SemaChecking.cpp revision bd5da9d1a711e90b3bc265b4798954085c0b48d9
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/Sema.h" 16#include "clang/Sema/SemaInternal.h" 17#include "clang/Sema/ScopeInfo.h" 18#include "clang/Analysis/Analyses/FormatString.h" 19#include "clang/AST/ASTContext.h" 20#include "clang/AST/CharUnits.h" 21#include "clang/AST/DeclCXX.h" 22#include "clang/AST/DeclObjC.h" 23#include "clang/AST/ExprCXX.h" 24#include "clang/AST/ExprObjC.h" 25#include "clang/AST/EvaluatedExprVisitor.h" 26#include "clang/AST/DeclObjC.h" 27#include "clang/AST/StmtCXX.h" 28#include "clang/AST/StmtObjC.h" 29#include "clang/Lex/Preprocessor.h" 30#include "llvm/ADT/BitVector.h" 31#include "llvm/ADT/STLExtras.h" 32#include "llvm/Support/raw_ostream.h" 33#include "clang/Basic/TargetBuiltins.h" 34#include "clang/Basic/TargetInfo.h" 35#include "clang/Basic/ConvertUTF.h" 36#include <limits> 37using namespace clang; 38using namespace sema; 39 40SourceLocation Sema::getLocationOfStringLiteralByte(const StringLiteral *SL, 41 unsigned ByteNo) const { 42 return SL->getLocationOfByte(ByteNo, PP.getSourceManager(), 43 PP.getLangOptions(), PP.getTargetInfo()); 44} 45 46 47/// CheckablePrintfAttr - does a function call have a "printf" attribute 48/// and arguments that merit checking? 49bool Sema::CheckablePrintfAttr(const FormatAttr *Format, CallExpr *TheCall) { 50 if (Format->getType() == "printf") return true; 51 if (Format->getType() == "printf0") { 52 // printf0 allows null "format" string; if so don't check format/args 53 unsigned format_idx = Format->getFormatIdx() - 1; 54 // Does the index refer to the implicit object argument? 55 if (isa<CXXMemberCallExpr>(TheCall)) { 56 if (format_idx == 0) 57 return false; 58 --format_idx; 59 } 60 if (format_idx < TheCall->getNumArgs()) { 61 Expr *Format = TheCall->getArg(format_idx)->IgnoreParenCasts(); 62 if (!Format->isNullPointerConstant(Context, 63 Expr::NPC_ValueDependentIsNull)) 64 return true; 65 } 66 } 67 return false; 68} 69 70/// Checks that a call expression's argument count is the desired number. 71/// This is useful when doing custom type-checking. Returns true on error. 72static bool checkArgCount(Sema &S, CallExpr *call, unsigned desiredArgCount) { 73 unsigned argCount = call->getNumArgs(); 74 if (argCount == desiredArgCount) return false; 75 76 if (argCount < desiredArgCount) 77 return S.Diag(call->getLocEnd(), diag::err_typecheck_call_too_few_args) 78 << 0 /*function call*/ << desiredArgCount << argCount 79 << call->getSourceRange(); 80 81 // Highlight all the excess arguments. 82 SourceRange range(call->getArg(desiredArgCount)->getLocStart(), 83 call->getArg(argCount - 1)->getLocEnd()); 84 85 return S.Diag(range.getBegin(), diag::err_typecheck_call_too_many_args) 86 << 0 /*function call*/ << desiredArgCount << argCount 87 << call->getArg(1)->getSourceRange(); 88} 89 90ExprResult 91Sema::CheckBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) { 92 ExprResult TheCallResult(Owned(TheCall)); 93 94 // Find out if any arguments are required to be integer constant expressions. 95 unsigned ICEArguments = 0; 96 ASTContext::GetBuiltinTypeError Error; 97 Context.GetBuiltinType(BuiltinID, Error, &ICEArguments); 98 if (Error != ASTContext::GE_None) 99 ICEArguments = 0; // Don't diagnose previously diagnosed errors. 100 101 // If any arguments are required to be ICE's, check and diagnose. 102 for (unsigned ArgNo = 0; ICEArguments != 0; ++ArgNo) { 103 // Skip arguments not required to be ICE's. 104 if ((ICEArguments & (1 << ArgNo)) == 0) continue; 105 106 llvm::APSInt Result; 107 if (SemaBuiltinConstantArg(TheCall, ArgNo, Result)) 108 return true; 109 ICEArguments &= ~(1 << ArgNo); 110 } 111 112 switch (BuiltinID) { 113 case Builtin::BI__builtin___CFStringMakeConstantString: 114 assert(TheCall->getNumArgs() == 1 && 115 "Wrong # arguments to builtin CFStringMakeConstantString"); 116 if (CheckObjCString(TheCall->getArg(0))) 117 return ExprError(); 118 break; 119 case Builtin::BI__builtin_stdarg_start: 120 case Builtin::BI__builtin_va_start: 121 if (SemaBuiltinVAStart(TheCall)) 122 return ExprError(); 123 break; 124 case Builtin::BI__builtin_isgreater: 125 case Builtin::BI__builtin_isgreaterequal: 126 case Builtin::BI__builtin_isless: 127 case Builtin::BI__builtin_islessequal: 128 case Builtin::BI__builtin_islessgreater: 129 case Builtin::BI__builtin_isunordered: 130 if (SemaBuiltinUnorderedCompare(TheCall)) 131 return ExprError(); 132 break; 133 case Builtin::BI__builtin_fpclassify: 134 if (SemaBuiltinFPClassification(TheCall, 6)) 135 return ExprError(); 136 break; 137 case Builtin::BI__builtin_isfinite: 138 case Builtin::BI__builtin_isinf: 139 case Builtin::BI__builtin_isinf_sign: 140 case Builtin::BI__builtin_isnan: 141 case Builtin::BI__builtin_isnormal: 142 if (SemaBuiltinFPClassification(TheCall, 1)) 143 return ExprError(); 144 break; 145 case Builtin::BI__builtin_shufflevector: 146 return SemaBuiltinShuffleVector(TheCall); 147 // TheCall will be freed by the smart pointer here, but that's fine, since 148 // SemaBuiltinShuffleVector guts it, but then doesn't release it. 149 case Builtin::BI__builtin_prefetch: 150 if (SemaBuiltinPrefetch(TheCall)) 151 return ExprError(); 152 break; 153 case Builtin::BI__builtin_object_size: 154 if (SemaBuiltinObjectSize(TheCall)) 155 return ExprError(); 156 break; 157 case Builtin::BI__builtin_longjmp: 158 if (SemaBuiltinLongjmp(TheCall)) 159 return ExprError(); 160 break; 161 162 case Builtin::BI__builtin_classify_type: 163 if (checkArgCount(*this, TheCall, 1)) return true; 164 TheCall->setType(Context.IntTy); 165 break; 166 case Builtin::BI__builtin_constant_p: 167 if (checkArgCount(*this, TheCall, 1)) return true; 168 TheCall->setType(Context.IntTy); 169 break; 170 case Builtin::BI__sync_fetch_and_add: 171 case Builtin::BI__sync_fetch_and_sub: 172 case Builtin::BI__sync_fetch_and_or: 173 case Builtin::BI__sync_fetch_and_and: 174 case Builtin::BI__sync_fetch_and_xor: 175 case Builtin::BI__sync_add_and_fetch: 176 case Builtin::BI__sync_sub_and_fetch: 177 case Builtin::BI__sync_and_and_fetch: 178 case Builtin::BI__sync_or_and_fetch: 179 case Builtin::BI__sync_xor_and_fetch: 180 case Builtin::BI__sync_val_compare_and_swap: 181 case Builtin::BI__sync_bool_compare_and_swap: 182 case Builtin::BI__sync_lock_test_and_set: 183 case Builtin::BI__sync_lock_release: 184 case Builtin::BI__sync_swap: 185 return SemaBuiltinAtomicOverloaded(move(TheCallResult)); 186 } 187 188 // Since the target specific builtins for each arch overlap, only check those 189 // of the arch we are compiling for. 190 if (BuiltinID >= Builtin::FirstTSBuiltin) { 191 switch (Context.Target.getTriple().getArch()) { 192 case llvm::Triple::arm: 193 case llvm::Triple::thumb: 194 if (CheckARMBuiltinFunctionCall(BuiltinID, TheCall)) 195 return ExprError(); 196 break; 197 default: 198 break; 199 } 200 } 201 202 return move(TheCallResult); 203} 204 205// Get the valid immediate range for the specified NEON type code. 206static unsigned RFT(unsigned t, bool shift = false) { 207 bool quad = t & 0x10; 208 209 switch (t & 0x7) { 210 case 0: // i8 211 return shift ? 7 : (8 << (int)quad) - 1; 212 case 1: // i16 213 return shift ? 15 : (4 << (int)quad) - 1; 214 case 2: // i32 215 return shift ? 31 : (2 << (int)quad) - 1; 216 case 3: // i64 217 return shift ? 63 : (1 << (int)quad) - 1; 218 case 4: // f32 219 assert(!shift && "cannot shift float types!"); 220 return (2 << (int)quad) - 1; 221 case 5: // poly8 222 return shift ? 7 : (8 << (int)quad) - 1; 223 case 6: // poly16 224 return shift ? 15 : (4 << (int)quad) - 1; 225 case 7: // float16 226 assert(!shift && "cannot shift float types!"); 227 return (4 << (int)quad) - 1; 228 } 229 return 0; 230} 231 232bool Sema::CheckARMBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) { 233 llvm::APSInt Result; 234 235 unsigned mask = 0; 236 unsigned TV = 0; 237 switch (BuiltinID) { 238#define GET_NEON_OVERLOAD_CHECK 239#include "clang/Basic/arm_neon.inc" 240#undef GET_NEON_OVERLOAD_CHECK 241 } 242 243 // For NEON intrinsics which are overloaded on vector element type, validate 244 // the immediate which specifies which variant to emit. 245 if (mask) { 246 unsigned ArgNo = TheCall->getNumArgs()-1; 247 if (SemaBuiltinConstantArg(TheCall, ArgNo, Result)) 248 return true; 249 250 TV = Result.getLimitedValue(32); 251 if ((TV > 31) || (mask & (1 << TV)) == 0) 252 return Diag(TheCall->getLocStart(), diag::err_invalid_neon_type_code) 253 << TheCall->getArg(ArgNo)->getSourceRange(); 254 } 255 256 // For NEON intrinsics which take an immediate value as part of the 257 // instruction, range check them here. 258 unsigned i = 0, l = 0, u = 0; 259 switch (BuiltinID) { 260 default: return false; 261 case ARM::BI__builtin_arm_ssat: i = 1; l = 1; u = 31; break; 262 case ARM::BI__builtin_arm_usat: i = 1; u = 31; break; 263 case ARM::BI__builtin_arm_vcvtr_f: 264 case ARM::BI__builtin_arm_vcvtr_d: i = 1; u = 1; break; 265#define GET_NEON_IMMEDIATE_CHECK 266#include "clang/Basic/arm_neon.inc" 267#undef GET_NEON_IMMEDIATE_CHECK 268 }; 269 270 // Check that the immediate argument is actually a constant. 271 if (SemaBuiltinConstantArg(TheCall, i, Result)) 272 return true; 273 274 // Range check against the upper/lower values for this isntruction. 275 unsigned Val = Result.getZExtValue(); 276 if (Val < l || Val > (u + l)) 277 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 278 << l << u+l << TheCall->getArg(i)->getSourceRange(); 279 280 // FIXME: VFP Intrinsics should error if VFP not present. 281 return false; 282} 283 284/// CheckFunctionCall - Check a direct function call for various correctness 285/// and safety properties not strictly enforced by the C type system. 286bool Sema::CheckFunctionCall(FunctionDecl *FDecl, CallExpr *TheCall) { 287 // Get the IdentifierInfo* for the called function. 288 IdentifierInfo *FnInfo = FDecl->getIdentifier(); 289 290 // None of the checks below are needed for functions that don't have 291 // simple names (e.g., C++ conversion functions). 292 if (!FnInfo) 293 return false; 294 295 // FIXME: This mechanism should be abstracted to be less fragile and 296 // more efficient. For example, just map function ids to custom 297 // handlers. 298 299 // Printf and scanf checking. 300 for (specific_attr_iterator<FormatAttr> 301 i = FDecl->specific_attr_begin<FormatAttr>(), 302 e = FDecl->specific_attr_end<FormatAttr>(); i != e ; ++i) { 303 304 const FormatAttr *Format = *i; 305 const bool b = Format->getType() == "scanf"; 306 if (b || CheckablePrintfAttr(Format, TheCall)) { 307 bool HasVAListArg = Format->getFirstArg() == 0; 308 CheckPrintfScanfArguments(TheCall, HasVAListArg, 309 Format->getFormatIdx() - 1, 310 HasVAListArg ? 0 : Format->getFirstArg() - 1, 311 !b); 312 } 313 } 314 315 for (specific_attr_iterator<NonNullAttr> 316 i = FDecl->specific_attr_begin<NonNullAttr>(), 317 e = FDecl->specific_attr_end<NonNullAttr>(); i != e; ++i) { 318 CheckNonNullArguments(*i, TheCall->getArgs(), 319 TheCall->getCallee()->getLocStart()); 320 } 321 322 // Builtin handling 323 int CMF = -1; 324 switch (FDecl->getBuiltinID()) { 325 case Builtin::BI__builtin_memset: 326 case Builtin::BI__builtin___memset_chk: 327 case Builtin::BImemset: 328 CMF = CMF_Memset; 329 break; 330 331 case Builtin::BI__builtin_memcpy: 332 case Builtin::BI__builtin___memcpy_chk: 333 case Builtin::BImemcpy: 334 CMF = CMF_Memcpy; 335 break; 336 337 case Builtin::BI__builtin_memmove: 338 case Builtin::BI__builtin___memmove_chk: 339 case Builtin::BImemmove: 340 CMF = CMF_Memmove; 341 break; 342 343 case Builtin::BIstrlcpy: 344 case Builtin::BIstrlcat: 345 CheckStrlcpycatArguments(TheCall, FnInfo); 346 break; 347 348 case Builtin::BI__builtin_memcmp: 349 CMF = CMF_Memcmp; 350 break; 351 352 default: 353 if (FDecl->getLinkage() == ExternalLinkage && 354 (!getLangOptions().CPlusPlus || FDecl->isExternC())) { 355 if (FnInfo->isStr("memset")) 356 CMF = CMF_Memset; 357 else if (FnInfo->isStr("memcpy")) 358 CMF = CMF_Memcpy; 359 else if (FnInfo->isStr("memmove")) 360 CMF = CMF_Memmove; 361 else if (FnInfo->isStr("memcmp")) 362 CMF = CMF_Memcmp; 363 } 364 break; 365 } 366 367 // Memset/memcpy/memmove handling 368 if (CMF != -1) 369 CheckMemaccessArguments(TheCall, CheckedMemoryFunction(CMF), FnInfo); 370 371 return false; 372} 373 374bool Sema::CheckBlockCall(NamedDecl *NDecl, CallExpr *TheCall) { 375 // Printf checking. 376 const FormatAttr *Format = NDecl->getAttr<FormatAttr>(); 377 if (!Format) 378 return false; 379 380 const VarDecl *V = dyn_cast<VarDecl>(NDecl); 381 if (!V) 382 return false; 383 384 QualType Ty = V->getType(); 385 if (!Ty->isBlockPointerType()) 386 return false; 387 388 const bool b = Format->getType() == "scanf"; 389 if (!b && !CheckablePrintfAttr(Format, TheCall)) 390 return false; 391 392 bool HasVAListArg = Format->getFirstArg() == 0; 393 CheckPrintfScanfArguments(TheCall, HasVAListArg, Format->getFormatIdx() - 1, 394 HasVAListArg ? 0 : Format->getFirstArg() - 1, !b); 395 396 return false; 397} 398 399/// SemaBuiltinAtomicOverloaded - We have a call to a function like 400/// __sync_fetch_and_add, which is an overloaded function based on the pointer 401/// type of its first argument. The main ActOnCallExpr routines have already 402/// promoted the types of arguments because all of these calls are prototyped as 403/// void(...). 404/// 405/// This function goes through and does final semantic checking for these 406/// builtins, 407ExprResult 408Sema::SemaBuiltinAtomicOverloaded(ExprResult TheCallResult) { 409 CallExpr *TheCall = (CallExpr *)TheCallResult.get(); 410 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 411 FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl()); 412 413 // Ensure that we have at least one argument to do type inference from. 414 if (TheCall->getNumArgs() < 1) { 415 Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args_at_least) 416 << 0 << 1 << TheCall->getNumArgs() 417 << TheCall->getCallee()->getSourceRange(); 418 return ExprError(); 419 } 420 421 // Inspect the first argument of the atomic builtin. This should always be 422 // a pointer type, whose element is an integral scalar or pointer type. 423 // Because it is a pointer type, we don't have to worry about any implicit 424 // casts here. 425 // FIXME: We don't allow floating point scalars as input. 426 Expr *FirstArg = TheCall->getArg(0); 427 const PointerType *pointerType = FirstArg->getType()->getAs<PointerType>(); 428 if (!pointerType) { 429 Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer) 430 << FirstArg->getType() << FirstArg->getSourceRange(); 431 return ExprError(); 432 } 433 434 QualType ValType = pointerType->getPointeeType(); 435 if (!ValType->isIntegerType() && !ValType->isAnyPointerType() && 436 !ValType->isBlockPointerType()) { 437 Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer_intptr) 438 << FirstArg->getType() << FirstArg->getSourceRange(); 439 return ExprError(); 440 } 441 442 switch (ValType.getObjCLifetime()) { 443 case Qualifiers::OCL_None: 444 case Qualifiers::OCL_ExplicitNone: 445 // okay 446 break; 447 448 case Qualifiers::OCL_Weak: 449 case Qualifiers::OCL_Strong: 450 case Qualifiers::OCL_Autoreleasing: 451 Diag(DRE->getLocStart(), diag::err_arc_atomic_ownership) 452 << ValType << FirstArg->getSourceRange(); 453 return ExprError(); 454 } 455 456 // The majority of builtins return a value, but a few have special return 457 // types, so allow them to override appropriately below. 