy] and Block_copy(^{...}). 5642 if (ObjCMessageExpr *ME = dyn_cast<ObjCMessageExpr>(e)) { 5643 Selector Cmd = ME->getSelector(); 5644 if (Cmd.isUnarySelector() && Cmd.getNameForSlot(0) == "copy") { 5645 e = ME->getInstanceReceiver(); 5646 if (!e) 5647 return 0; 5648 e = e->IgnoreParenCasts(); 5649 } 5650 } else if (CallExpr *CE = dyn_cast<CallExpr>(e)) { 5651 if (CE->getNumArgs() == 1) { 5652 FunctionDecl *Fn = dyn_cast_or_null<FunctionDecl>(CE->getCalleeDecl()); 5653 if (Fn) { 5654 const IdentifierInfo *FnI = Fn->getIdentifier(); 5655 if (FnI && FnI->isStr("_Block_copy")) { 5656 e = CE->getArg(0)->IgnoreParenCasts(); 5657 } 5658 } 5659 } 5660 } 5661 5662 BlockExpr *block = dyn_cast<BlockExpr>(e); 5663 if (!block || !block->getBlockDecl()->capturesVariable(owner.Variable)) 5664 return 0; 5665 5666 FindCaptureVisitor visitor(S.Context, owner.Variable); 5667 visitor.Visit(block->getBlockDecl()->getBody()); 5668 return visitor.Capturer; 5669} 5670 5671static void diagnoseRetainCycle(Sema &S, Expr *capturer, 5672 RetainCycleOwner &owner) { 5673 assert(capturer); 5674 assert(owner.Variable && owner.Loc.isValid()); 5675 5676 S.Diag(capturer->getExprLoc(), diag::warn_arc_retain_cycle) 5677 << owner.Variable << capturer->getSourceRange(); 5678 S.Diag(owner.Loc, diag::note_arc_retain_cycle_owner) 5679 << owner.Indirect << owner.Range; 5680} 5681 5682/// Check for a keyword selector that starts with the word 'add' or 5683/// 'set'. 5684static bool isSetterLikeSelector(Selector sel) { 5685 if (sel.isUnarySelector()) return false; 5686 5687 StringRef str = sel.getNameForSlot(0); 5688 while (!str.empty() && str.front() == '_') str = str.substr(1); 5689 if (str.startswith("set")) 5690 str = str.substr(3); 5691 else if (str.startswith("add")) { 5692 // Specially whitelist 'addOperationWithBlock:'. 5693 if (sel.getNumArgs() == 1 && str.startswith("addOperationWithBlock")) 5694 return false; 5695 str = str.substr(3); 5696 } 5697 else 5698 return false; 5699 5700 if (str.empty()) return true; 5701 return !islower(str.front()); 5702} 5703 5704/// Check a message send to see if it's likely to cause a retain cycle. 5705void Sema::checkRetainCycles(ObjCMessageExpr *msg) { 5706 // Only check instance methods whose selector looks like a setter. 5707 if (!msg->isInstanceMessage() || !isSetterLikeSelector(msg->getSelector())) 5708 return; 5709 5710 // Try to find a variable that the receiver is strongly owned by. 5711 RetainCycleOwner owner; 5712 if (msg->getReceiverKind() == ObjCMessageExpr::Instance) { 5713 if (!findRetainCycleOwner(*this, msg->getInstanceReceiver(), owner)) 5714 return; 5715 } else { 5716 assert(msg->getReceiverKind() == ObjCMessageExpr::SuperInstance); 5717 owner.Variable = getCurMethodDecl()->getSelfDecl(); 5718 owner.Loc = msg->getSuperLoc(); 5719 owner.Range = msg->getSuperLoc(); 5720 } 5721 5722 // Check whether the receiver is captured by any of the arguments. 