458 QualType ResultType = ValType; 459 460 // We need to figure out which concrete builtin this maps onto. For example, 461 // __sync_fetch_and_add with a 2 byte object turns into 462 // __sync_fetch_and_add_2. 463#define BUILTIN_ROW(x) \ 464 { Builtin::BI##x##_1, Builtin::BI##x##_2, Builtin::BI##x##_4, \ 465 Builtin::BI##x##_8, Builtin::BI##x##_16 } 466 467 static const unsigned BuiltinIndices[][5] = { 468 BUILTIN_ROW(__sync_fetch_and_add), 469 BUILTIN_ROW(__sync_fetch_and_sub), 470 BUILTIN_ROW(__sync_fetch_and_or), 471 BUILTIN_ROW(__sync_fetch_and_and), 472 BUILTIN_ROW(__sync_fetch_and_xor), 473 474 BUILTIN_ROW(__sync_add_and_fetch), 475 BUILTIN_ROW(__sync_sub_and_fetch), 476 BUILTIN_ROW(__sync_and_and_fetch), 477 BUILTIN_ROW(__sync_or_and_fetch), 478 BUILTIN_ROW(__sync_xor_and_fetch), 479 480 BUILTIN_ROW(__sync_val_compare_and_swap), 481 BUILTIN_ROW(__sync_bool_compare_and_swap), 482 BUILTIN_ROW(__sync_lock_test_and_set), 483 BUILTIN_ROW(__sync_lock_release), 484 BUILTIN_ROW(__sync_swap) 485 }; 486#undef BUILTIN_ROW 487 488 // Determine the index of the size. 489 unsigned SizeIndex; 490 switch (Context.getTypeSizeInChars(ValType).getQuantity()) { 491 case 1: SizeIndex = 0; break; 492 case 2: SizeIndex = 1; break; 493 case 4: SizeIndex = 2; break; 494 case 8: SizeIndex = 3; break; 495 case 16: SizeIndex = 4; break; 496 default: 497 Diag(DRE->getLocStart(), diag::err_atomic_builtin_pointer_size) 498 << FirstArg->getType() << FirstArg->getSourceRange(); 499 return ExprError(); 500 } 501 502 // Each of these builtins has one pointer argument, followed by some number of 503 // values (0, 1 or 2) followed by a potentially empty varags list of stuff 504 // that we ignore. Find out which row of BuiltinIndices to read from as well 505 // as the number of fixed args. 506 unsigned BuiltinID = FDecl->getBuiltinID(); 507 unsigned BuiltinIndex, NumFixed = 1; 508 switch (BuiltinID) { 509 default: assert(0 && "Unknown overloaded atomic builtin!"); 510 case Builtin::BI__sync_fetch_and_add: BuiltinIndex = 0; break; 511 case Builtin::BI__sync_fetch_and_sub: BuiltinIndex = 1; break; 512 case Builtin::BI__sync_fetch_and_or: BuiltinIndex = 2; break; 513 case Builtin::BI__sync_fetch_and_and: BuiltinIndex = 3; break; 514 case Builtin::BI__sync_fetch_and_xor: BuiltinIndex = 4; break; 515 516 case Builtin::BI__sync_add_and_fetch: BuiltinIndex = 5; break; 517 case Builtin::BI__sync_sub_and_fetch: BuiltinIndex = 6; break; 518 case Builtin::BI__sync_and_and_fetch: BuiltinIndex = 7; break; 519 case Builtin::BI__sync_or_and_fetch: BuiltinIndex = 8; break; 520 case Builtin::BI__sync_xor_and_fetch: BuiltinIndex = 9; break; 521 522 case Builtin::BI__sync_val_compare_and_swap: 523 BuiltinIndex = 10; 524 NumFixed = 2; 525 break; 526 case Builtin::BI__sync_bool_compare_and_swap: 527 BuiltinIndex = 11; 528 NumFixed = 2; 529 ResultType = Context.BoolTy; 530 break; 531 case Builtin::BI__sync_lock_test_and_set: BuiltinIndex = 12; break; 532 case Builtin::BI__sync_lock_release: 533 BuiltinIndex = 13; 534 NumFixed = 0; 535 ResultType = Context.VoidTy; 536 break; 537 case Builtin::BI__sync_swap: BuiltinIndex = 14; break; 538 } 539 540 // Now that we know how many fixed arguments we expect, first check that we 541 // have at least that many. 542 if (TheCall->getNumArgs() < 1+NumFixed) { 543 Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args_at_least) 544 << 0 << 1+NumFixed << TheCall->getNumArgs() 545 << TheCall->getCallee()->getSourceRange(); 546 return ExprError(); 547 } 548 549 // Get the decl for the concrete builtin from this, we can tell what the 550 // concrete integer type we should convert to is. 551 unsigned NewBuiltinID = BuiltinIndices[BuiltinIndex][SizeIndex]; 552 const char *NewBuiltinName = Context.BuiltinInfo.GetName(NewBuiltinID); 553 IdentifierInfo *NewBuiltinII = PP.getIdentifierInfo(NewBuiltinName); 554 FunctionDecl *NewBuiltinDecl = 555 cast<FunctionDecl>(LazilyCreateBuiltin(NewBuiltinII, NewBuiltinID, 556 TUScope, false, DRE->getLocStart())); 557 558 // The first argument --- the pointer --- has a fixed type; we 559 // deduce the types of the rest of the arguments accordingly. Walk 560 // the remaining arguments, converting them to the deduced value type. 561 for (unsigned i = 0; i != NumFixed; ++i) { 562 ExprResult Arg = TheCall->getArg(i+1); 563 564 // If the argument is an implicit cast, then there was a promotion due to 565 // "...", just remove it now. 566 if (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(Arg.get())) { 567 Arg = ICE->getSubExpr(); 568 ICE->setSubExpr(0); 569 TheCall->setArg(i+1, Arg.get()); 570 } 571 572 // GCC does an implicit conversion to the pointer or integer ValType. This 573 // can fail in some cases (1i -> int**), check for this error case now. 574 CastKind Kind = CK_Invalid; 575 ExprValueKind VK = VK_RValue; 576 CXXCastPath BasePath; 577 Arg = CheckCastTypes(Arg.get()->getLocStart(), Arg.get()->getSourceRange(), 578 ValType, Arg.take(), Kind, VK, BasePath); 579 if (Arg.isInvalid()) 580 return ExprError(); 581 582 // Okay, we have something that *can* be converted to the right type. Check 583 // to see if there is a potentially weird extension going on here. This can 584 // happen when you do an atomic operation on something like an char* and 585 // pass in 42. The 42 gets converted to char. This is even more strange 586 // for things like 45.123 -> char, etc. 587 // FIXME: Do this check. 588 Arg = ImpCastExprToType(Arg.take(), ValType, Kind, VK, &BasePath); 589 TheCall->setArg(i+1, Arg.get()); 590 } 591 592 // Switch the DeclRefExpr to refer to the new decl. 593 DRE->setDecl(NewBuiltinDecl); 594 DRE->setType(NewBuiltinDecl->getType()); 595 596 // Set the callee in the CallExpr. 597 // FIXME: This leaks the original parens and implicit casts. 598 ExprResult PromotedCall = UsualUnaryConversions(DRE); 599 if (PromotedCall.isInvalid()) 600 return ExprError(); 601 TheCall->setCallee(PromotedCall.take()); 602 603 // Change the result type of the call to match the original value type. This 604 // is arbitrary, but the codegen for these builtins ins design to handle it 605 // gracefully. 606 TheCall->setType(ResultType); 607 608 return move(TheCallResult); 609} 610 611 612/// CheckObjCString - Checks that the argument to the builtin 613/// CFString constructor is correct 614/// Note: It might also make sense to do the UTF-16 conversion here (would 615/// simplify the backend). 616bool Sema::CheckObjCString(Expr *Arg) { 617 Arg = Arg->IgnoreParenCasts(); 618 StringLiteral *Literal = dyn_cast<StringLiteral>(Arg); 619 620 if (!Literal || !Literal->isAscii()) { 621 Diag(Arg->getLocStart(), diag::err_cfstring_literal_not_string_constant) 622 << Arg->getSourceRange(); 623 return true; 624 } 625 626 if (Literal->containsNonAsciiOrNull()) { 627 StringRef String = Literal->getString(); 628 unsigned NumBytes = String.size(); 629 SmallVector<UTF16, 128> ToBuf(NumBytes); 630 const UTF8 *FromPtr = (UTF8 *)String.data(); 631 UTF16 *ToPtr = &ToBuf[0]; 632 633 ConversionResult Result = ConvertUTF8toUTF16(&FromPtr, FromPtr + NumBytes, 634 &ToPtr, ToPtr + NumBytes, 635 strictConversion); 636 // Check for conversion failure. 637 if (Result != conversionOK) 638 Diag(Arg->getLocStart(), 639 diag::warn_cfstring_truncated) << Arg->getSourceRange(); 640 } 641 return false; 642} 643 644/// SemaBuiltinVAStart - Check the arguments to __builtin_va_start for validity. 645/// Emit an error and return true on failure, return false on success. 646bool Sema::SemaBuiltinVAStart(CallExpr *TheCall) { 647 Expr *Fn = TheCall->getCallee(); 648 if (TheCall->getNumArgs() > 2) { 649 Diag(TheCall->getArg(2)->getLocStart(), 650 diag::err_typecheck_call_too_many_args) 651 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 652 << Fn->getSourceRange() 653 << SourceRange(TheCall->getArg(2)->getLocStart(), 654 (*(TheCall->arg_end()-1))->getLocEnd()); 655 return true; 656 } 657 658 if (TheCall->getNumArgs() < 2) { 659 return Diag(TheCall->getLocEnd(), 660 diag::err_typecheck_call_too_few_args_at_least) 661 << 0 /*function call*/ << 2 << TheCall->getNumArgs(); 662 } 663 664 // Determine whether the current function is variadic or not. 665 BlockScopeInfo *CurBlock = getCurBlock(); 666 bool isVariadic; 667 if (CurBlock) 668 isVariadic = CurBlock->TheDecl->isVariadic(); 669 else if (FunctionDecl *FD = getCurFunctionDecl()) 670 isVariadic = FD->isVariadic(); 671 else 672 isVariadic = getCurMethodDecl()->isVariadic(); 673 674 if (!isVariadic) { 675 Diag(Fn->getLocStart(), diag::err_va_start_used_in_non_variadic_function); 676 return true; 677 } 678 679 // Verify that the second argument to the builtin is the last argument of the 680 // current function or method. 681 bool SecondArgIsLastNamedArgument = false; 682 const Expr *Arg = TheCall->getArg(1)->IgnoreParenCasts(); 683 684 if (const DeclRefExpr *DR = dyn_cast<DeclRefExpr>(Arg)) { 685 if (const ParmVarDecl *PV = dyn_cast<ParmVarDecl>(DR->getDecl())) { 686 // FIXME: This isn't correct for methods (results in bogus warning). 687 // Get the last formal in the current function. 688 const ParmVarDecl *LastArg; 689 if (CurBlock) 690 LastArg = *(CurBlock->TheDecl->param_end()-1); 691 else if (FunctionDecl *FD = getCurFunctionDecl()) 692 LastArg = *(FD->param_end()-1); 693 else 694 LastArg = *(getCurMethodDecl()->param_end()-1); 695 SecondArgIsLastNamedArgument = PV == LastArg; 696 } 697 } 698 699 if (!SecondArgIsLastNamedArgument) 700 Diag(TheCall->getArg(1)->getLocStart(), 701 diag::warn_second_parameter_of_va_start_not_last_named_argument); 702 return false; 703} 704 705/// SemaBuiltinUnorderedCompare - Handle functions like __builtin_isgreater and 706/// friends. This is declared to take (...), so we have to check everything. 707bool Sema::SemaBuiltinUnorderedCompare(CallExpr *TheCall) { 708 if (TheCall->getNumArgs() < 2) 709 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 710 << 0 << 2 << TheCall->getNumArgs()/*function call*/; 711 if (TheCall->getNumArgs() > 2) 712 return Diag(TheCall->getArg(2)->getLocStart(), 713 diag::err_typecheck_call_too_many_args) 714 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 715 << SourceRange(TheCall->getArg(2)->getLocStart(), 716 (*(TheCall->arg_end()-1))->getLocEnd()); 717 718 ExprResult OrigArg0 = TheCall->getArg(0); 719 ExprResult OrigArg1 = TheCall->getArg(1); 720 721 // Do standard promotions between the two arguments, returning their common 722 // type. 723 QualType Res = UsualArithmeticConversions(OrigArg0, OrigArg1, false); 724 if (OrigArg0.isInvalid() || OrigArg1.isInvalid()) 725 return true; 726 727 // Make sure any conversions are pushed back into the call; this is 728 // type safe since unordered compare builtins are declared as "_Bool 729 // foo(...)". 730 TheCall->setArg(0, OrigArg0.get()); 731 TheCall->setArg(1, OrigArg1.get()); 732 733 if (OrigArg0.get()->isTypeDependent() || OrigArg1.get()->isTypeDependent()) 734 return false; 735 736 // If the common type isn't a real floating type, then the arguments were 737 // invalid for this operation. 738 if (!Res->isRealFloatingType()) 739 return Diag(OrigArg0.get()->getLocStart(), 740 diag::err_typecheck_call_invalid_ordered_compare) 741 << OrigArg0.get()->getType() << OrigArg1.get()->getType() 742 << SourceRange(OrigArg0.get()->getLocStart(), OrigArg1.get()->getLocEnd()); 743 744 return false; 745} 746 747/// SemaBuiltinSemaBuiltinFPClassification - Handle functions like 748/// __builtin_isnan and friends. This is declared to take (...), so we have 749/// to check everything. We expect the last argument to be a floating point 750/// value. 751bool Sema::SemaBuiltinFPClassification(CallExpr *TheCall, unsigned NumArgs) { 752 if (TheCall->getNumArgs() < NumArgs) 753 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 754 << 0 << NumArgs << TheCall->getNumArgs()/*function call*/; 755 if (TheCall->getNumArgs() > NumArgs) 756 return Diag(TheCall->getArg(NumArgs)->getLocStart(), 757 diag::err_typecheck_call_too_many_args) 758 << 0 /*function call*/ << NumArgs << TheCall->getNumArgs() 759 << SourceRange(TheCall->getArg(NumArgs)->getLocStart(), 760 (*(TheCall->arg_end()-1))->getLocEnd()); 761 762 Expr *OrigArg = TheCall->getArg(NumArgs-1); 763 764 if (OrigArg->isTypeDependent()) 765 return false; 766 767 // This operation requires a non-_Complex floating-point number. 768 if (!OrigArg->getType()->isRealFloatingType()) 769 return Diag(OrigArg->getLocStart(), 770 diag::err_typecheck_call_invalid_unary_fp) 771 << OrigArg->getType() << OrigArg->getSourceRange(); 772 773 // If this is an implicit conversion from float -> double, remove it. 774 if (ImplicitCastExpr *Cast = dyn_cast<ImplicitCastExpr>(OrigArg)) { 775 Expr *CastArg = Cast->getSubExpr(); 776 if (CastArg->getType()->isSpecificBuiltinType(BuiltinType::Float)) { 777 assert(Cast->getType()->isSpecificBuiltinType(BuiltinType::Double) && 778 "promotion from float to double is the only expected cast here"); 779 Cast->setSubExpr(0); 780 TheCall->setArg(NumArgs-1, CastArg); 781 OrigArg = CastArg; 782 } 783 } 784 785 return false; 786} 787 788/// SemaBuiltinShuffleVector - Handle __builtin_shufflevector. 789// This is declared to take (...), so we have to check everything. 790ExprResult Sema::SemaBuiltinShuffleVector(CallExpr *TheCall) { 791 if (TheCall->getNumArgs() < 2) 792 return ExprError(Diag(TheCall->getLocEnd(), 793 diag::err_typecheck_call_too_few_args_at_least) 794 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 795 << TheCall->getSourceRange()); 796 797 // Determine which of the following types of shufflevector we're checking: 798 // 1) unary, vector mask: (lhs, mask) 799 // 2) binary, vector mask: (lhs, rhs, mask) 800 // 3) binary, scalar mask: (lhs, rhs, index, ..., index) 801 QualType resType = TheCall->getArg(0)->getType(); 802 unsigned numElements = 0; 803 804 if (!TheCall->getArg(0)->isTypeDependent() && 805 !TheCall->getArg(1)->isTypeDependent()) { 806 QualType LHSType = TheCall->getArg(0)->getType(); 807 QualType RHSType = TheCall->getArg(1)->getType(); 808 809 if (!LHSType->isVectorType() || !RHSType->isVectorType()) { 810 Diag(TheCall->getLocStart(), diag::err_shufflevector_non_vector) 811 << SourceRange(TheCall->getArg(0)->getLocStart(), 812 TheCall->getArg(1)->getLocEnd()); 813 return ExprError(); 814 } 815 816 numElements = LHSType->getAs<VectorType>()->getNumElements(); 817 unsigned numResElements = TheCall->getNumArgs() - 2; 818 819 // Check to see if we have a call with 2 vector arguments, the unary shuffle 820 // with mask. If so, verify that RHS is an integer vector type with the 821 // same number of elts as lhs. 822 if (TheCall->getNumArgs() == 2) { 823 if (!RHSType->hasIntegerRepresentation() || 824 RHSType->getAs<VectorType>()->getNumElements() != numElements) 825 Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector) 826 << SourceRange(TheCall->getArg(1)->getLocStart(), 827 TheCall->getArg(1)->getLocEnd()); 828 numResElements = numElements; 829 } 830 else if (!Context.hasSameUnqualifiedType(LHSType, RHSType)) { 831 Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector) 832 << SourceRange(TheCall->getArg(0)->getLocStart(), 833 TheCall->getArg(1)->getLocEnd()); 834 return ExprError(); 835 } else if (numElements != numResElements) { 836 QualType eltType = LHSType->getAs<VectorType>()->getElementType(); 837 resType = Context.getVectorType(eltType, numResElements, 838 VectorType::GenericVector); 839 } 840 } 841 842 for (unsigned i = 2; i < TheCall->getNumArgs(); i++) { 843 if (TheCall->getArg(i)->isTypeDependent() || 844 TheCall->getArg(i)->isValueDependent()) 845 continue; 846 847 llvm::APSInt Result(32); 848 if (!TheCall->getArg(i)->isIntegerConstantExpr(Result, Context)) 849 return ExprError(Diag(TheCall->getLocStart(), 850 diag::err_shufflevector_nonconstant_argument) 851 << TheCall->getArg(i)->getSourceRange()); 852 853 if (Result.getActiveBits() > 64 || Result.getZExtValue() >= numElements*2) 854 return ExprError(Diag(TheCall->getLocStart(), 855 diag::err_shufflevector_argument_too_large) 856 << TheCall->getArg(i)->getSourceRange()); 857 } 858 859 SmallVector<Expr*, 32> exprs; 860 861 for (unsigned i = 0, e = TheCall->getNumArgs(); i != e; i++) { 862 exprs.push_back(TheCall->getArg(i)); 863 TheCall->setArg(i, 0); 864 } 865 866 return Owned(new (Context) ShuffleVectorExpr(Context, exprs.begin(), 867 exprs.size(), resType, 868 TheCall->getCallee()->getLocStart(), 869 TheCall->getRParenLoc())); 870} 871 872/// SemaBuiltinPrefetch - Handle __builtin_prefetch. 873// This is declared to take (const void*, ...) and can take two 874// optional constant int args. 875bool Sema::SemaBuiltinPrefetch(CallExpr *TheCall) { 876 unsigned NumArgs = TheCall->getNumArgs(); 877 878 if (NumArgs > 3) 879 return Diag(TheCall->getLocEnd(), 880 diag::err_typecheck_call_too_many_args_at_most) 881 << 0 /*function call*/ << 3 << NumArgs 882 << TheCall->getSourceRange(); 883 884 // Argument 0 is checked for us and the remaining arguments must be 885 // constant integers. 886 for (unsigned i = 1; i != NumArgs; ++i) { 887 Expr *Arg = TheCall->getArg(i); 888 889 llvm::APSInt Result; 890 if (SemaBuiltinConstantArg(TheCall, i, Result)) 891 return true; 892 893 // FIXME: gcc issues a warning and rewrites these to 0. These 894 // seems especially odd for the third argument since the default 895 // is 3. 896 if (i == 1) { 897 if (Result.getLimitedValue() > 1) 898 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 899 << "0" << "1" << Arg->getSourceRange(); 900 } else { 901 if (Result.getLimitedValue() > 3) 902 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 903 << "0" << "3" << Arg->getSourceRange(); 904 } 905 } 906 907 return false; 908} 909 910/// SemaBuiltinConstantArg - Handle a check if argument ArgNum of CallExpr 911/// TheCall is a constant expression. 912bool Sema::SemaBuiltinConstantArg(CallExpr *TheCall, int ArgNum, 913 llvm::APSInt &Result) { 914 Expr *Arg = TheCall->getArg(ArgNum); 915 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 916 FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl()); 917 918 if (Arg->isTypeDependent() || Arg->isValueDependent()) return false; 919 920 if (!Arg->isIntegerConstantExpr(Result, Context)) 921 return Diag(TheCall->getLocStart(), diag::err_constant_integer_arg_type) 922 << FDecl->getDeclName() << Arg->getSourceRange(); 923 924 return false; 925} 926 927/// SemaBuiltinObjectSize - Handle __builtin_object_size(void *ptr, 928/// int type). This simply type checks that type is one of the defined 929/// constants (0-3). 930// For compatibility check 0-3, llvm only handles 0 and 2. 931bool Sema::SemaBuiltinObjectSize(CallExpr *TheCall) { 932 llvm::APSInt Result; 933 934 // Check constant-ness first. 935 if (SemaBuiltinConstantArg(TheCall, 1, Result)) 936 return true; 937 938 Expr *Arg = TheCall->getArg(1); 939 if (Result.getSExtValue() < 0 || Result.getSExtValue() > 3) { 940 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 941 << "0" << "3" << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 942 } 943 944 return false; 945} 946 947/// SemaBuiltinLongjmp - Handle __builtin_longjmp(void *env[5], int val). 948/// This checks that val is a constant 1. 949bool Sema::SemaBuiltinLongjmp(CallExpr *TheCall) { 950 Expr *Arg = TheCall->getArg(1); 951 llvm::APSInt Result; 952 953 // TODO: This is less than ideal. Overload this to take a value. 954 if (SemaBuiltinConstantArg(TheCall, 1, Result)) 955 return true; 956 957 if (Result != 1) 958 return Diag(TheCall->getLocStart(), diag::err_builtin_longjmp_invalid_val) 959 << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 960 961 return false; 962} 963 964// Handle i > 1 ? "x" : "y", recursively. 965bool Sema::SemaCheckStringLiteral(const Expr *E, const CallExpr *TheCall, 966 bool HasVAListArg, 967 unsigned format_idx, unsigned firstDataArg, 968 bool isPrintf) { 969 tryAgain: 970 if (E->isTypeDependent() || E->isValueDependent()) 971 return false; 972 973 E = E->IgnoreParens(); 974 975 switch (E->getStmtClass()) { 976 case Stmt::BinaryConditionalOperatorClass: 977 case Stmt::ConditionalOperatorClass: { 978 const AbstractConditionalOperator *C = cast<AbstractConditionalOperator>(E); 979 return SemaCheckStringLiteral(C->getTrueExpr(), TheCall, HasVAListArg, 980 format_idx, firstDataArg, isPrintf) 981 && SemaCheckStringLiteral(C->getFalseExpr(), TheCall, HasVAListArg, 982 format_idx, firstDataArg, isPrintf); 983 } 984 985 case Stmt::IntegerLiteralClass: 986 // Technically -Wformat-nonliteral does not warn about this case. 987 // The behavior of printf and friends in this case is implementation 988 // dependent. Ideally if the format string cannot be null then 989 // it should have a 'nonnull' attribute in the function prototype. 990 return true; 991 992 case Stmt::ImplicitCastExprClass: { 993 E = cast<ImplicitCastExpr>(E)->getSubExpr(); 994 goto tryAgain; 995 } 996 997 case Stmt::OpaqueValueExprClass: 998 if (const Expr *src = cast<OpaqueValueExpr>(E)->getSourceExpr()) { 999 E = src; 1000 goto tryAgain; 1001 } 1002 return false; 1003 1004 case Stmt::PredefinedExprClass: 1005 // While __func__, etc., are technically not string literals, they 1006 // cannot contain format specifiers and thus are not a security 1007 // liability. 1008 return true; 1009 1010 case Stmt::DeclRefExprClass: { 1011 const DeclRefExpr *DR = cast<DeclRefExpr>(E); 1012 1013 // As an exception, do not flag errors for variables binding to 1014 // const string literals. 1015 if (const VarDecl *VD = dyn_cast<VarDecl>(DR->getDecl())) { 1016 bool isConstant = false; 1017 QualType T = DR->getType(); 1018 1019 if (const ArrayType *AT = Context.getAsArrayType(T)) { 1020 isConstant = AT->getElementType().isConstant(Context); 1021 } else if (const PointerType *PT = T->getAs<PointerType>()) { 1022 isConstant = T.isConstant(Context) && 1023 PT->getPointeeType().isConstant(Context); 1024 } 1025 1026 if (isConstant) { 1027 if (const Expr *Init = VD->getAnyInitializer()) 1028 return SemaCheckStringLiteral(Init, TheCall, 1029 HasVAListArg, format_idx, firstDataArg, 1030 isPrintf); 1031 } 1032 1033 // For vprintf* functions (i.e., HasVAListArg==true), we add a 1034 // special check to see if the format string is a function parameter 1035 // of the function calling the printf function. If the function 1036 // has an attribute indicating it is a printf-like function, then we 1037 // should suppress warnings concerning non-literals being used in a call 1038 // to a vprintf function. For example: 1039 // 1040 // void 1041 // logmessage(char const *fmt __attribute__ (format (printf, 1, 2)), ...){ 1042 // va_list ap; 1043 // va_start(ap, fmt); 1044 // vprintf(fmt, ap); // Do NOT emit a warning about "fmt". 1045 // ... 1046 // 1047 // 1048 // FIXME: We don't have full attribute support yet, so just check to see 1049 // if the argument is a DeclRefExpr that references a parameter. We'll 1050 // add proper support for checking the attribute later. 1051 if (HasVAListArg) 1052 if (isa<ParmVarDecl>(VD)) 1053 return true; 1054 } 1055 1056 return false; 1057 } 1058 1059 case Stmt::CallExprClass: { 1060 const CallExpr *CE = cast<CallExpr>(E); 1061 if (const ImplicitCastExpr *ICE 1062 = dyn_cast<ImplicitCastExpr>(CE->getCallee())) { 1063 if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(ICE->getSubExpr())) { 1064 if (const FunctionDecl *FD = dyn_cast<FunctionDecl>(DRE->getDecl())) { 1065 if (const FormatArgAttr *FA = FD->getAttr<FormatArgAttr>()) { 1066 unsigned ArgIndex = FA->getFormatIdx(); 1067 const Expr *Arg = CE->getArg(ArgIndex - 1); 1068 1069 return SemaCheckStringLiteral(Arg, TheCall, HasVAListArg, 1070 format_idx, firstDataArg, isPrintf); 1071 } 1072 } 1073 } 1074 } 1075 1076 return false; 1077 } 1078 case Stmt::ObjCStringLiteralClass: 1079 case Stmt::StringLiteralClass: { 1080 const StringLiteral *StrE = NULL; 1081 1082 if (const ObjCStringLiteral *ObjCFExpr = dyn_cast<ObjCStringLiteral>(E)) 1083 StrE = ObjCFExpr->getString(); 1084 else 1085 StrE = cast<StringLiteral>(E); 1086 1087 if (StrE) { 1088 CheckFormatString(StrE, E, TheCall, HasVAListArg, format_idx, 1089 firstDataArg, isPrintf); 1090 return true; 1091 } 1092 1093 return false; 1094 } 1095 1096 default: 1097 return false; 1098 } 1099} 1100 1101void 1102Sema::CheckNonNullArguments(const NonNullAttr *NonNull, 1103 const Expr * const *ExprArgs, 1104 SourceLocation CallSiteLoc) { 1105 for (NonNullAttr::args_iterator i = NonNull->args_begin(), 1106 e = NonNull->args_end(); 1107 i != e; ++i) { 1108 const Expr *ArgExpr = ExprArgs[*i]; 1109 if (ArgExpr->isNullPointerConstant(Context, 1110 Expr::NPC_ValueDependentIsNotNull)) 1111 Diag(CallSiteLoc, diag::warn_null_arg) << ArgExpr->getSourceRange(); 1112 } 1113} 1114 1115/// CheckPrintfScanfArguments - Check calls to printf and scanf (and similar 1116/// functions) for correct use of format strings. 1117void 1118Sema::CheckPrintfScanfArguments(const CallExpr *TheCall, bool HasVAListArg, 1119 unsigned format_idx, unsigned firstDataArg, 1120 bool isPrintf) { 1121 1122 const Expr *Fn = TheCall->getCallee(); 1123 1124 // The way the format attribute works in GCC, the implicit this argument 1125 // of member functions is counted. However, it doesn't appear in our own 1126 // lists, so decrement format_idx in that case. 1127 if (isa<CXXMemberCallExpr>(TheCall)) { 1128 const CXXMethodDecl *method_decl = 1129 dyn_cast<CXXMethodDecl>(TheCall->getCalleeDecl()); 1130 if (method_decl && method_decl->isInstance()) { 1131 // Catch a format attribute mistakenly referring to the object argument. 1132 if (format_idx == 0) 1133 return; 1134 --format_idx; 1135 if(firstDataArg != 0) 1136 --firstDataArg; 1137 } 1138 } 1139 1140 // CHECK: printf/scanf-like function is called with no format string. 1141 if (format_idx >= TheCall->getNumArgs()) { 1142 Diag(TheCall->getRParenLoc(), diag::warn_missing_format_string) 1143 << Fn->getSourceRange(); 1144 return; 1145 } 1146 1147 const Expr *OrigFormatExpr = TheCall->getArg(format_idx)->IgnoreParenCasts(); 1148 1149 // CHECK: format string is not a string literal. 1150 // 1151 // Dynamically generated format strings are difficult to 1152 // automatically vet at compile time. Requiring that format strings 1153 // are string literals: (1) permits the checking of format strings by 1154 // the compiler and thereby (2) can practically remove the source of 1155 // many format string exploits. 1156 1157 // Format string can be either ObjC string (e.g. @"%d") or 1158 // C string (e.g. "%d") 1159 // ObjC string uses the same format specifiers as C string, so we can use 1160 // the same format string checking logic for both ObjC and C strings. 1161 if (SemaCheckStringLiteral(OrigFormatExpr, TheCall, HasVAListArg, format_idx, 1162 firstDataArg, isPrintf)) 1163 return; // Literal format string found, check done! 1164 1165 // If there are no arguments specified, warn with -Wformat-security, otherwise 1166 // warn only with -Wformat-nonliteral. 1167 if (TheCall->getNumArgs() == format_idx+1) 1168 Diag(TheCall->getArg(format_idx)->getLocStart(), 1169 diag::warn_format_nonliteral_noargs) 1170 << OrigFormatExpr->getSourceRange(); 1171 else 1172 Diag(TheCall->getArg(format_idx)->getLocStart(), 1173 diag::warn_format_nonliteral) 1174 << OrigFormatExpr->getSourceRange(); 1175} 1176 1177namespace { 1178class CheckFormatHandler : public analyze_format_string::FormatStringHandler { 1179protected: 1180 Sema &S; 1181 const StringLiteral *FExpr; 1182 const Expr *OrigFormatExpr; 1183 const unsigned FirstDataArg; 1184 const unsigned NumDataArgs; 1185 const bool IsObjCLiteral; 1186 const char *Beg; // Start of format string. 1187 const bool HasVAListArg; 1188 const CallExpr *TheCall; 1189 unsigned FormatIdx; 1190 llvm::BitVector CoveredArgs; 1191 bool usesPositionalArgs; 1192 bool atFirstArg; 1193public: 1194 CheckFormatHandler(Sema &s, const StringLiteral *fexpr, 1195 const Expr *origFormatExpr, unsigned firstDataArg, 1196 unsigned numDataArgs, bool isObjCLiteral, 1197 const char *beg, bool hasVAListArg, 1198 const CallExpr *theCall, unsigned formatIdx) 1199 : S(s), FExpr(fexpr), OrigFormatExpr(origFormatExpr), 1200 FirstDataArg(firstDataArg), 1201 NumDataArgs(numDataArgs), 1202 IsObjCLiteral(isObjCLiteral), Beg(beg), 1203 HasVAListArg(hasVAListArg), 1204 TheCall(theCall), FormatIdx(formatIdx), 1205 usesPositionalArgs(false), atFirstArg(true) { 1206 CoveredArgs.resize(numDataArgs); 1207 CoveredArgs.reset(); 1208 } 1209 1210 void DoneProcessing(); 1211 1212 void HandleIncompleteSpecifier(const char *startSpecifier, 1213 unsigned specifierLen); 1214 1215 virtual void HandleInvalidPosition(const char *startSpecifier, 1216 unsigned specifierLen, 1217 analyze_format_string::PositionContext p); 1218 1219 virtual void HandleZeroPosition(const char *startPos, unsigned posLen); 1220 1221 void HandleNullChar(const char *nullCharacter); 1222 1223protected: 1224 bool HandleInvalidConversionSpecifier(unsigned argIndex, SourceLocation Loc, 1225 const char *startSpec, 1226 unsigned specifierLen, 1227 const char *csStart, unsigned csLen); 1228 1229 SourceRange getFormatStringRange(); 1230 CharSourceRange getSpecifierRange(const char *startSpecifier, 1231 unsigned specifierLen); 1232 SourceLocation getLocationOfByte(const char *x); 1233 1234 const Expr *getDataArg(unsigned i) const; 1235 1236 bool CheckNumArgs(const analyze_format_string::FormatSpecifier &FS, 1237 const analyze_format_string::ConversionSpecifier &CS, 1238 const char *startSpecifier, unsigned specifierLen, 1239 unsigned argIndex); 1240}; 1241} 1242 1243SourceRange CheckFormatHandler::getFormatStringRange() { 1244 return OrigFormatExpr->getSourceRange(); 1245} 1246 1247CharSourceRange CheckFormatHandler:: 1248getSpecifierRange(const char *startSpecifier, unsigned specifierLen) { 1249 SourceLocation Start = getLocationOfByte(startSpecifier); 1250 SourceLocation End = getLocationOfByte(startSpecifier + specifierLen - 1); 1251 1252 // Advance the end SourceLocation by one due to half-open ranges. 