5723 for (unsigned i = 0, e = msg->getNumArgs(); i != e; ++i) 5724 if (Expr *capturer = findCapturingExpr(*this, msg->getArg(i), owner)) 5725 return diagnoseRetainCycle(*this, capturer, owner); 5726} 5727 5728/// Check a property assign to see if it's likely to cause a retain cycle. 5729void Sema::checkRetainCycles(Expr *receiver, Expr *argument) { 5730 RetainCycleOwner owner; 5731 if (!findRetainCycleOwner(*this, receiver, owner)) 5732 return; 5733 5734 if (Expr *capturer = findCapturingExpr(*this, argument, owner)) 5735 diagnoseRetainCycle(*this, capturer, owner); 5736} 5737 5738void Sema::checkRetainCycles(VarDecl *Var, Expr *Init) { 5739 RetainCycleOwner Owner; 5740 if (!considerVariable(Var, /*DeclRefExpr=*/0, Owner)) 5741 return; 5742 5743 // Because we don't have an expression for the variable, we have to set the 5744 // location explicitly here. 5745 Owner.Loc = Var->getLocation(); 5746 Owner.Range = Var->getSourceRange(); 5747 5748 if (Expr *Capturer = findCapturingExpr(*this, Init, Owner)) 5749 diagnoseRetainCycle(*this, Capturer, Owner); 5750} 5751 5752bool Sema::checkUnsafeAssigns(SourceLocation Loc, 5753 QualType LHS, Expr *RHS) { 5754 Qualifiers::ObjCLifetime LT = LHS.getObjCLifetime(); 5755 if (LT != Qualifiers::OCL_Weak && LT != Qualifiers::OCL_ExplicitNone) 5756 return false; 5757 // strip off any implicit cast added to get to the one arc-specific 5758 while (ImplicitCastExpr *cast = dyn_cast<ImplicitCastExpr>(RHS)) { 5759 if (cast->getCastKind() == CK_ARCConsumeObject) { 5760 Diag(Loc, diag::warn_arc_retained_assign) 5761 << (LT == Qualifiers::OCL_ExplicitNone) << 1 5762 << RHS->getSourceRange(); 5763 return true; 5764 } 5765 RHS = cast->getSubExpr(); 5766 } 5767 return false; 5768} 5769 5770void Sema::checkUnsafeExprAssigns(SourceLocation Loc, 5771 Expr *LHS, Expr *RHS) { 5772 QualType LHSType; 5773 // PropertyRef on LHS type need be directly obtained from 5774 // its declaration as it has a PsuedoType. 5775 ObjCPropertyRefExpr *PRE 5776 = dyn_cast<ObjCPropertyRefExpr>(LHS->IgnoreParens()); 5777 if (PRE && !PRE->isImplicitProperty()) { 5778 const ObjCPropertyDecl *PD = PRE->getExplicitProperty(); 5779 if (PD) 5780 LHSType = PD->getType(); 5781 } 5782 5783 if (LHSType.isNull()) 5784 LHSType = LHS->getType(); 5785 5786 Qualifiers::ObjCLifetime LT = LHSType.getObjCLifetime(); 5787 5788 if (LT == Qualifiers::OCL_Weak) { 5789 DiagnosticsEngine::Level Level = 5790 Diags.getDiagnosticLevel(diag::warn_arc_repeated_use_of_weak, Loc); 5791 if (Level != DiagnosticsEngine::Ignored) 5792 getCurFunction()->markSafeWeakUse(LHS); 5793 } 5794 5795 if (checkUnsafeAssigns(Loc, LHSType, RHS)) 5796 return; 5797 5798 // FIXME. Check for other life times. 5799 if (LT != Qualifiers::OCL_None) 5800 return; 5801 5802 if (PRE) { 5803 if (PRE->isImplicitProperty()) 5804 return; 5805 const ObjCPropertyDecl *PD = PRE->getExplicitProperty(); 5806 if (!PD) 5807 return; 5808 5809 unsigned Attributes = PD->getPropertyAttributes(); 5810 if (Attributes & ObjCPropertyDecl::OBJC_PR_assign) { 5811 // when 'assign' attribute was not explicitly specified 5812 // by user, ignore it and rely on property type itself 5813 // for lifetime info. 5814 unsigned AsWrittenAttr = PD->getPropertyAttributesAsWritten(); 5815 if (!