1253 End = End.getFileLocWithOffset(1); 1254 1255 return CharSourceRange::getCharRange(Start, End); 1256} 1257 1258SourceLocation CheckFormatHandler::getLocationOfByte(const char *x) { 1259 return S.getLocationOfStringLiteralByte(FExpr, x - Beg); 1260} 1261 1262void CheckFormatHandler::HandleIncompleteSpecifier(const char *startSpecifier, 1263 unsigned specifierLen){ 1264 SourceLocation Loc = getLocationOfByte(startSpecifier); 1265 S.Diag(Loc, diag::warn_printf_incomplete_specifier) 1266 << getSpecifierRange(startSpecifier, specifierLen); 1267} 1268 1269void 1270CheckFormatHandler::HandleInvalidPosition(const char *startPos, unsigned posLen, 1271 analyze_format_string::PositionContext p) { 1272 SourceLocation Loc = getLocationOfByte(startPos); 1273 S.Diag(Loc, diag::warn_format_invalid_positional_specifier) 1274 << (unsigned) p << getSpecifierRange(startPos, posLen); 1275} 1276 1277void CheckFormatHandler::HandleZeroPosition(const char *startPos, 1278 unsigned posLen) { 1279 SourceLocation Loc = getLocationOfByte(startPos); 1280 S.Diag(Loc, diag::warn_format_zero_positional_specifier) 1281 << getSpecifierRange(startPos, posLen); 1282} 1283 1284void CheckFormatHandler::HandleNullChar(const char *nullCharacter) { 1285 if (!IsObjCLiteral) { 1286 // The presence of a null character is likely an error. 1287 S.Diag(getLocationOfByte(nullCharacter), 1288 diag::warn_printf_format_string_contains_null_char) 1289 << getFormatStringRange(); 1290 } 1291} 1292 1293const Expr *CheckFormatHandler::getDataArg(unsigned i) const { 1294 return TheCall->getArg(FirstDataArg + i); 1295} 1296 1297void CheckFormatHandler::DoneProcessing() { 1298 // Does the number of data arguments exceed the number of 1299 // format conversions in the format string? 1300 if (!HasVAListArg) { 1301 // Find any arguments that weren't covered. 1302 CoveredArgs.flip(); 1303 signed notCoveredArg = CoveredArgs.find_first(); 1304 if (notCoveredArg >= 0) { 1305 assert((unsigned)notCoveredArg < NumDataArgs); 1306 S.Diag(getDataArg((unsigned) notCoveredArg)->getLocStart(), 1307 diag::warn_printf_data_arg_not_used) 1308 << getFormatStringRange(); 1309 } 1310 } 1311} 1312 1313bool 1314CheckFormatHandler::HandleInvalidConversionSpecifier(unsigned argIndex, 1315 SourceLocation Loc, 1316 const char *startSpec, 1317 unsigned specifierLen, 1318 const char *csStart, 1319 unsigned csLen) { 1320 1321 bool keepGoing = true; 1322 if (argIndex < NumDataArgs) { 1323 // Consider the argument coverered, even though the specifier doesn't 1324 // make sense. 1325 CoveredArgs.set(argIndex); 1326 } 1327 else { 1328 // If argIndex exceeds the number of data arguments we 1329 // don't issue a warning because that is just a cascade of warnings (and 1330 // they may have intended '%%' anyway). We don't want to continue processing 1331 // the format string after this point, however, as we will like just get 1332 // gibberish when trying to match arguments. 1333 keepGoing = false; 1334 } 1335 1336 S.Diag(Loc, diag::warn_format_invalid_conversion) 1337 << StringRef(csStart, csLen) 1338 << getSpecifierRange(startSpec, specifierLen); 1339 1340 return keepGoing; 1341} 1342 1343bool 1344CheckFormatHandler::CheckNumArgs( 1345 const analyze_format_string::FormatSpecifier &FS, 1346 const analyze_format_string::ConversionSpecifier &CS, 1347 const char *startSpecifier, unsigned specifierLen, unsigned argIndex) { 1348 1349 if (argIndex >= NumDataArgs) { 1350 if (FS.usesPositionalArg()) { 1351 S.Diag(getLocationOfByte(CS.getStart()), 1352 diag::warn_printf_positional_arg_exceeds_data_args) 1353 << (argIndex+1) << NumDataArgs 1354 << getSpecifierRange(startSpecifier, specifierLen); 1355 } 1356 else { 1357 S.Diag(getLocationOfByte(CS.getStart()), 1358 diag::warn_printf_insufficient_data_args) 1359 << getSpecifierRange(startSpecifier, specifierLen); 1360 } 1361 1362 return false; 1363 } 1364 return true; 1365} 1366 1367//===--- CHECK: Printf format string checking ------------------------------===// 1368 1369namespace { 1370class CheckPrintfHandler : public CheckFormatHandler { 1371public: 1372 CheckPrintfHandler(Sema &s, const StringLiteral *fexpr, 1373 const Expr *origFormatExpr, unsigned firstDataArg, 1374 unsigned numDataArgs, bool isObjCLiteral, 1375 const char *beg, bool hasVAListArg, 1376 const CallExpr *theCall, unsigned formatIdx) 1377 : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg, 1378 numDataArgs, isObjCLiteral, beg, hasVAListArg, 1379 theCall, formatIdx) {} 1380 1381 1382 bool HandleInvalidPrintfConversionSpecifier( 1383 const analyze_printf::PrintfSpecifier &FS, 1384 const char *startSpecifier, 1385 unsigned specifierLen); 1386 1387 bool HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier &FS, 1388 const char *startSpecifier, 1389 unsigned specifierLen); 1390 1391 bool HandleAmount(const analyze_format_string::OptionalAmount &Amt, unsigned k, 1392 const char *startSpecifier, unsigned specifierLen); 1393 void HandleInvalidAmount(const analyze_printf::PrintfSpecifier &FS, 1394 const analyze_printf::OptionalAmount &Amt, 1395 unsigned type, 1396 const char *startSpecifier, unsigned specifierLen); 1397 void HandleFlag(const analyze_printf::PrintfSpecifier &FS, 1398 const analyze_printf::OptionalFlag &flag, 1399 const char *startSpecifier, unsigned specifierLen); 1400 void HandleIgnoredFlag(const analyze_printf::PrintfSpecifier &FS, 1401 const analyze_printf::OptionalFlag &ignoredFlag, 1402 const analyze_printf::OptionalFlag &flag, 1403 const char *startSpecifier, unsigned specifierLen); 1404}; 1405} 1406 1407bool CheckPrintfHandler::HandleInvalidPrintfConversionSpecifier( 1408 const analyze_printf::PrintfSpecifier &FS, 1409 const char *startSpecifier, 1410 unsigned specifierLen) { 1411 const analyze_printf::PrintfConversionSpecifier &CS = 1412 FS.getConversionSpecifier(); 1413 1414 return HandleInvalidConversionSpecifier(FS.getArgIndex(), 1415 getLocationOfByte(CS.getStart()), 1416 startSpecifier, specifierLen, 1417 CS.getStart(), CS.getLength()); 1418} 1419 1420bool CheckPrintfHandler::HandleAmount( 1421 const analyze_format_string::OptionalAmount &Amt, 1422 unsigned k, const char *startSpecifier, 1423 unsigned specifierLen) { 1424 1425 if (Amt.hasDataArgument()) { 1426 if (!HasVAListArg) { 1427 unsigned argIndex = Amt.getArgIndex(); 1428 if (argIndex >= NumDataArgs) { 1429 S.Diag(getLocationOfByte(Amt.getStart()), 1430 diag::warn_printf_asterisk_missing_arg) 1431 << k << getSpecifierRange(startSpecifier, specifierLen); 1432 // Don't do any more checking. We will just emit 1433 // spurious errors. 1434 return false; 1435 } 1436 1437 // Type check the data argument. It should be an 'int'. 1438 // Although not in conformance with C99, we also allow the argument to be 1439 // an 'unsigned int' as that is a reasonably safe case. GCC also 1440 // doesn't emit a warning for that case. 1441 CoveredArgs.set(argIndex); 1442 const Expr *Arg = getDataArg(argIndex); 1443 QualType T = Arg->getType(); 1444 1445 const analyze_printf::ArgTypeResult &ATR = Amt.getArgType(S.Context); 1446 assert(ATR.isValid()); 1447 1448 if (!ATR.matchesType(S.Context, T)) { 1449 S.Diag(getLocationOfByte(Amt.getStart()), 1450 diag::warn_printf_asterisk_wrong_type) 1451 << k 1452 << ATR.getRepresentativeType(S.Context) << T 1453 << getSpecifierRange(startSpecifier, specifierLen) 1454 << Arg->getSourceRange(); 1455 // Don't do any more checking. We will just emit 1456 // spurious errors. 1457 return false; 1458 } 1459 } 1460 } 1461 return true; 1462} 1463 1464void CheckPrintfHandler::HandleInvalidAmount( 1465 const analyze_printf::PrintfSpecifier &FS, 1466 const analyze_printf::OptionalAmount &Amt, 1467 unsigned type, 1468 const char *startSpecifier, 1469 unsigned specifierLen) { 1470 const analyze_printf::PrintfConversionSpecifier &CS = 1471 FS.getConversionSpecifier(); 1472 switch (Amt.getHowSpecified()) { 1473 case analyze_printf::OptionalAmount::Constant: 1474 S.Diag(getLocationOfByte(Amt.getStart()), 1475 diag::warn_printf_nonsensical_optional_amount) 1476 << type 1477 << CS.toString() 1478 << getSpecifierRange(startSpecifier, specifierLen) 1479 << FixItHint::CreateRemoval(getSpecifierRange(Amt.getStart(), 1480 Amt.getConstantLength())); 1481 break; 1482 1483 default: 1484 S.Diag(getLocationOfByte(Amt.getStart()), 1485 diag::warn_printf_nonsensical_optional_amount) 1486 << type 1487 << CS.toString() 1488 << getSpecifierRange(startSpecifier, specifierLen); 1489 break; 1490 } 1491} 1492 1493void CheckPrintfHandler::HandleFlag(const analyze_printf::PrintfSpecifier &FS, 1494 const analyze_printf::OptionalFlag &flag, 1495 const char *startSpecifier, 1496 unsigned specifierLen) { 1497 // Warn about pointless flag with a fixit removal. 1498 const analyze_printf::PrintfConversionSpecifier &CS = 1499 FS.getConversionSpecifier(); 1500 S.Diag(getLocationOfByte(flag.getPosition()), 1501 diag::warn_printf_nonsensical_flag) 1502 << flag.toString() << CS.toString() 1503 << getSpecifierRange(startSpecifier, specifierLen) 1504 << FixItHint::CreateRemoval(getSpecifierRange(flag.getPosition(), 1)); 1505} 1506 1507void CheckPrintfHandler::HandleIgnoredFlag( 1508 const analyze_printf::PrintfSpecifier &FS, 1509 const analyze_printf::OptionalFlag &ignoredFlag, 1510 const analyze_printf::OptionalFlag &flag, 1511 const char *startSpecifier, 1512 unsigned specifierLen) { 1513 // Warn about ignored flag with a fixit removal. 1514 S.Diag(getLocationOfByte(ignoredFlag.getPosition()), 1515 diag::warn_printf_ignored_flag) 1516 << ignoredFlag.toString() << flag.toString() 1517 << getSpecifierRange(startSpecifier, specifierLen) 1518 << FixItHint::CreateRemoval(getSpecifierRange( 1519 ignoredFlag.getPosition(), 1)); 1520} 1521 1522bool 1523CheckPrintfHandler::HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier 1524 &FS, 1525 const char *startSpecifier, 1526 unsigned specifierLen) { 1527 1528 using namespace analyze_format_string; 1529 using namespace analyze_printf; 1530 const PrintfConversionSpecifier &CS = FS.getConversionSpecifier(); 1531 1532 if (FS.consumesDataArgument()) { 1533 if (atFirstArg) { 1534 atFirstArg = false; 1535 usesPositionalArgs = FS.usesPositionalArg(); 1536 } 1537 else if (usesPositionalArgs != FS.usesPositionalArg()) { 1538 // Cannot mix-and-match positional and non-positional arguments. 1539 S.Diag(getLocationOfByte(CS.getStart()), 1540 diag::warn_format_mix_positional_nonpositional_args) 1541 << getSpecifierRange(startSpecifier, specifierLen); 1542 return false; 1543 } 1544 } 1545 1546 // First check if the field width, precision, and conversion specifier 1547 // have matching data arguments. 1548 if (!HandleAmount(FS.getFieldWidth(), /* field width */ 0, 1549 startSpecifier, specifierLen)) { 1550 return false; 1551 } 1552 1553 if (!HandleAmount(FS.getPrecision(), /* precision */ 1, 1554 startSpecifier, specifierLen)) { 1555 return false; 1556 } 1557 1558 if (!CS.consumesDataArgument()) { 1559 // FIXME: Technically specifying a precision or field width here 1560 // makes no sense. Worth issuing a warning at some point. 1561 return true; 1562 } 1563 1564 // Consume the argument. 1565 unsigned argIndex = FS.getArgIndex(); 1566 if (argIndex < NumDataArgs) { 1567 // The check to see if the argIndex is valid will come later. 1568 // We set the bit here because we may exit early from this 1569 // function if we encounter some other error. 1570 CoveredArgs.set(argIndex); 1571 } 1572 1573 // Check for using an Objective-C specific conversion specifier 1574 // in a non-ObjC literal. 1575 if (!IsObjCLiteral && CS.isObjCArg()) { 1576 return HandleInvalidPrintfConversionSpecifier(FS, startSpecifier, 1577 specifierLen); 1578 } 1579 1580 // Check for invalid use of field width 1581 if (!FS.hasValidFieldWidth()) { 1582 HandleInvalidAmount(FS, FS.getFieldWidth(), /* field width */ 0, 1583 startSpecifier, specifierLen); 1584 } 1585 1586 // Check for invalid use of precision 1587 if (!FS.hasValidPrecision()) { 1588 HandleInvalidAmount(FS, FS.getPrecision(), /* precision */ 1, 1589 startSpecifier, specifierLen); 1590 } 1591 1592 // Check each flag does not conflict with any other component. 1593 if (!FS.hasValidThousandsGroupingPrefix()) 1594 HandleFlag(FS, FS.hasThousandsGrouping(), startSpecifier, specifierLen); 1595 if (!FS.hasValidLeadingZeros()) 1596 HandleFlag(FS, FS.hasLeadingZeros(), startSpecifier, specifierLen); 1597 if (!FS.hasValidPlusPrefix()) 1598 HandleFlag(FS, FS.hasPlusPrefix(), startSpecifier, specifierLen); 1599 if (!FS.hasValidSpacePrefix()) 1600 HandleFlag(FS, FS.hasSpacePrefix(), startSpecifier, specifierLen); 1601 if (!FS.hasValidAlternativeForm()) 1602 HandleFlag(FS, FS.hasAlternativeForm(), startSpecifier, specifierLen); 1603 if (!FS.hasValidLeftJustified()) 1604 HandleFlag(FS, FS.isLeftJustified(), startSpecifier, specifierLen); 1605 1606 // Check that flags are not ignored by another flag 1607 if (FS.hasSpacePrefix() && FS.hasPlusPrefix()) // ' ' ignored by '+' 1608 HandleIgnoredFlag(FS, FS.hasSpacePrefix(), FS.hasPlusPrefix(), 1609 startSpecifier, specifierLen); 1610 if (FS.hasLeadingZeros() && FS.isLeftJustified()) // '0' ignored by '-' 1611 HandleIgnoredFlag(FS, FS.hasLeadingZeros(), FS.isLeftJustified(), 1612 startSpecifier, specifierLen); 1613 1614 // Check the length modifier is valid with the given conversion specifier. 1615 const LengthModifier &LM = FS.getLengthModifier(); 1616 if (!FS.hasValidLengthModifier()) 1617 S.Diag(getLocationOfByte(LM.getStart()), 1618 diag::warn_format_nonsensical_length) 1619 << LM.toString() << CS.toString() 1620 << getSpecifierRange(startSpecifier, specifierLen) 1621 << FixItHint::CreateRemoval(getSpecifierRange(LM.getStart(), 1622 LM.getLength())); 1623 1624 // Are we using '%n'? 1625 if (CS.getKind() == ConversionSpecifier::nArg) { 1626 // Issue a warning about this being a possible security issue. 1627 S.Diag(getLocationOfByte(CS.getStart()), diag::warn_printf_write_back) 1628 << getSpecifierRange(startSpecifier, specifierLen); 1629 // Continue checking the other format specifiers. 1630 return true; 1631 } 1632 1633 // The remaining checks depend on the data arguments. 1634 if (HasVAListArg) 1635 return true; 1636 1637 if (!CheckNumArgs(FS, CS, startSpecifier, specifierLen, argIndex)) 1638 return false; 1639 1640 // Now type check the data expression that matches the 1641 // format specifier. 1642 const Expr *Ex = getDataArg(argIndex); 1643 const analyze_printf::ArgTypeResult &ATR = FS.getArgType(S.Context); 1644 if (ATR.isValid() && !ATR.matchesType(S.Context, Ex->getType())) { 1645 // Check if we didn't match because of an implicit cast from a 'char' 1646 // or 'short' to an 'int'. This is done because printf is a varargs 1647 // function. 1648 if (const ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(Ex)) 1649 if (ICE->getType() == S.Context.IntTy) { 1650 // All further checking is done on the subexpression. 1651 Ex = ICE->getSubExpr(); 1652 if (ATR.matchesType(S.Context, Ex->getType())) 1653 return true; 1654 } 1655 1656 // We may be able to offer a FixItHint if it is a supported type. 1657 PrintfSpecifier fixedFS = FS; 1658 bool success = fixedFS.fixType(Ex->getType()); 1659 1660 if (success) { 1661 // Get the fix string from the fixed format specifier 1662 llvm::SmallString<128> buf; 1663 llvm::raw_svector_ostream os(buf); 1664 fixedFS.toString(os); 1665 1666 // FIXME: getRepresentativeType() perhaps should return a string 1667 // instead of a QualType to better handle when the representative 1668 // type is 'wint_t' (which is defined in the system headers). 1669 S.Diag(getLocationOfByte(CS.getStart()), 1670 diag::warn_printf_conversion_argument_type_mismatch) 1671 << ATR.getRepresentativeType(S.Context) << Ex->getType() 1672 << getSpecifierRange(startSpecifier, specifierLen) 1673 << Ex->getSourceRange() 1674 << FixItHint::CreateReplacement( 1675 getSpecifierRange(startSpecifier, specifierLen), 1676 os.str()); 1677 } 1678 else { 1679 S.Diag(getLocationOfByte(CS.getStart()), 1680 diag::warn_printf_conversion_argument_type_mismatch) 1681 << ATR.getRepresentativeType(S.Context) << Ex->getType() 1682 << getSpecifierRange(startSpecifier, specifierLen) 1683 << Ex->getSourceRange(); 1684 } 1685 } 1686 1687 return true; 1688} 1689 1690//===--- CHECK: Scanf format string checking ------------------------------===// 1691 1692namespace { 1693class CheckScanfHandler : public CheckFormatHandler { 1694public: 1695 CheckScanfHandler(Sema &s, const StringLiteral *fexpr, 1696 const Expr *origFormatExpr, unsigned firstDataArg, 1697 unsigned numDataArgs, bool isObjCLiteral, 1698 const char *beg, bool hasVAListArg, 1699 const CallExpr *theCall, unsigned formatIdx) 1700 : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg, 1701 numDataArgs, isObjCLiteral, beg, hasVAListArg, 1702 theCall, formatIdx) {} 1703 1704 bool HandleScanfSpecifier(const analyze_scanf::ScanfSpecifier &FS, 1705 const char *startSpecifier, 1706 unsigned specifierLen); 1707 1708 bool HandleInvalidScanfConversionSpecifier( 1709 const analyze_scanf::ScanfSpecifier &FS, 1710 const char *startSpecifier, 1711 unsigned specifierLen); 1712 1713 void HandleIncompleteScanList(const char *start, const char *end); 1714}; 1715} 1716 1717void CheckScanfHandler::HandleIncompleteScanList(const char *start, 1718 const char *end) { 1719 S.Diag(getLocationOfByte(end), diag::warn_scanf_scanlist_incomplete) 1720 << getSpecifierRange(start, end - start); 1721} 1722 1723bool CheckScanfHandler::HandleInvalidScanfConversionSpecifier( 1724 const analyze_scanf::ScanfSpecifier &FS, 1725 const char *startSpecifier, 1726 unsigned specifierLen) { 1727 1728 const analyze_scanf::ScanfConversionSpecifier &CS = 1729 FS.getConversionSpecifier(); 1730 1731 return HandleInvalidConversionSpecifier(FS.getArgIndex(), 1732 getLocationOfByte(CS.getStart()), 1733 startSpecifier, specifierLen, 1734 CS.getStart(), CS.getLength()); 1735} 1736 1737bool CheckScanfHandler::HandleScanfSpecifier( 1738 const analyze_scanf::ScanfSpecifier &FS, 1739 const char *startSpecifier, 1740 unsigned specifierLen) { 1741 1742 using namespace analyze_scanf; 1743 using namespace analyze_format_string; 1744 1745 const ScanfConversionSpecifier &CS = FS.getConversionSpecifier(); 1746 1747 // Handle case where '%' and '*' don't consume an argument. These shouldn't 1748 // be used to decide if we are using positional arguments consistently. 1749 if (FS.consumesDataArgument()) { 1750 if (atFirstArg) { 1751 atFirstArg = false; 1752 usesPositionalArgs = FS.usesPositionalArg(); 1753 } 1754 else if (usesPositionalArgs != FS.usesPositionalArg()) { 1755 // Cannot mix-and-match positional and non-positional arguments. 1756 S.Diag(getLocationOfByte(CS.getStart()), 1757 diag::warn_format_mix_positional_nonpositional_args) 1758 << getSpecifierRange(startSpecifier, specifierLen); 1759 return false; 1760 } 1761 } 1762 1763 // Check if the field with is non-zero. 1764 const OptionalAmount &Amt = FS.getFieldWidth(); 1765 if (Amt.getHowSpecified() == OptionalAmount::Constant) { 1766 if (Amt.getConstantAmount() == 0) { 1767 const CharSourceRange &R = getSpecifierRange(Amt.getStart(), 1768 Amt.getConstantLength()); 1769 S.Diag(getLocationOfByte(Amt.getStart()), 1770 diag::warn_scanf_nonzero_width) 1771 << R << FixItHint::CreateRemoval(R); 1772 } 1773 } 1774 1775 if (!FS.consumesDataArgument()) { 1776 // FIXME: Technically specifying a precision or field width here 1777 // makes no sense. Worth issuing a warning at some point. 1778 return true; 1779 } 1780 1781 // Consume the argument. 1782 unsigned argIndex = FS.getArgIndex(); 1783 if (argIndex < NumDataArgs) { 1784 // The check to see if the argIndex is valid will come later. 1785 // We set the bit here because we may exit early from this 1786 // function if we encounter some other error. 1787 CoveredArgs.set(argIndex); 1788 } 1789 1790 // Check the length modifier is valid with the given conversion specifier. 1791 const LengthModifier &LM = FS.getLengthModifier(); 1792 if (!FS.hasValidLengthModifier()) { 1793 S.Diag(getLocationOfByte(LM.getStart()), 1794 diag::warn_format_nonsensical_length) 1795 << LM.toString() << CS.toString() 1796 << getSpecifierRange(startSpecifier, specifierLen) 1797 << FixItHint::CreateRemoval(getSpecifierRange(LM.getStart(), 1798 LM.getLength())); 1799 } 1800 1801 // The remaining checks depend on the data arguments. 1802 if (HasVAListArg) 1803 return true; 1804 1805 if (!CheckNumArgs(FS, CS, startSpecifier, specifierLen, argIndex)) 1806 return false; 1807 1808 // FIXME: Check that the argument type matches the format specifier. 1809 1810 return true; 1811} 1812 1813void Sema::CheckFormatString(const StringLiteral *FExpr, 1814 const Expr *OrigFormatExpr, 1815 const CallExpr *TheCall, bool HasVAListArg, 1816 unsigned format_idx, unsigned firstDataArg, 1817 bool isPrintf) { 1818 1819 // CHECK: is the format string a wide literal? 1820 if (!FExpr->isAscii()) { 1821 Diag(FExpr->getLocStart(), 1822 diag::warn_format_string_is_wide_literal) 1823 << OrigFormatExpr->getSourceRange(); 1824 return; 1825 } 1826 1827 // Str - The format string. NOTE: this is NOT null-terminated! 1828 StringRef StrRef = FExpr->getString(); 1829 const char *Str = StrRef.data(); 1830 unsigned StrLen = StrRef.size(); 1831 1832 // CHECK: empty format string? 1833 if (StrLen == 0) { 1834 Diag(FExpr->getLocStart(), diag::warn_empty_format_string) 1835 << OrigFormatExpr->getSourceRange(); 1836 return; 1837 } 1838 1839 if (isPrintf) { 1840 CheckPrintfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg, 1841 TheCall->getNumArgs() - firstDataArg, 1842 isa<ObjCStringLiteral>(OrigFormatExpr), Str, 1843 HasVAListArg, TheCall, format_idx); 1844 1845 if (!analyze_format_string::ParsePrintfString(H, Str, Str + StrLen)) 1846 H.DoneProcessing(); 1847 } 1848 else { 1849 CheckScanfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg, 1850 TheCall->getNumArgs() - firstDataArg, 1851 isa<ObjCStringLiteral>(OrigFormatExpr), Str, 1852 HasVAListArg, TheCall, format_idx); 1853 1854 if (!analyze_format_string::ParseScanfString(H, Str, Str + StrLen)) 1855 H.DoneProcessing(); 1856 } 1857} 1858 1859//===--- CHECK: Standard memory functions ---------------------------------===// 1860 1861/// \brief Determine whether the given type is a dynamic class type (e.g., 1862/// whether it has a vtable). 1863static bool isDynamicClassType(QualType T) { 1864 if (CXXRecordDecl *Record = T->getAsCXXRecordDecl()) 1865 if (CXXRecordDecl *Definition = Record->getDefinition()) 1866 if (Definition->isDynamicClass()) 1867 return true; 1868 1869 return false; 1870} 1871 1872/// \brief If E is a sizeof expression, returns its argument expression, 1873/// otherwise returns NULL. 1874static const Expr *getSizeOfExprArg(const Expr* E) { 1875 if (const UnaryExprOrTypeTraitExpr *SizeOf = 1876 dyn_cast<UnaryExprOrTypeTraitExpr>(E)) 1877 if (SizeOf->getKind() == clang::UETT_SizeOf && !SizeOf->isArgumentType()) 1878 return SizeOf->getArgumentExpr()->IgnoreParenImpCasts(); 1879 1880 return 0; 1881} 1882 1883/// \brief If E is a sizeof expression, returns its argument type. 1884static QualType getSizeOfArgType(const Expr* E) { 1885 if (const UnaryExprOrTypeTraitExpr *SizeOf = 1886 dyn_cast<UnaryExprOrTypeTraitExpr>(E)) 1887 if (SizeOf->getKind() == clang::UETT_SizeOf) 1888 return SizeOf->getTypeOfArgument(); 1889 1890 return QualType(); 1891} 1892 1893/// \brief Check for dangerous or invalid arguments to memset(). 1894/// 1895/// This issues warnings on known problematic, dangerous or unspecified 1896/// arguments to the standard 'memset', 'memcpy', 'memmove', and 'memcmp' 1897/// function calls. 1898/// 1899/// \param Call The call expression to diagnose. 1900void Sema::CheckMemaccessArguments(const CallExpr *Call, 1901 CheckedMemoryFunction CMF, 1902 IdentifierInfo *FnName) { 1903 // It is possible to have a non-standard definition of memset. Validate 1904 // we have enough arguments, and if not, abort further checking. 1905 if (Call->getNumArgs() < 3) 1906 return; 1907 1908 unsigned LastArg = (CMF == CMF_Memset? 1 : 2); 1909 const Expr *LenExpr = Call->getArg(2)->IgnoreParenImpCasts(); 1910 1911 // We have special checking when the length is a sizeof expression. 1912 QualType SizeOfArgTy = getSizeOfArgType(LenExpr); 1913 const Expr *SizeOfArg = getSizeOfExprArg(LenExpr); 1914 llvm::FoldingSetNodeID SizeOfArgID; 1915 1916 for (unsigned ArgIdx = 0; ArgIdx != LastArg; ++ArgIdx) { 1917 const Expr *Dest = Call->getArg(ArgIdx)->IgnoreParenImpCasts(); 1918 SourceRange ArgRange = Call->getArg(ArgIdx)->getSourceRange(); 1919 1920 QualType DestTy = Dest->getType(); 1921 if (const PointerType *DestPtrTy = DestTy->getAs<PointerType>()) { 1922 QualType PointeeTy = DestPtrTy->getPointeeType(); 1923 1924 // Never warn about void type pointers. This can be used to suppress 1925 // false positives. 1926 if (PointeeTy->isVoidType()) 1927 continue; 1928 1929 // Catch "memset(p, 0, sizeof(p))" -- needs to be sizeof(*p). Do this by 1930 // actually comparing the expressions for equality. Because computing the 1931 // expression IDs can be expensive, we only do this if the diagnostic is 1932 // enabled. 1933 if (SizeOfArg && 1934 Diags.getDiagnosticLevel(diag::warn_sizeof_pointer_expr_memaccess, 1935 SizeOfArg->getExprLoc())) { 1936 // We only compute IDs for expressions if the warning is enabled, and 1937 // cache the sizeof arg's ID. 1938 if (SizeOfArgID == llvm::FoldingSetNodeID()) 1939 SizeOfArg->Profile(SizeOfArgID, Context, true); 1940 llvm::FoldingSetNodeID DestID; 1941 Dest->Profile(DestID, Context, true); 1942 if (DestID == SizeOfArgID) { 1943 unsigned ActionIdx = 0; // Default is to suggest dereferencing. 1944 if (const UnaryOperator *UnaryOp = dyn_cast<UnaryOperator>(Dest)) 1945 if (UnaryOp->getOpcode() == UO_AddrOf) 1946 ActionIdx = 1; // If its an address-of operator, just remove it. 1947 if (Context.getTypeSize(PointeeTy) == Context.getCharWidth()) 1948 ActionIdx = 2; // If the pointee's size is sizeof(char), 1949 // suggest an explicit length. 1950 DiagRuntimeBehavior(SizeOfArg->getExprLoc(), Dest, 1951 PDiag(diag::warn_sizeof_pointer_expr_memaccess) 1952 << FnName << ArgIdx << ActionIdx 1953 << Dest->getSourceRange() 1954 << SizeOfArg->getSourceRange()); 1955 break; 1956 } 1957 } 1958 1959 // Also check for cases where the sizeof argument is the exact same 1960 // type as the memory argument, and where it points to a user-defined 1961 // record type. 1962 if (SizeOfArgTy != QualType()) { 1963 if (PointeeTy->isRecordType() && 1964 Context.typesAreCompatible(SizeOfArgTy, DestTy)) { 1965 DiagRuntimeBehavior(LenExpr->getExprLoc(), Dest, 1966 PDiag(diag::warn_sizeof_pointer_type_memaccess) 1967 << FnName << SizeOfArgTy << ArgIdx 1968 << PointeeTy << Dest->getSourceRange() 1969 << LenExpr->getSourceRange()); 1970 break; 1971 } 1972 } 1973 1974 unsigned DiagID; 1975 1976 // Always complain about dynamic classes. 1977 if (isDynamicClassType(PointeeTy)) 1978 DiagID = diag::warn_dyn_class_memaccess; 1979 else if (PointeeTy.hasNonTrivialObjCLifetime() && CMF != CMF_Memset) 1980 DiagID = diag::warn_arc_object_memaccess; 1981 else 1982 continue; 1983 1984 DiagRuntimeBehavior( 1985 Dest->getExprLoc(), Dest, 1986 PDiag(DiagID) 1987 << ArgIdx << FnName << PointeeTy 1988 << Call->getCallee()->getSourceRange()); 1989 1990 DiagRuntimeBehavior( 1991 Dest->getExprLoc(), Dest, 1992 PDiag(diag::note_bad_memaccess_silence) 1993 << FixItHint::CreateInsertion(ArgRange.getBegin(), "(void*)")); 1994 break; 1995 } 1996 } 1997} 1998 1999// A little helper routine: ignore addition and subtraction of integer literals. 2000// This intentionally does not ignore all integer constant expressions because 2001// we don't want to remove sizeof(). 2002static const Expr *ignoreLiteralAdditions(const Expr *Ex, ASTContext &Ctx) { 2003 Ex = Ex->IgnoreParenCasts(); 2004 2005 for (;;) { 2006 const BinaryOperator * BO = dyn_cast<BinaryOperator>(Ex); 2007 if (!BO || !BO->isAdditiveOp()) 2008 break; 2009 2010 const Expr *RHS = BO->getRHS()->IgnoreParenCasts(); 2011 const Expr *LHS = BO->getLHS()->IgnoreParenCasts(); 2012 2013 if (isa<IntegerLiteral>(RHS)) 2014 Ex = LHS; 2015 else if (isa<IntegerLiteral>(LHS)) 2016 Ex = RHS; 2017 else 2018 break; 2019 } 2020 2021 return Ex; 2022} 2023 2024// Warn if the user has made the 'size' argument to strlcpy or strlcat 2025// be the size of the source, instead of the destination. 