(AsWrittenAttr & ObjCPropertyDecl::OBJC_PR_assign) && 5816 LHSType->isObjCRetainableType()) 5817 return; 5818 5819 while (ImplicitCastExpr *cast = dyn_cast<ImplicitCastExpr>(RHS)) { 5820 if (cast->getCastKind() == CK_ARCConsumeObject) { 5821 Diag(Loc, diag::warn_arc_retained_property_assign) 5822 << RHS->getSourceRange(); 5823 return; 5824 } 5825 RHS = cast->getSubExpr(); 5826 } 5827 } 5828 else if (Attributes & ObjCPropertyDecl::OBJC_PR_weak) { 5829 while (ImplicitCastExpr *cast = dyn_cast<ImplicitCastExpr>(RHS)) { 5830 if (cast->getCastKind() == CK_ARCConsumeObject) { 5831 Diag(Loc, diag::warn_arc_retained_assign) 5832 << 0 << 0<< RHS->getSourceRange(); 5833 return; 5834 } 5835 RHS = cast->getSubExpr(); 5836 } 5837 } 5838 } 5839} 5840 5841//===--- CHECK: Empty statement body (-Wempty-body) ---------------------===// 5842 5843namespace { 5844bool ShouldDiagnoseEmptyStmtBody(const SourceManager &SourceMgr, 5845 SourceLocation StmtLoc, 5846 const NullStmt *Body) { 5847 // Do not warn if the body is a macro that expands to nothing, e.g: 5848 // 5849 // #define CALL(x) 5850 // if (condition) 5851 // CALL(0); 5852 // 5853 if (Body->hasLeadingEmptyMacro()) 5854 return false; 5855 5856 // Get line numbers of statement and body. 5857 bool StmtLineInvalid; 5858 unsigned StmtLine = SourceMgr.getSpellingLineNumber(StmtLoc, 5859 &StmtLineInvalid); 5860 if (StmtLineInvalid) 5861 return false; 5862 5863 bool BodyLineInvalid; 5864 unsigned BodyLine = SourceMgr.getSpellingLineNumber(Body->getSemiLoc(), 5865 &BodyLineInvalid); 5866 if (BodyLineInvalid) 5867 return false; 5868 5869 // Warn if null statement and body are on the same line. 5870 if (StmtLine != BodyLine) 5871 return false; 5872 5873 return true; 5874} 5875} // Unnamed namespace 5876 5877void Sema::DiagnoseEmptyStmtBody(SourceLocation StmtLoc, 5878 const Stmt *Body, 5879 unsigned DiagID) { 5880 // Since this is a syntactic check, don't emit diagnostic for template 5881 // instantiations, this just adds noise. 5882 if (CurrentInstantiationScope) 5883 return; 5884 5885 // The body should be a null statement. 5886 const NullStmt *NBody = dyn_cast<NullStmt>(Body); 5887 if (!NBody) 5888 return; 5889 5890 // Do the usual checks. 5891 if (!ShouldDiagnoseEmptyStmtBody(SourceMgr, StmtLoc, NBody)) 5892 return; 5893 5894 Diag(NBody->getSemiLoc(), DiagID); 5895 Diag(NBody->getSemiLoc(), diag::note_empty_body_on_separate_line); 5896} 5897 5898void Sema::DiagnoseEmptyLoopBody(const Stmt *S, 5899 const Stmt *PossibleBody) { 5900 assert(!CurrentInstantiationScope); // Ensured by caller 5901 5902 SourceLocation StmtLoc; 5903 const Stmt *Body; 5904 unsigned DiagID; 5905 if (const ForStmt *FS = dyn_cast<ForStmt>(S)) { 5906 StmtLoc = FS->getRParenLoc(); 5907 Body = FS->getBody(); 5908 DiagID = diag::warn_empty_for_body; 5909 } else if (const WhileStmt *WS = dyn_cast<WhileStmt>(S)) { 5910 StmtLoc = WS->getCond()->getSourceRange().getEnd(); 5911 Body = WS->getBody(); 5912 DiagID = diag::warn_empty_while_body; 5913 } else 5914 return; // Neither `for' nor `while'. 