2026void Sema::CheckStrlcpycatArguments(const CallExpr *Call, 2027 IdentifierInfo *FnName) { 2028 2029 // Don't crash if the user has the wrong number of arguments 2030 if (Call->getNumArgs() != 3) 2031 return; 2032 2033 const Expr *SrcArg = ignoreLiteralAdditions(Call->getArg(1), Context); 2034 const Expr *SizeArg = ignoreLiteralAdditions(Call->getArg(2), Context); 2035 const Expr *CompareWithSrc = NULL; 2036 2037 // Look for 'strlcpy(dst, x, sizeof(x))' 2038 if (const Expr *Ex = getSizeOfExprArg(SizeArg)) 2039 CompareWithSrc = Ex; 2040 else { 2041 // Look for 'strlcpy(dst, x, strlen(x))' 2042 if (const CallExpr *SizeCall = dyn_cast<CallExpr>(SizeArg)) { 2043 if (SizeCall->isBuiltinCall(Context) == Builtin::BIstrlen 2044 && SizeCall->getNumArgs() == 1) 2045 CompareWithSrc = ignoreLiteralAdditions(SizeCall->getArg(0), Context); 2046 } 2047 } 2048 2049 if (!CompareWithSrc) 2050 return; 2051 2052 // Determine if the argument to sizeof/strlen is equal to the source 2053 // argument. In principle there's all kinds of things you could do 2054 // here, for instance creating an == expression and evaluating it with 2055 // EvaluateAsBooleanCondition, but this uses a more direct technique: 2056 const DeclRefExpr *SrcArgDRE = dyn_cast<DeclRefExpr>(SrcArg); 2057 if (!SrcArgDRE) 2058 return; 2059 2060 const DeclRefExpr *CompareWithSrcDRE = dyn_cast<DeclRefExpr>(CompareWithSrc); 2061 if (!CompareWithSrcDRE || 2062 SrcArgDRE->getDecl() != CompareWithSrcDRE->getDecl()) 2063 return; 2064 2065 const Expr *OriginalSizeArg = Call->getArg(2); 2066 Diag(CompareWithSrcDRE->getLocStart(), diag::warn_strlcpycat_wrong_size) 2067 << OriginalSizeArg->getSourceRange() << FnName; 2068 2069 // Output a FIXIT hint if the destination is an array (rather than a 2070 // pointer to an array). This could be enhanced to handle some 2071 // pointers if we know the actual size, like if DstArg is 'array+2' 2072 // we could say 'sizeof(array)-2'. 2073 const Expr *DstArg = Call->getArg(0)->IgnoreParenImpCasts(); 2074 2075 if (DstArg->getType()->isArrayType()) { 2076 llvm::SmallString<128> sizeString; 2077 llvm::raw_svector_ostream OS(sizeString); 2078 OS << "sizeof("; 2079 DstArg->printPretty(OS, Context, 0, Context.PrintingPolicy); 2080 OS << ")"; 2081 2082 Diag(OriginalSizeArg->getLocStart(), diag::note_strlcpycat_wrong_size) 2083 << FixItHint::CreateReplacement(OriginalSizeArg->getSourceRange(), 2084 OS.str()); 2085 } 2086} 2087 2088//===--- CHECK: Return Address of Stack Variable --------------------------===// 2089 2090static Expr *EvalVal(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars); 2091static Expr *EvalAddr(Expr* E, SmallVectorImpl<DeclRefExpr *> &refVars); 2092 2093/// CheckReturnStackAddr - Check if a return statement returns the address 2094/// of a stack variable. 2095void 2096Sema::CheckReturnStackAddr(Expr *RetValExp, QualType lhsType, 2097 SourceLocation ReturnLoc) { 2098 2099 Expr *stackE = 0; 2100 SmallVector<DeclRefExpr *, 8> refVars; 2101 2102 // Perform checking for returned stack addresses, local blocks, 2103 // label addresses or references to temporaries. 2104 if (lhsType->isPointerType() || 2105 (!getLangOptions().ObjCAutoRefCount && lhsType->isBlockPointerType())) { 2106 stackE = EvalAddr(RetValExp, refVars); 2107 } else if (lhsType->isReferenceType()) { 2108 stackE = EvalVal(RetValExp, refVars); 2109 } 2110 2111 if (stackE == 0) 2112 return; // Nothing suspicious was found. 2113 2114 SourceLocation diagLoc; 2115 SourceRange diagRange; 2116 if (refVars.empty()) { 2117 diagLoc = stackE->getLocStart(); 2118 diagRange = stackE->getSourceRange(); 2119 } else { 2120 // We followed through a reference variable. 'stackE' contains the 2121 // problematic expression but we will warn at the return statement pointing 2122 // at the reference variable. We will later display the "trail" of 2123 // reference variables using notes. 2124 diagLoc = refVars[0]->getLocStart(); 2125 diagRange = refVars[0]->getSourceRange(); 2126 } 2127 2128 if (DeclRefExpr *DR = dyn_cast<DeclRefExpr>(stackE)) { //address of local var. 2129 Diag(diagLoc, lhsType->isReferenceType() ? diag::warn_ret_stack_ref 2130 : diag::warn_ret_stack_addr) 2131 << DR->getDecl()->getDeclName() << diagRange; 2132 } else if (isa<BlockExpr>(stackE)) { // local block. 2133 Diag(diagLoc, diag::err_ret_local_block) << diagRange; 2134 } else if (isa<AddrLabelExpr>(stackE)) { // address of label. 2135 Diag(diagLoc, diag::warn_ret_addr_label) << diagRange; 2136 } else { // local temporary. 2137 Diag(diagLoc, lhsType->isReferenceType() ? diag::warn_ret_local_temp_ref 2138 : diag::warn_ret_local_temp_addr) 2139 << diagRange; 2140 } 2141 2142 // Display the "trail" of reference variables that we followed until we 2143 // found the problematic expression using notes. 2144 for (unsigned i = 0, e = refVars.size(); i != e; ++i) { 2145 VarDecl *VD = cast<VarDecl>(refVars[i]->getDecl()); 2146 // If this var binds to another reference var, show the range of the next 2147 // var, otherwise the var binds to the problematic expression, in which case 2148 // show the range of the expression. 2149 SourceRange range = (i < e-1) ? refVars[i+1]->getSourceRange() 2150 : stackE->getSourceRange(); 2151 Diag(VD->getLocation(), diag::note_ref_var_local_bind) 2152 << VD->getDeclName() << range; 2153 } 2154} 2155 2156/// EvalAddr - EvalAddr and EvalVal are mutually recursive functions that 2157/// check if the expression in a return statement evaluates to an address 2158/// to a location on the stack, a local block, an address of a label, or a 2159/// reference to local temporary. The recursion is used to traverse the 2160/// AST of the return expression, with recursion backtracking when we 2161/// encounter a subexpression that (1) clearly does not lead to one of the 2162/// above problematic expressions (2) is something we cannot determine leads to 2163/// a problematic expression based on such local checking. 2164/// 2165/// Both EvalAddr and EvalVal follow through reference variables to evaluate 2166/// the expression that they point to. Such variables are added to the 2167/// 'refVars' vector so that we know what the reference variable "trail" was. 2168/// 2169/// EvalAddr processes expressions that are pointers that are used as 2170/// references (and not L-values). EvalVal handles all other values. 2171/// At the base case of the recursion is a check for the above problematic 2172/// expressions. 2173/// 2174/// This implementation handles: 2175/// 2176/// * pointer-to-pointer casts 2177/// * implicit conversions from array references to pointers 2178/// * taking the address of fields 2179/// * arbitrary interplay between "&" and "*" operators 2180/// * pointer arithmetic from an address of a stack variable 2181/// * taking the address of an array element where the array is on the stack 2182static Expr *EvalAddr(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars) { 2183 if (E->isTypeDependent()) 2184 return NULL; 2185 2186 // We should only be called for evaluating pointer expressions. 2187 assert((E->getType()->isAnyPointerType() || 2188 E->getType()->isBlockPointerType() || 2189 E->getType()->isObjCQualifiedIdType()) && 2190 "EvalAddr only works on pointers"); 2191 2192 E = E->IgnoreParens(); 2193 2194 // Our "symbolic interpreter" is just a dispatch off the currently 2195 // viewed AST node. We then recursively traverse the AST by calling 2196 // EvalAddr and EvalVal appropriately. 2197 switch (E->getStmtClass()) { 2198 case Stmt::DeclRefExprClass: { 2199 DeclRefExpr *DR = cast<DeclRefExpr>(E); 2200 2201 if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl())) 2202 // If this is a reference variable, follow through to the expression that 2203 // it points to. 2204 if (V->hasLocalStorage() && 2205 V->getType()->isReferenceType() && V->hasInit()) { 2206 // Add the reference variable to the "trail". 2207 refVars.push_back(DR); 2208 return EvalAddr(V->getInit(), refVars); 2209 } 2210 2211 return NULL; 2212 } 2213 2214 case Stmt::UnaryOperatorClass: { 2215 // The only unary operator that make sense to handle here 2216 // is AddrOf. All others don't make sense as pointers. 2217 UnaryOperator *U = cast<UnaryOperator>(E); 2218 2219 if (U->getOpcode() == UO_AddrOf) 2220 return EvalVal(U->getSubExpr(), refVars); 2221 else 2222 return NULL; 2223 } 2224 2225 case Stmt::BinaryOperatorClass: { 2226 // Handle pointer arithmetic. All other binary operators are not valid 2227 // in this context. 2228 BinaryOperator *B = cast<BinaryOperator>(E); 2229 BinaryOperatorKind op = B->getOpcode(); 2230 2231 if (op != BO_Add && op != BO_Sub) 2232 return NULL; 2233 2234 Expr *Base = B->getLHS(); 2235 2236 // Determine which argument is the real pointer base. It could be 2237 // the RHS argument instead of the LHS. 2238 if (!Base->getType()->isPointerType()) Base = B->getRHS(); 2239 2240 assert (Base->getType()->isPointerType()); 2241 return EvalAddr(Base, refVars); 2242 } 2243 2244 // For conditional operators we need to see if either the LHS or RHS are 2245 // valid DeclRefExpr*s. If one of them is valid, we return it. 2246 case Stmt::ConditionalOperatorClass: { 2247 ConditionalOperator *C = cast<ConditionalOperator>(E); 2248 2249 // Handle the GNU extension for missing LHS. 2250 if (Expr *lhsExpr = C->getLHS()) { 2251 // In C++, we can have a throw-expression, which has 'void' type. 2252 if (!lhsExpr->getType()->isVoidType()) 2253 if (Expr* LHS = EvalAddr(lhsExpr, refVars)) 2254 return LHS; 2255 } 2256 2257 // In C++, we can have a throw-expression, which has 'void' type. 2258 if (C->getRHS()->getType()->isVoidType()) 2259 return NULL; 2260 2261 return EvalAddr(C->getRHS(), refVars); 2262 } 2263 2264 case Stmt::BlockExprClass: 2265 if (cast<BlockExpr>(E)->getBlockDecl()->hasCaptures()) 2266 return E; // local block. 2267 return NULL; 2268 2269 case Stmt::AddrLabelExprClass: 2270 return E; // address of label. 2271 2272 // For casts, we need to handle conversions from arrays to 2273 // pointer values, and pointer-to-pointer conversions. 2274 case Stmt::ImplicitCastExprClass: 2275 case Stmt::CStyleCastExprClass: 2276 case Stmt::CXXFunctionalCastExprClass: 2277 case Stmt::ObjCBridgedCastExprClass: { 2278 Expr* SubExpr = cast<CastExpr>(E)->getSubExpr(); 2279 QualType T = SubExpr->getType(); 2280 2281 if (SubExpr->getType()->isPointerType() || 2282 SubExpr->getType()->isBlockPointerType() || 2283 SubExpr->getType()->isObjCQualifiedIdType()) 2284 return EvalAddr(SubExpr, refVars); 2285 else if (T->isArrayType()) 2286 return EvalVal(SubExpr, refVars); 2287 else 2288 return 0; 2289 } 2290 2291 // C++ casts. For dynamic casts, static casts, and const casts, we 2292 // are always converting from a pointer-to-pointer, so we just blow 2293 // through the cast. In the case the dynamic cast doesn't fail (and 2294 // return NULL), we take the conservative route and report cases 2295 // where we return the address of a stack variable. For Reinterpre 2296 // FIXME: The comment about is wrong; we're not always converting 2297 // from pointer to pointer. I'm guessing that this code should also 2298 // handle references to objects. 2299 case Stmt::CXXStaticCastExprClass: 2300 case Stmt::CXXDynamicCastExprClass: 2301 case Stmt::CXXConstCastExprClass: 2302 case Stmt::CXXReinterpretCastExprClass: { 2303 Expr *S = cast<CXXNamedCastExpr>(E)->getSubExpr(); 2304 if (S->getType()->isPointerType() || S->getType()->isBlockPointerType()) 2305 return EvalAddr(S, refVars); 2306 else 2307 return NULL; 2308 } 2309 2310 case Stmt::MaterializeTemporaryExprClass: 2311 if (Expr *Result = EvalAddr( 2312 cast<MaterializeTemporaryExpr>(E)->GetTemporaryExpr(), 2313 refVars)) 2314 return Result; 2315 2316 return E; 2317 2318 // Everything else: we simply don't reason about them. 2319 default: 2320 return NULL; 2321 } 2322} 2323 2324 2325/// EvalVal - This function is complements EvalAddr in the mutual recursion. 2326/// See the comments for EvalAddr for more details. 2327static Expr *EvalVal(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars) { 2328do { 2329 // We should only be called for evaluating non-pointer expressions, or 2330 // expressions with a pointer type that are not used as references but instead 2331 // are l-values (e.g., DeclRefExpr with a pointer type). 2332 2333 // Our "symbolic interpreter" is just a dispatch off the currently 2334 // viewed AST node. We then recursively traverse the AST by calling 2335 // EvalAddr and EvalVal appropriately. 2336 2337 E = E->IgnoreParens(); 2338 switch (E->getStmtClass()) { 2339 case Stmt::ImplicitCastExprClass: { 2340 ImplicitCastExpr *IE = cast<ImplicitCastExpr>(E); 2341 if (IE->getValueKind() == VK_LValue) { 2342 E = IE->getSubExpr(); 2343 continue; 2344 } 2345 return NULL; 2346 } 2347 2348 case Stmt::DeclRefExprClass: { 2349 // When we hit a DeclRefExpr we are looking at code that refers to a 2350 // variable's name. If it's not a reference variable we check if it has 2351 // local storage within the function, and if so, return the expression. 2352 DeclRefExpr *DR = cast<DeclRefExpr>(E); 2353 2354 if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl())) 2355 if (V->hasLocalStorage()) { 2356 if (!V->getType()->isReferenceType()) 2357 return DR; 2358 2359 // Reference variable, follow through to the expression that 2360 // it points to. 2361 if (V->hasInit()) { 2362 // Add the reference variable to the "trail". 2363 refVars.push_back(DR); 2364 return EvalVal(V->getInit(), refVars); 2365 } 2366 } 2367 2368 return NULL; 2369 } 2370 2371 case Stmt::UnaryOperatorClass: { 2372 // The only unary operator that make sense to handle here 2373 // is Deref. All others don't resolve to a "name." This includes 2374 // handling all sorts of rvalues passed to a unary operator. 2375 UnaryOperator *U = cast<UnaryOperator>(E); 2376 2377 if (U->getOpcode() == UO_Deref) 2378 return EvalAddr(U->getSubExpr(), refVars); 2379 2380 return NULL; 2381 } 2382 2383 case Stmt::ArraySubscriptExprClass: { 2384 // Array subscripts are potential references to data on the stack. We 2385 // retrieve the DeclRefExpr* for the array variable if it indeed 2386 // has local storage. 2387 return EvalAddr(cast<ArraySubscriptExpr>(E)->getBase(), refVars); 2388 } 2389 2390 case Stmt::ConditionalOperatorClass: { 2391 // For conditional operators we need to see if either the LHS or RHS are 2392 // non-NULL Expr's. If one is non-NULL, we return it. 2393 ConditionalOperator *C = cast<ConditionalOperator>(E); 2394 2395 // Handle the GNU extension for missing LHS. 2396 if (Expr *lhsExpr = C->getLHS()) 2397 if (Expr *LHS = EvalVal(lhsExpr, refVars)) 2398 return LHS; 2399 2400 return EvalVal(C->getRHS(), refVars); 2401 } 2402 2403 // Accesses to members are potential references to data on the stack. 2404 case Stmt::MemberExprClass: { 2405 MemberExpr *M = cast<MemberExpr>(E); 2406 2407 // Check for indirect access. We only want direct field accesses. 2408 if (M->isArrow()) 2409 return NULL; 2410 2411 // Check whether the member type is itself a reference, in which case 2412 // we're not going to refer to the member, but to what the member refers to. 2413 if (M->getMemberDecl()->getType()->isReferenceType()) 2414 return NULL; 2415 2416 return EvalVal(M->getBase(), refVars); 2417 } 2418 2419 case Stmt::MaterializeTemporaryExprClass: 2420 if (Expr *Result = EvalVal( 2421 cast<MaterializeTemporaryExpr>(E)->GetTemporaryExpr(), 2422 refVars)) 2423 return Result; 2424 2425 return E; 2426 2427 default: 2428 // Check that we don't return or take the address of a reference to a 2429 // temporary. This is only useful in C++. 