5915 5916 // The body should be a null statement. 5917 const NullStmt *NBody = dyn_cast<NullStmt>(Body); 5918 if (!NBody) 5919 return; 5920 5921 // Skip expensive checks if diagnostic is disabled. 5922 if (Diags.getDiagnosticLevel(DiagID, NBody->getSemiLoc()) == 5923 DiagnosticsEngine::Ignored) 5924 return; 5925 5926 // Do the usual checks. 5927 if (!ShouldDiagnoseEmptyStmtBody(SourceMgr, StmtLoc, NBody)) 5928 return; 5929 5930 // `for(...);' and `while(...);' are popular idioms, so in order to keep 5931 // noise level low, emit diagnostics only if for/while is followed by a 5932 // CompoundStmt, e.g.: 5933 // for (int i = 0; i < n; i++); 5934 // { 5935 // a(i); 5936 // } 5937 // or if for/while is followed by a statement with more indentation 5938 // than for/while itself: 5939 // for (int i = 0; i < n; i++); 5940 // a(i); 5941 bool ProbableTypo = isa<CompoundStmt>(PossibleBody); 5942 if (!ProbableTypo) { 5943 bool BodyColInvalid; 5944 unsigned BodyCol = SourceMgr.getPresumedColumnNumber( 5945 PossibleBody->getLocStart(), 5946 &BodyColInvalid); 5947 if (BodyColInvalid) 5948 return; 5949 5950 bool StmtColInvalid; 5951 unsigned StmtCol = SourceMgr.getPresumedColumnNumber( 5952 S->getLocStart(), 5953 &StmtColInvalid); 5954 if (StmtColInvalid) 5955 return; 5956 5957 if (BodyCol > StmtCol) 5958 ProbableTypo = true; 5959 } 5960 5961 if (ProbableTypo) { 5962 Diag(NBody->getSemiLoc(), DiagID); 5963 Diag(NBody->getSemiLoc(), diag::note_empty_body_on_separate_line); 5964 } 5965} 5966 5967//===--- Layout compatibility ----------------------------------------------// 5968 5969namespace { 5970 5971bool isLayoutCompatible(ASTContext &C, QualType T1, QualType T2); 5972 5973/// \brief Check if two enumeration types are layout-compatible. 5974bool isLayoutCompatible(ASTContext &C, EnumDecl *ED1, EnumDecl *ED2) { 5975 // C++11 [dcl.enum] p8: 5976 // Two enumeration types are layout-compatible if they have the same 5977 // underlying type. 5978 return ED1->isComplete() && ED2->isComplete() && 5979 C.hasSameType(ED1->getIntegerType(), ED2->getIntegerType()); 5980} 5981 5982/// \brief Check if two fields are layout-compatible. 5983bool isLayoutCompatible(ASTContext &C, FieldDecl *Field1, FieldDecl *Field2) { 5984 if (!isLayoutCompatible(C, Field1->getType(), Field2->getType())) 5985 return false; 5986 5987 if (Field1->isBitField() != Field2->isBitField()) 5988 return false; 5989 5990 if (Field1->isBitField()) { 5991 // Make sure that the bit-fields are the same length. 5992 unsigned Bits1 = Field1->getBitWidthValue(C); 5993 unsigned Bits2 = Field2->getBitWidthValue(C); 5994 5995 if (Bits1 != Bits2) 5996 return false; 5997 } 5998 5999 return true; 6000} 6001 6002/// \brief Check if two standard-layout structs are layout-compatible. 6003/// (C++11 [class.mem] p17) 6004bool isLayoutCompatibleStruct(ASTContext &C, 6005 RecordDecl *RD1, 6006 RecordDecl *RD2) { 6007 // If both records are C++ classes, check that base classes match. 6008 if (const CXXRecordDecl *D1CXX = dyn_cast<CXXRecordDecl>(RD1)) { 6009 // If one of records is a CXXRecordDecl we are in C++ mode, 6010 // thus the other one is a CXXRecordDecl, too. 6011 const CXXRecordDecl *D2CXX = cast<CXXRecordDecl>(RD2); 6012 // Check number of base classes. 6013 if (D1CXX->getNumBases() != D2CXX->getNumBases()) 6014 return false; 6015 6016 // Check the base classes. 6017 for (CXXRecordDecl::base_class_const_iterator 6018 Base1 = D1CXX->bases_begin(), 6019 BaseEnd1 = D1CXX->bases_end(), 6020 Base2 = D2CXX->bases_begin(); 6021 Base1 != BaseEnd1; 6022 ++Base1, ++Base2) { 6023 if (!isLayoutCompatible(C, Base1->getType(), Base2->getType())) 6024 return false; 6025 } 6026 } else if (const CXXRecordDecl *D2CXX = dyn_cast<CXXRecordDecl>(RD2)) { 6027 // If only RD2 is a C++ class, it should have zero base classes. 6028 if (D2CXX->getNumBases() > 0) 6029 return false; 6030 } 6031 6032 // Check the fields. 6033 RecordDecl::field_iterator Field2 = RD2->field_begin(), 6034 Field2End = RD2->field_end(), 6035 Field1 = RD1->field_begin(), 6036 Field1End = RD1->field_end(); 6037 for ( ; Field1 != Field1End && Field2 != Field2End; ++Field1, ++Field2) { 6038 if (!isLayoutCompatible(C, *Field1, *Field2)) 6039 return false; 6040 } 6041 if (Field1 != Field1End || Field2 != Field2End) 6042 return false; 6043 6044 return true; 6045} 6046 6047/// \brief Check if two standard-layout unions are layout-compatible. 6048/// (C++11 [class.mem] p18) 6049bool isLayoutCompatibleUnion(ASTContext &C, 6050 RecordDecl *RD1, 6051 RecordDecl *RD2) { 6052 llvm::SmallPtrSet<FieldDecl *, 8> UnmatchedFields; 6053 for (RecordDecl::field_iterator Field2 = RD2->field_begin(), 6054 Field2End = RD2->field_end(); 6055 Field2 != Field2End; ++Field2) { 6056 UnmatchedFields.insert(*Field2); 6057 } 6058 6059 for (RecordDecl::field_iterator Field1 = RD1->field_begin(), 6060 Field1End = RD1->field_end(); 6061 Field1 != Field1End; ++Field1) { 6062 llvm::SmallPtrSet<FieldDecl *, 8>::iterator 6063 I = UnmatchedFields.begin(), 6064 E = UnmatchedFields.end(); 6065 6066 for ( ; I != E; ++I) { 6067 if (isLayoutCompatible(C, *Field1, *I)) { 6068 bool Result = UnmatchedFields.erase(*I); 6069 (void) Result; 6070 assert(Result); 6071 break; 6072 } 6073 } 6074 if (I == E) 6075 return false; 6076 } 6077 6078 return UnmatchedFields.empty(); 6079} 6080 6081bool isLayoutCompatible(ASTContext &C, RecordDecl *RD1, RecordDecl *RD2) { 6082 if (RD1->isUnion() != RD2->isUnion()) 6083 return false; 6084 6085 if (RD1->isUnion()) 6086 return isLayoutCompatibleUnion(C, RD1, RD2); 6087 else 6088 return isLayoutCompatibleStruct(C, RD1, RD2); 6089} 6090 6091/// \brief Check if two types are layout-compatible in C++11 sense. 