2430 if (!E->isTypeDependent() && E->isRValue()) 2431 return E; 2432 2433 // Everything else: we simply don't reason about them. 2434 return NULL; 2435 } 2436} while (true); 2437} 2438 2439//===--- CHECK: Floating-Point comparisons (-Wfloat-equal) ---------------===// 2440 2441/// Check for comparisons of floating point operands using != and ==. 2442/// Issue a warning if these are no self-comparisons, as they are not likely 2443/// to do what the programmer intended. 2444void Sema::CheckFloatComparison(SourceLocation loc, Expr* lex, Expr *rex) { 2445 bool EmitWarning = true; 2446 2447 Expr* LeftExprSansParen = lex->IgnoreParenImpCasts(); 2448 Expr* RightExprSansParen = rex->IgnoreParenImpCasts(); 2449 2450 // Special case: check for x == x (which is OK). 2451 // Do not emit warnings for such cases. 2452 if (DeclRefExpr* DRL = dyn_cast<DeclRefExpr>(LeftExprSansParen)) 2453 if (DeclRefExpr* DRR = dyn_cast<DeclRefExpr>(RightExprSansParen)) 2454 if (DRL->getDecl() == DRR->getDecl()) 2455 EmitWarning = false; 2456 2457 2458 // Special case: check for comparisons against literals that can be exactly 2459 // represented by APFloat. In such cases, do not emit a warning. This 2460 // is a heuristic: often comparison against such literals are used to 2461 // detect if a value in a variable has not changed. This clearly can 2462 // lead to false negatives. 2463 if (EmitWarning) { 2464 if (FloatingLiteral* FLL = dyn_cast<FloatingLiteral>(LeftExprSansParen)) { 2465 if (FLL->isExact()) 2466 EmitWarning = false; 2467 } else 2468 if (FloatingLiteral* FLR = dyn_cast<FloatingLiteral>(RightExprSansParen)){ 2469 if (FLR->isExact()) 2470 EmitWarning = false; 2471 } 2472 } 2473 2474 // Check for comparisons with builtin types. 2475 if (EmitWarning) 2476 if (CallExpr* CL = dyn_cast<CallExpr>(LeftExprSansParen)) 2477 if (CL->isBuiltinCall(Context)) 2478 EmitWarning = false; 2479 2480 if (EmitWarning) 2481 if (CallExpr* CR = dyn_cast<CallExpr>(RightExprSansParen)) 2482 if (CR->isBuiltinCall(Context)) 2483 EmitWarning = false; 2484 2485 // Emit the diagnostic. 2486 if (EmitWarning) 2487 Diag(loc, diag::warn_floatingpoint_eq) 2488 << lex->getSourceRange() << rex->getSourceRange(); 2489} 2490 2491//===--- CHECK: Integer mixed-sign comparisons (-Wsign-compare) --------===// 2492//===--- CHECK: Lossy implicit conversions (-Wconversion) --------------===// 2493 2494namespace { 2495 2496/// Structure recording the 'active' range of an integer-valued 2497/// expression. 2498struct IntRange { 2499 /// The number of bits active in the int. 2500 unsigned Width; 2501 2502 /// True if the int is known not to have negative values. 2503 bool NonNegative; 2504 2505 IntRange(unsigned Width, bool NonNegative) 2506 : Width(Width), NonNegative(NonNegative) 2507 {} 2508 2509 /// Returns the range of the bool type. 2510 static IntRange forBoolType() { 2511 return IntRange(1, true); 2512 } 2513 2514 /// Returns the range of an opaque value of the given integral type. 2515 static IntRange forValueOfType(ASTContext &C, QualType T) { 2516 return forValueOfCanonicalType(C, 2517 T->getCanonicalTypeInternal().getTypePtr()); 2518 } 2519 2520 /// Returns the range of an opaque value of a canonical integral type. 2521 static IntRange forValueOfCanonicalType(ASTContext &C, const Type *T) { 2522 assert(T->isCanonicalUnqualified()); 2523 2524 if (const VectorType *VT = dyn_cast<VectorType>(T)) 2525 T = VT->getElementType().getTypePtr(); 2526 if (const ComplexType *CT = dyn_cast<ComplexType>(T)) 2527 T = CT->getElementType().getTypePtr(); 2528 2529 // For enum types, use the known bit width of the enumerators. 2530 if (const EnumType *ET = dyn_cast<EnumType>(T)) { 2531 EnumDecl *Enum = ET->getDecl(); 2532 if (!Enum->isDefinition()) 2533 return IntRange(C.getIntWidth(QualType(T, 0)), false); 2534 2535 unsigned NumPositive = Enum->getNumPositiveBits(); 2536 unsigned NumNegative = Enum->getNumNegativeBits(); 2537 2538 return IntRange(std::max(NumPositive, NumNegative), NumNegative == 0); 2539 } 2540 2541 const BuiltinType *BT = cast<BuiltinType>(T); 2542 assert(BT->isInteger()); 2543 2544 return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger()); 2545 } 2546 2547 /// Returns the "target" range of a canonical integral type, i.e. 2548 /// the range of values expressible in the type. 2549 /// 2550 /// This matches forValueOfCanonicalType except that enums have the 2551 /// full range of their type, not the range of their enumerators. 2552 static IntRange forTargetOfCanonicalType(ASTContext &C, const Type *T) { 2553 assert(T->isCanonicalUnqualified()); 2554 2555 if (const VectorType *VT = dyn_cast<VectorType>(T)) 2556 T = VT->getElementType().getTypePtr(); 2557 if (const ComplexType *CT = dyn_cast<ComplexType>(T)) 2558 T = CT->getElementType().getTypePtr(); 2559 if (const EnumType *ET = dyn_cast<EnumType>(T)) 2560 T = ET->getDecl()->getIntegerType().getTypePtr(); 2561 2562 const BuiltinType *BT = cast<BuiltinType>(T); 2563 assert(BT->isInteger()); 2564 2565 return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger()); 2566 } 2567 2568 /// Returns the supremum of two ranges: i.e. their conservative merge. 2569 static IntRange join(IntRange L, IntRange R) { 2570 return IntRange(std::max(L.Width, R.Width), 2571 L.NonNegative && R.NonNegative); 2572 } 2573 2574 /// Returns the infinum of two ranges: i.e. their aggressive merge. 2575 static IntRange meet(IntRange L, IntRange R) { 2576 return IntRange(std::min(L.Width, R.Width), 2577 L.NonNegative || R.NonNegative); 2578 } 2579}; 2580 2581IntRange GetValueRange(ASTContext &C, llvm::APSInt &value, unsigned MaxWidth) { 2582 if (value.isSigned() && value.isNegative()) 2583 return IntRange(value.getMinSignedBits(), false); 2584 2585 if (value.getBitWidth() > MaxWidth) 2586 value = value.trunc(MaxWidth); 2587 2588 // isNonNegative() just checks the sign bit without considering 2589 // signedness. 2590 return IntRange(value.getActiveBits(), true); 2591} 2592 2593IntRange GetValueRange(ASTContext &C, APValue &result, QualType Ty, 2594 unsigned MaxWidth) { 2595 if (result.isInt()) 2596 return GetValueRange(C, result.getInt(), MaxWidth); 2597 2598 if (result.isVector()) { 2599 IntRange R = GetValueRange(C, result.getVectorElt(0), Ty, MaxWidth); 2600 for (unsigned i = 1, e = result.getVectorLength(); i != e; ++i) { 2601 IntRange El = GetValueRange(C, result.getVectorElt(i), Ty, MaxWidth); 2602 R = IntRange::join(R, El); 2603 } 2604 return R; 2605 } 2606 2607 if (result.isComplexInt()) { 2608 IntRange R = GetValueRange(C, result.getComplexIntReal(), MaxWidth); 2609 IntRange I = GetValueRange(C, result.getComplexIntImag(), MaxWidth); 2610 return IntRange::join(R, I); 2611 } 2612 2613 // This can happen with lossless casts to intptr_t of "based" lvalues. 2614 // Assume it might use arbitrary bits. 2615 // FIXME: The only reason we need to pass the type in here is to get 2616 // the sign right on this one case. It would be nice if APValue 2617 // preserved this. 2618 assert(result.isLValue()); 2619 return IntRange(MaxWidth, Ty->isUnsignedIntegerOrEnumerationType()); 2620} 2621 2622/// Pseudo-evaluate the given integer expression, estimating the 2623/// range of values it might take. 2624/// 2625/// \param MaxWidth - the width to which the value will be truncated 2626IntRange GetExprRange(ASTContext &C, Expr *E, unsigned MaxWidth) { 2627 E = E->IgnoreParens(); 2628 2629 // Try a full evaluation first. 2630 Expr::EvalResult result; 2631 if (E->Evaluate(result, C)) 2632 return GetValueRange(C, result.Val, E->getType(), MaxWidth); 2633 2634 // I think we only want to look through implicit casts here; if the 2635 // user has an explicit widening cast, we should treat the value as 2636 // being of the new, wider type. 2637 if (ImplicitCastExpr *CE = dyn_cast<ImplicitCastExpr>(E)) { 2638 if (CE->getCastKind() == CK_NoOp) 2639 return GetExprRange(C, CE->getSubExpr(), MaxWidth); 2640 2641 IntRange OutputTypeRange = IntRange::forValueOfType(C, CE->getType()); 2642 2643 bool isIntegerCast = (CE->getCastKind() == CK_IntegralCast); 2644 2645 // Assume that non-integer casts can span the full range of the type. 2646 if (!isIntegerCast) 2647 return OutputTypeRange; 2648 2649 IntRange SubRange 2650 = GetExprRange(C, CE->getSubExpr(), 2651 std::min(MaxWidth, OutputTypeRange.Width)); 2652 2653 // Bail out if the subexpr's range is as wide as the cast type. 2654 if (SubRange.Width >= OutputTypeRange.Width) 2655 return OutputTypeRange; 2656 2657 // Otherwise, we take the smaller width, and we're non-negative if 2658 // either the output type or the subexpr is. 2659 return IntRange(SubRange.Width, 2660 SubRange.NonNegative || OutputTypeRange.NonNegative); 2661 } 2662 2663 if (ConditionalOperator *CO = dyn_cast<ConditionalOperator>(E)) { 2664 // If we can fold the condition, just take that operand. 2665 bool CondResult; 2666 if (CO->getCond()->EvaluateAsBooleanCondition(CondResult, C)) 2667 return GetExprRange(C, CondResult ? CO->getTrueExpr() 2668 : CO->getFalseExpr(), 2669 MaxWidth); 2670 2671 // Otherwise, conservatively merge. 2672 IntRange L = GetExprRange(C, CO->getTrueExpr(), MaxWidth); 2673 IntRange R = GetExprRange(C, CO->getFalseExpr(), MaxWidth); 2674 return IntRange::join(L, R); 2675 } 2676 2677 if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) { 2678 switch (BO->getOpcode()) { 2679 2680 // Boolean-valued operations are single-bit and positive. 2681 case BO_LAnd: 2682 case BO_LOr: 2683 case BO_LT: 2684 case BO_GT: 2685 case BO_LE: 2686 case BO_GE: 2687 case BO_EQ: 2688 case BO_NE: 2689 return IntRange::forBoolType(); 2690 2691 // The type of the assignments is the type of the LHS, so the RHS 2692 // is not necessarily the same type. 2693 case BO_MulAssign: 2694 case BO_DivAssign: 2695 case BO_RemAssign: 2696 case BO_AddAssign: 2697 case BO_SubAssign: 2698 case BO_XorAssign: 2699 case BO_OrAssign: 2700 // TODO: bitfields? 2701 return IntRange::forValueOfType(C, E->getType()); 2702 2703 // Simple assignments just pass through the RHS, which will have 2704 // been coerced to the LHS type. 2705 case BO_Assign: 2706 // TODO: bitfields? 2707 return GetExprRange(C, BO->getRHS(), MaxWidth); 2708 2709 // Operations with opaque sources are black-listed. 2710 case BO_PtrMemD: 2711 case BO_PtrMemI: 2712 return IntRange::forValueOfType(C, E->getType()); 2713 2714 // Bitwise-and uses the *infinum* of the two source ranges. 2715 case BO_And: 2716 case BO_AndAssign: 2717 return IntRange::meet(GetExprRange(C, BO->getLHS(), MaxWidth), 2718 GetExprRange(C, BO->getRHS(), MaxWidth)); 2719 2720 // Left shift gets black-listed based on a judgement call. 2721 case BO_Shl: 2722 // ...except that we want to treat '1 << (blah)' as logically 2723 // positive. It's an important idiom. 2724 if (IntegerLiteral *I 2725 = dyn_cast<IntegerLiteral>(BO->getLHS()->IgnoreParenCasts())) { 2726 if (I->getValue() == 1) { 2727 IntRange R = IntRange::forValueOfType(C, E->getType()); 2728 return IntRange(R.Width, /*NonNegative*/ true); 2729 } 2730 } 2731 // fallthrough 2732 2733 case BO_ShlAssign: 2734 return IntRange::forValueOfType(C, E->getType()); 2735 2736 // Right shift by a constant can narrow its left argument. 2737 case BO_Shr: 2738 case BO_ShrAssign: { 2739 IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth); 2740 2741 // If the shift amount is a positive constant, drop the width by 2742 // that much. 2743 llvm::APSInt shift; 2744 if (BO->getRHS()->isIntegerConstantExpr(shift, C) && 2745 shift.isNonNegative()) { 2746 unsigned zext = shift.getZExtValue(); 2747 if (zext >= L.Width) 2748 L.Width = (L.NonNegative ? 0 : 1); 2749 else 2750 L.Width -= zext; 2751 } 2752 2753 return L; 2754 } 2755 2756 // Comma acts as its right operand. 2757 case BO_Comma: 2758 return GetExprRange(C, BO->getRHS(), MaxWidth); 2759 2760 // Black-list pointer subtractions. 2761 case BO_Sub: 2762 if (BO->getLHS()->getType()->isPointerType()) 2763 return IntRange::forValueOfType(C, E->getType()); 2764 break; 2765 2766 // The width of a division result is mostly determined by the size 2767 // of the LHS. 2768 case BO_Div: { 2769 // Don't 'pre-truncate' the operands. 2770 unsigned opWidth = C.getIntWidth(E->getType()); 2771 IntRange L = GetExprRange(C, BO->getLHS(), opWidth); 2772 2773 // If the divisor is constant, use that. 2774 llvm::APSInt divisor; 2775 if (BO->getRHS()->isIntegerConstantExpr(divisor, C)) { 2776 unsigned log2 = divisor.logBase2(); // floor(log_2(divisor)) 2777 if (log2 >= L.Width) 2778 L.Width = (L.NonNegative ? 0 : 1); 2779 else 2780 L.Width = std::min(L.Width - log2, MaxWidth); 2781 return L; 2782 } 2783 2784 // Otherwise, just use the LHS's width. 2785 IntRange R = GetExprRange(C, BO->getRHS(), opWidth); 2786 return IntRange(L.Width, L.NonNegative && R.NonNegative); 2787 } 2788 2789 // The result of a remainder can't be larger than the result of 2790 // either side. 2791 case BO_Rem: { 2792 // Don't 'pre-truncate' the operands. 2793 unsigned opWidth = C.getIntWidth(E->getType()); 2794 IntRange L = GetExprRange(C, BO->getLHS(), opWidth); 2795 IntRange R = GetExprRange(C, BO->getRHS(), opWidth); 2796 2797 IntRange meet = IntRange::meet(L, R); 2798 meet.Width = std::min(meet.Width, MaxWidth); 2799 return meet; 2800 } 2801 2802 // The default behavior is okay for these. 2803 case BO_Mul: 2804 case BO_Add: 2805 case BO_Xor: 2806 case BO_Or: 2807 break; 2808 } 2809 2810 // The default case is to treat the operation as if it were closed 2811 // on the narrowest type that encompasses both operands. 2812 IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth); 2813 IntRange R = GetExprRange(C, BO->getRHS(), MaxWidth); 2814 return IntRange::join(L, R); 2815 } 2816 2817 if (UnaryOperator *UO = dyn_cast<UnaryOperator>(E)) { 2818 switch (UO->getOpcode()) { 2819 // Boolean-valued operations are white-listed. 2820 case UO_LNot: 2821 return IntRange::forBoolType(); 2822 2823 // Operations with opaque sources are black-listed. 2824 case UO_Deref: 2825 case UO_AddrOf: // should be impossible 2826 return IntRange::forValueOfType(C, E->getType()); 2827 2828 default: 2829 return GetExprRange(C, UO->getSubExpr(), MaxWidth); 2830 } 2831 } 2832 2833 if (dyn_cast<OffsetOfExpr>(E)) { 2834 IntRange::forValueOfType(C, E->getType()); 2835 } 2836 2837 FieldDecl *BitField = E->getBitField(); 2838 if (BitField) { 2839 llvm::APSInt BitWidthAP = BitField->getBitWidth()->EvaluateAsInt(C); 2840 unsigned BitWidth = BitWidthAP.getZExtValue(); 2841 2842 return IntRange(BitWidth, 2843 BitField->getType()->isUnsignedIntegerOrEnumerationType()); 2844 } 2845 2846 return IntRange::forValueOfType(C, E->getType()); 2847} 2848 2849IntRange GetExprRange(ASTContext &C, Expr *E) { 2850 return GetExprRange(C, E, C.getIntWidth(E->getType())); 2851} 2852 2853/// Checks whether the given value, which currently has the given 2854/// source semantics, has the same value when coerced through the 2855/// target semantics. 