6092bool isLayoutCompatible(ASTContext &C, QualType T1, QualType T2) { 6093 if (T1.isNull() || T2.isNull()) 6094 return false; 6095 6096 // C++11 [basic.types] p11: 6097 // If two types T1 and T2 are the same type, then T1 and T2 are 6098 // layout-compatible types. 6099 if (C.hasSameType(T1, T2)) 6100 return true; 6101 6102 T1 = T1.getCanonicalType().getUnqualifiedType(); 6103 T2 = T2.getCanonicalType().getUnqualifiedType(); 6104 6105 const Type::TypeClass TC1 = T1->getTypeClass(); 6106 const Type::TypeClass TC2 = T2->getTypeClass(); 6107 6108 if (TC1 != TC2) 6109 return false; 6110 6111 if (TC1 == Type::Enum) { 6112 return isLayoutCompatible(C, 6113 cast<EnumType>(T1)->getDecl(), 6114 cast<EnumType>(T2)->getDecl()); 6115 } else if (TC1 == Type::Record) { 6116 if (!T1->isStandardLayoutType() || !T2->isStandardLayoutType()) 6117 return false; 6118 6119 return isLayoutCompatible(C, 6120 cast<RecordType>(T1)->getDecl(), 6121 cast<RecordType>(T2)->getDecl()); 6122 } 6123 6124 return false; 6125} 6126} 6127 6128//===--- CHECK: pointer_with_type_tag attribute: datatypes should match ----// 6129 6130namespace { 6131/// \brief Given a type tag expression find the type tag itself. 6132/// 6133/// \param TypeExpr Type tag expression, as it appears in user's code. 6134/// 6135/// \param VD Declaration of an identifier that appears in a type tag. 6136/// 6137/// \param MagicValue Type tag magic value. 6138bool FindTypeTagExpr(const Expr *TypeExpr, const ASTContext &Ctx, 6139 const ValueDecl **VD, uint64_t *MagicValue) { 6140 while(true) { 6141 if (!TypeExpr) 6142 return false; 6143 6144 TypeExpr = TypeExpr->IgnoreParenImpCasts()->IgnoreParenCasts(); 6145 6146 switch (TypeExpr->getStmtClass()) { 6147 case Stmt::UnaryOperatorClass: { 6148 const UnaryOperator *UO = cast<UnaryOperator>(TypeExpr); 6149 if (UO->getOpcode() == UO_AddrOf || UO->getOpcode() == UO_Deref) { 6150 TypeExpr = UO->getSubExpr(); 6151 continue; 6152 } 6153 return false; 6154 } 6155 6156 case Stmt::DeclRefExprClass: { 6157 const DeclRefExpr *DRE = cast<DeclRefExpr>(TypeExpr); 6158 *VD = DRE->getDecl(); 6159 return true; 6160 } 6161 6162 case Stmt::IntegerLiteralClass: { 6163 const IntegerLiteral *IL = cast<IntegerLiteral>(TypeExpr); 6164 llvm::APInt MagicValueAPInt = IL->getValue(); 6165 if (MagicValueAPInt.getActiveBits() <= 64) { 6166 *MagicValue = MagicValueAPInt.getZExtValue(); 6167 return true; 6168 } else 6169 return false; 6170 } 6171 6172 case Stmt::BinaryConditionalOperatorClass: 6173 case Stmt::ConditionalOperatorClass: { 6174 const AbstractConditionalOperator *ACO = 6175 cast<AbstractConditionalOperator>(TypeExpr); 6176 bool Result; 6177 if (ACO->getCond()->EvaluateAsBooleanCondition(Result, Ctx)) { 6178 if (Result) 6179 TypeExpr = ACO->getTrueExpr(); 6180 else 6181 TypeExpr = ACO->getFalseExpr(); 6182 continue; 6183 } 6184 return false; 6185 } 6186 6187 case Stmt::BinaryOperatorClass: { 6188 const BinaryOperator *BO = cast<BinaryOperator>(TypeExpr); 6189 if (BO->getOpcode() == BO_Comma) { 6190 TypeExpr = BO->getRHS(); 6191 continue; 6192 } 6193 return false; 6194 } 6195 6196 default: 6197 return false; 6198 } 6199 } 6200} 6201 6202/// \brief Retrieve the C type corresponding to type tag TypeExpr. 6203/// 6204/// \param TypeExpr Expression that specifies a type tag. 6205/// 6206/// \param MagicValues Registered magic values. 6207/// 6208/// \param FoundWrongKind Set to true if a type tag was found, but of a wrong 6209/// kind. 6210/// 6211/// \param TypeInfo Information about the corresponding C type. 6212/// 6213/// \returns true if the corresponding C type was found. 6214bool GetMatchingCType( 6215 const IdentifierInfo *ArgumentKind, 6216 const Expr *TypeExpr, const ASTContext &Ctx, 6217 const llvm::DenseMap<Sema::TypeTagMagicValue, 6218 Sema::TypeTagData> *MagicValues, 6219 bool &FoundWrongKind, 6220 Sema::TypeTagData &TypeInfo) { 6221 FoundWrongKind = false; 6222 6223 // Variable declaration that has type_tag_for_datatype attribute. 6224 const ValueDecl *VD = NULL; 6225 6226 uint64_t MagicValue; 6227 6228 if (!FindTypeTagExpr(TypeExpr, Ctx, &VD, &MagicValue)) 6229 return false; 6230 6231 if (VD) { 6232 for (specific_attr_iterator<TypeTagForDatatypeAttr> 6233 I = VD->specific_attr_begin<TypeTagForDatatypeAttr>(), 6234 E = VD->specific_attr_end<TypeTagForDatatypeAttr>(); 6235 I != E; ++I) { 6236 if (I->getArgumentKind() != ArgumentKind) { 6237 FoundWrongKind = true; 6238 return false; 6239 } 6240 TypeInfo.Type = I->getMatchingCType(); 6241 TypeInfo.LayoutCompatible = I->getLayoutCompatible(); 6242 TypeInfo.MustBeNull = I->getMustBeNull(); 6243 return true; 6244 } 6245 return false; 6246 } 6247 6248 if (!MagicValues) 6249 return false; 6250 6251 llvm::DenseMap<Sema::TypeTagMagicValue, 6252 Sema::TypeTagData>::const_iterator I = 6253 MagicValues->find(std::make_pair(ArgumentKind, MagicValue)); 6254 if (I == MagicValues->end()) 6255 return false; 6256 6257 TypeInfo = I->second; 6258 return true; 6259} 6260} // unnamed namespace 6261 6262void Sema::RegisterTypeTagForDatatype(const IdentifierInfo *ArgumentKind, 6263 uint64_t MagicValue, QualType Type, 6264 bool LayoutCompatible, 6265 bool MustBeNull) { 6266 if (!TypeTagForDatatypeMagicValues) 6267 TypeTagForDatatypeMagicValues.reset( 6268 new llvm::DenseMap<TypeTagMagicValue, TypeTagData>); 6269 6270 TypeTagMagicValue Magic(ArgumentKind, MagicValue); 6271 (*TypeTagForDatatypeMagicValues)[Magic] = 6272 TypeTagData(Type, LayoutCompatible, MustBeNull); 6273} 6274 6275namespace { 6276bool IsSameCharType(QualType T1, QualType T2) { 6277 const BuiltinType *BT1 = T1->getAs<BuiltinType>(); 6278 if (!BT1) 6279 return false; 6280 6281 const BuiltinType *BT2 = T2->getAs<BuiltinType>(); 6282 if (!BT2) 6283 return false; 6284 6285 BuiltinType::Kind T1Kind = BT1->getKind(); 6286 BuiltinType::Kind T2Kind = BT2->getKind(); 6287 6288 return (T1Kind == BuiltinType::SChar && T2Kind == BuiltinType::Char_S) || 6289 (T1Kind == BuiltinType::UChar && T2Kind == BuiltinType::Char_U) || 6290 (T1Kind == BuiltinType::Char_U && T2Kind == BuiltinType::UChar) || 6291 (T1Kind == BuiltinType::Char_S && T2Kind == BuiltinType::SChar); 6292} 6293} // unnamed namespace 6294 6295void Sema::CheckArgumentWithTypeTag(const ArgumentWithTypeTagAttr *Attr, 6296 const Expr * const *ExprArgs) { 6297 const IdentifierInfo *ArgumentKind = Attr->getArgumentKind(); 6298 bool IsPointerAttr = Attr->getIsPointer(); 6299 6300 const Expr *TypeTagExpr = ExprArgs[Attr->getTypeTagIdx()]; 6301 bool FoundWrongKind; 6302 TypeTagData TypeInfo; 6303 if (!GetMatchingCType(ArgumentKind, TypeTagExpr, Context, 6304 TypeTagForDatatypeMagicValues.get(), 6305 FoundWrongKind, TypeInfo)) { 6306 if (FoundWrongKind) 6307 Diag(TypeTagExpr->getExprLoc(), 6308 diag::warn_type_tag_for_datatype_wrong_kind) 6309 << TypeTagExpr->getSourceRange(); 6310 return; 6311 } 6312 6313 const Expr *ArgumentExpr = ExprArgs[Attr->getArgumentIdx()]; 6314 if (IsPointerAttr) { 6315 // Skip implicit cast of pointer to `void *' (as a function argument). 6316 if (const ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(ArgumentExpr)) 6317 if (ICE->getType()->isVoidPointerType() && 6318 ICE->getCastKind() == CK_BitCast) 6319 ArgumentExpr = ICE->getSubExpr(); 6320 } 6321 QualType ArgumentType = ArgumentExpr->getType(); 6322 6323 // Passing a `void*' pointer shouldn't trigger a warning. 6324 if (IsPointerAttr && ArgumentType->isVoidPointerType()) 6325 return; 6326 6327 if (TypeInfo.MustBeNull) { 6328 // Type tag with matching void type requires a null pointer. 6329 if (!ArgumentExpr->isNullPointerConstant(Context, 6330 Expr::NPC_ValueDependentIsNotNull)) { 6331 Diag(ArgumentExpr->getExprLoc(), 6332 diag::warn_type_safety_null_pointer_required) 6333 << ArgumentKind->getName() 6334 << ArgumentExpr->getSourceRange() 6335 << TypeTagExpr->getSourceRange(); 6336 } 6337 return; 6338 } 6339 6340 QualType RequiredType = TypeInfo.Type; 6341 if (IsPointerAttr) 6342 RequiredType = Context.getPointerType(RequiredType); 6343 6344 bool mismatch = false; 6345 if (!TypeInfo.LayoutCompatible) { 6346 mismatch = !Context.hasSameType(ArgumentType, RequiredType); 6347 6348 // C++11 [basic.fundamental] p1: 6349 // Plain char, signed char, and unsigned char are three distinct types. 6350 // 6351 // But we treat plain `char' as equivalent to `signed char' or `unsigned 6352 // char' depending on the current char signedness mode. 6353 if (mismatch) 6354 if ((IsPointerAttr && IsSameCharType(ArgumentType->getPointeeType(), 6355 RequiredType->getPointeeType())) || 6356 (!IsPointerAttr && IsSameCharType(ArgumentType, RequiredType))) 6357 mismatch = false; 6358 } else 6359 if (IsPointerAttr) 6360 mismatch = !isLayoutCompatible(Context, 6361 ArgumentType->getPointeeType(), 6362 RequiredType->getPointeeType()); 6363 else 6364 mismatch = !isLayoutCompatible(Context, ArgumentType, RequiredType); 6365 6366 if (mismatch) 6367 Diag(ArgumentExpr->getExprLoc(), diag::warn_type_safety_type_mismatch) 6368 << ArgumentType << ArgumentKind->getName() 6369 << TypeInfo.LayoutCompatible << RequiredType 6370 << ArgumentExpr->getSourceRange() 6371 << TypeTagExpr->getSourceRange(); 6372} 6373