2856bool IsSameFloatAfterCast(const llvm::APFloat &value, 2857 const llvm::fltSemantics &Src, 2858 const llvm::fltSemantics &Tgt) { 2859 llvm::APFloat truncated = value; 2860 2861 bool ignored; 2862 truncated.convert(Src, llvm::APFloat::rmNearestTiesToEven, &ignored); 2863 truncated.convert(Tgt, llvm::APFloat::rmNearestTiesToEven, &ignored); 2864 2865 return truncated.bitwiseIsEqual(value); 2866} 2867 2868/// Checks whether the given value, which currently has the given 2869/// source semantics, has the same value when coerced through the 2870/// target semantics. 2871/// 2872/// The value might be a vector of floats (or a complex number). 2873bool IsSameFloatAfterCast(const APValue &value, 2874 const llvm::fltSemantics &Src, 2875 const llvm::fltSemantics &Tgt) { 2876 if (value.isFloat()) 2877 return IsSameFloatAfterCast(value.getFloat(), Src, Tgt); 2878 2879 if (value.isVector()) { 2880 for (unsigned i = 0, e = value.getVectorLength(); i != e; ++i) 2881 if (!IsSameFloatAfterCast(value.getVectorElt(i), Src, Tgt)) 2882 return false; 2883 return true; 2884 } 2885 2886 assert(value.isComplexFloat()); 2887 return (IsSameFloatAfterCast(value.getComplexFloatReal(), Src, Tgt) && 2888 IsSameFloatAfterCast(value.getComplexFloatImag(), Src, Tgt)); 2889} 2890 2891void AnalyzeImplicitConversions(Sema &S, Expr *E, SourceLocation CC); 2892 2893static bool IsZero(Sema &S, Expr *E) { 2894 // Suppress cases where we are comparing against an enum constant. 2895 if (const DeclRefExpr *DR = 2896 dyn_cast<DeclRefExpr>(E->IgnoreParenImpCasts())) 2897 if (isa<EnumConstantDecl>(DR->getDecl())) 2898 return false; 2899 2900 // Suppress cases where the '0' value is expanded from a macro. 2901 if (E->getLocStart().isMacroID()) 2902 return false; 2903 2904 llvm::APSInt Value; 2905 return E->isIntegerConstantExpr(Value, S.Context) && Value == 0; 2906} 2907 2908static bool HasEnumType(Expr *E) { 2909 // Strip off implicit integral promotions. 2910 while (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(E)) { 2911 if (ICE->getCastKind() != CK_IntegralCast && 2912 ICE->getCastKind() != CK_NoOp) 2913 break; 2914 E = ICE->getSubExpr(); 2915 } 2916 2917 return E->getType()->isEnumeralType(); 2918} 2919 2920void CheckTrivialUnsignedComparison(Sema &S, BinaryOperator *E) { 2921 BinaryOperatorKind op = E->getOpcode(); 2922 if (E->isValueDependent()) 2923 return; 2924 2925 if (op == BO_LT && IsZero(S, E->getRHS())) { 2926 S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison) 2927 << "< 0" << "false" << HasEnumType(E->getLHS()) 2928 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 2929 } else if (op == BO_GE && IsZero(S, E->getRHS())) { 2930 S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison) 2931 << ">= 0" << "true" << HasEnumType(E->getLHS()) 2932 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 2933 } else if (op == BO_GT && IsZero(S, E->getLHS())) { 2934 S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison) 2935 << "0 >" << "false" << HasEnumType(E->getRHS()) 2936 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 2937 } else if (op == BO_LE && IsZero(S, E->getLHS())) { 2938 S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison) 2939 << "0 <=" << "true" << HasEnumType(E->getRHS()) 2940 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 2941 } 2942} 2943 2944/// Analyze the operands of the given comparison. Implements the 2945/// fallback case from AnalyzeComparison. 2946void AnalyzeImpConvsInComparison(Sema &S, BinaryOperator *E) { 2947 AnalyzeImplicitConversions(S, E->getLHS(), E->getOperatorLoc()); 2948 AnalyzeImplicitConversions(S, E->getRHS(), E->getOperatorLoc()); 2949} 2950 2951/// \brief Implements -Wsign-compare. 2952/// 2953/// \param lex the left-hand expression 2954/// \param rex the right-hand expression 2955/// \param OpLoc the location of the joining operator 2956/// \param BinOpc binary opcode or 0 2957void AnalyzeComparison(Sema &S, BinaryOperator *E) { 2958 // The type the comparison is being performed in. 2959 QualType T = E->getLHS()->getType(); 2960 assert(S.Context.hasSameUnqualifiedType(T, E->getRHS()->getType()) 2961 && "comparison with mismatched types"); 2962 2963 // We don't do anything special if this isn't an unsigned integral 2964 // comparison: we're only interested in integral comparisons, and 2965 // signed comparisons only happen in cases we don't care to warn about. 2966 // 2967 // We also don't care about value-dependent expressions or expressions 2968 // whose result is a constant. 2969 if (!T->hasUnsignedIntegerRepresentation() 2970 || E->isValueDependent() || E->isIntegerConstantExpr(S.Context)) 2971 return AnalyzeImpConvsInComparison(S, E); 2972 2973 Expr *lex = E->getLHS()->IgnoreParenImpCasts(); 2974 Expr *rex = E->getRHS()->IgnoreParenImpCasts(); 2975 2976 // Check to see if one of the (unmodified) operands is of different 2977 // signedness. 2978 Expr *signedOperand, *unsignedOperand; 2979 if (lex->getType()->hasSignedIntegerRepresentation()) { 2980 assert(!rex->getType()->hasSignedIntegerRepresentation() && 2981 "unsigned comparison between two signed integer expressions?"); 2982 signedOperand = lex; 2983 unsignedOperand = rex; 2984 } else if (rex->getType()->hasSignedIntegerRepresentation()) { 2985 signedOperand = rex; 2986 unsignedOperand = lex; 2987 } else { 2988 CheckTrivialUnsignedComparison(S, E); 2989 return AnalyzeImpConvsInComparison(S, E); 2990 } 2991 2992 // Otherwise, calculate the effective range of the signed operand. 2993 IntRange signedRange = GetExprRange(S.Context, signedOperand); 2994 2995 // Go ahead and analyze implicit conversions in the operands. Note 2996 // that we skip the implicit conversions on both sides. 2997 AnalyzeImplicitConversions(S, lex, E->getOperatorLoc()); 2998 AnalyzeImplicitConversions(S, rex, E->getOperatorLoc()); 2999 3000 // If the signed range is non-negative, -Wsign-compare won't fire, 3001 // but we should still check for comparisons which are always true 3002 // or false. 3003 if (signedRange.NonNegative) 3004 return CheckTrivialUnsignedComparison(S, E); 3005 3006 // For (in)equality comparisons, if the unsigned operand is a 3007 // constant which cannot collide with a overflowed signed operand, 3008 // then reinterpreting the signed operand as unsigned will not 3009 // change the result of the comparison. 3010 if (E->isEqualityOp()) { 3011 unsigned comparisonWidth = S.Context.getIntWidth(T); 3012 IntRange unsignedRange = GetExprRange(S.Context, unsignedOperand); 3013 3014 // We should never be unable to prove that the unsigned operand is 3015 // non-negative. 3016 assert(unsignedRange.NonNegative && "unsigned range includes negative?"); 3017 3018 if (unsignedRange.Width < comparisonWidth) 3019 return; 3020 } 3021 3022 S.Diag(E->getOperatorLoc(), diag::warn_mixed_sign_comparison) 3023 << lex->getType() << rex->getType() 3024 << lex->getSourceRange() << rex->getSourceRange(); 3025} 3026 3027/// Analyzes an attempt to assign the given value to a bitfield. 3028/// 3029/// Returns true if there was something fishy about the attempt. 3030bool AnalyzeBitFieldAssignment(Sema &S, FieldDecl *Bitfield, Expr *Init, 3031 SourceLocation InitLoc) { 3032 assert(Bitfield->isBitField()); 3033 if (Bitfield->isInvalidDecl()) 3034 return false; 3035 3036 // White-list bool bitfields. 3037 if (Bitfield->getType()->isBooleanType()) 3038 return false; 3039 3040 // Ignore value- or type-dependent expressions. 3041 if (Bitfield->getBitWidth()->isValueDependent() || 3042 Bitfield->getBitWidth()->isTypeDependent() || 3043 Init->isValueDependent() || 3044 Init->isTypeDependent()) 3045 return false; 3046 3047 Expr *OriginalInit = Init->IgnoreParenImpCasts(); 3048 3049 llvm::APSInt Width(32); 3050 Expr::EvalResult InitValue; 3051 if (!Bitfield->getBitWidth()->isIntegerConstantExpr(Width, S.Context) || 3052 !OriginalInit->Evaluate(InitValue, S.Context) || 3053 !InitValue.Val.isInt()) 3054 return false; 3055 3056 const llvm::APSInt &Value = InitValue.Val.getInt(); 3057 unsigned OriginalWidth = Value.getBitWidth(); 3058 unsigned FieldWidth = Width.getZExtValue(); 3059 3060 if (OriginalWidth <= FieldWidth) 3061 return false; 3062 3063 llvm::APSInt TruncatedValue = Value.trunc(FieldWidth); 3064 3065 // It's fairly common to write values into signed bitfields 3066 // that, if sign-extended, would end up becoming a different 3067 // value. We don't want to warn about that. 3068 if (Value.isSigned() && Value.isNegative()) 3069 TruncatedValue = TruncatedValue.sext(OriginalWidth); 3070 else 3071 TruncatedValue = TruncatedValue.zext(OriginalWidth); 3072 3073 if (Value == TruncatedValue) 3074 return false; 3075 3076 std::string PrettyValue = Value.toString(10); 3077 std::string PrettyTrunc = TruncatedValue.toString(10); 3078 3079 S.Diag(InitLoc, diag::warn_impcast_bitfield_precision_constant) 3080 << PrettyValue << PrettyTrunc << OriginalInit->getType() 3081 << Init->getSourceRange(); 3082 3083 return true; 3084} 3085 3086/// Analyze the given simple or compound assignment for warning-worthy 3087/// operations. 3088void AnalyzeAssignment(Sema &S, BinaryOperator *E) { 3089 // Just recurse on the LHS. 3090 AnalyzeImplicitConversions(S, E->getLHS(), E->getOperatorLoc()); 3091 3092 // We want to recurse on the RHS as normal unless we're assigning to 3093 // a bitfield. 3094 if (FieldDecl *Bitfield = E->getLHS()->getBitField()) { 3095 if (AnalyzeBitFieldAssignment(S, Bitfield, E->getRHS(), 3096 E->getOperatorLoc())) { 3097 // Recurse, ignoring any implicit conversions on the RHS. 3098 return AnalyzeImplicitConversions(S, E->getRHS()->IgnoreParenImpCasts(), 3099 E->getOperatorLoc()); 3100 } 3101 } 3102 3103 AnalyzeImplicitConversions(S, E->getRHS(), E->getOperatorLoc()); 3104} 3105 3106/// Diagnose an implicit cast; purely a helper for CheckImplicitConversion. 3107void DiagnoseImpCast(Sema &S, Expr *E, QualType SourceType, QualType T, 3108 SourceLocation CContext, unsigned diag) { 3109 S.Diag(E->getExprLoc(), diag) 3110 << SourceType << T << E->getSourceRange() << SourceRange(CContext); 3111} 3112 3113/// Diagnose an implicit cast; purely a helper for CheckImplicitConversion. 3114void DiagnoseImpCast(Sema &S, Expr *E, QualType T, SourceLocation CContext, 3115 unsigned diag) { 3116 DiagnoseImpCast(S, E, E->getType(), T, CContext, diag); 3117} 3118 3119/// Diagnose an implicit cast from a literal expression. Also attemps to supply 3120/// fixit hints when the cast wouldn't lose information to simply write the 3121/// expression with the expected type. 3122void DiagnoseFloatingLiteralImpCast(Sema &S, FloatingLiteral *FL, QualType T, 3123 SourceLocation CContext) { 3124 // Emit the primary warning first, then try to emit a fixit hint note if 3125 // reasonable. 3126 S.Diag(FL->getExprLoc(), diag::warn_impcast_literal_float_to_integer) 3127 << FL->getType() << T << FL->getSourceRange() << SourceRange(CContext); 3128 3129 const llvm::APFloat &Value = FL->getValue(); 3130 3131 // Don't attempt to fix PPC double double literals. 3132 if (&Value.getSemantics() == &llvm::APFloat::PPCDoubleDouble) 3133 return; 3134 3135 // Try to convert this exactly to an integer. 3136 bool isExact = false; 3137 llvm::APSInt IntegerValue(S.Context.getIntWidth(T), 3138 T->hasUnsignedIntegerRepresentation()); 3139 if (Value.convertToInteger(IntegerValue, 3140 llvm::APFloat::rmTowardZero, &isExact) 3141 != llvm::APFloat::opOK || !isExact) 3142 return; 3143 3144 std::string LiteralValue = IntegerValue.toString(10); 3145 S.Diag(FL->getExprLoc(), diag::note_fix_integral_float_as_integer) 3146 << FixItHint::CreateReplacement(FL->getSourceRange(), LiteralValue); 3147} 3148 3149std::string PrettyPrintInRange(const llvm::APSInt &Value, IntRange Range) { 3150 if (!Range.Width) return "0"; 3151 3152 llvm::APSInt ValueInRange = Value; 3153 ValueInRange.setIsSigned(!Range.NonNegative); 3154 ValueInRange = ValueInRange.trunc(Range.Width); 3155 return ValueInRange.toString(10); 3156} 3157 3158static bool isFromSystemMacro(Sema &S, SourceLocation loc) { 3159 SourceManager &smgr = S.Context.getSourceManager(); 3160 return loc.isMacroID() && smgr.isInSystemHeader(smgr.getSpellingLoc(loc)); 3161} 3162 3163void CheckImplicitConversion(Sema &S, Expr *E, QualType T, 3164 SourceLocation CC, bool *ICContext = 0) { 3165 if (E->isTypeDependent() || E->isValueDependent()) return; 3166 3167 const Type *Source = S.Context.getCanonicalType(E->getType()).getTypePtr(); 3168 const Type *Target = S.Context.getCanonicalType(T).getTypePtr(); 3169 if (Source == Target) return; 3170 if (Target->isDependentType()) return; 3171 3172 // If the conversion context location is invalid don't complain. We also 3173 // don't want to emit a warning if the issue occurs from the expansion of 3174 // a system macro. The problem is that 'getSpellingLoc()' is slow, so we 3175 // delay this check as long as possible. Once we detect we are in that 3176 // scenario, we just return. 3177 if (CC.isInvalid()) 3178 return; 3179 3180 // Never diagnose implicit casts to bool. 3181 if (Target->isSpecificBuiltinType(BuiltinType::Bool)) 3182 return; 3183 3184 // Strip vector types. 3185 if (isa<VectorType>(Source)) { 3186 if (!isa<VectorType>(Target)) { 3187 if (isFromSystemMacro(S, CC)) 3188 return; 3189 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_vector_scalar); 3190 } 3191 3192 // If the vector cast is cast between two vectors of the same size, it is 3193 // a bitcast, not a conversion. 3194 if (S.Context.getTypeSize(Source) == S.Context.getTypeSize(Target)) 3195 return; 3196 3197 Source = cast<VectorType>(Source)->getElementType().getTypePtr(); 3198 Target = cast<VectorType>(Target)->getElementType().getTypePtr(); 3199 } 3200 3201 // Strip complex types. 3202 if (isa<ComplexType>(Source)) { 3203 if (!isa<ComplexType>(Target)) { 3204 if (isFromSystemMacro(S, CC)) 3205 return; 3206 3207 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_complex_scalar); 3208 } 3209 3210 Source = cast<ComplexType>(Source)->getElementType().getTypePtr(); 3211 Target = cast<ComplexType>(Target)->getElementType().getTypePtr(); 3212 } 3213 3214 const BuiltinType *SourceBT = dyn_cast<BuiltinType>(Source); 3215 const BuiltinType *TargetBT = dyn_cast<BuiltinType>(Target); 3216 3217 // If the source is floating point... 3218 if (SourceBT && SourceBT->isFloatingPoint()) { 3219 // ...and the target is floating point... 3220 if (TargetBT && TargetBT->isFloatingPoint()) { 3221 // ...then warn if we're dropping FP rank. 3222 3223 // Builtin FP kinds are ordered by increasing FP rank. 3224 if (SourceBT->getKind() > TargetBT->getKind()) { 3225 // Don't warn about float constants that are precisely 3226 // representable in the target type. 3227 Expr::EvalResult result; 3228 if (E->Evaluate(result, S.Context)) { 3229 // Value might be a float, a float vector, or a float complex. 3230 if (IsSameFloatAfterCast(result.Val, 3231 S.Context.getFloatTypeSemantics(QualType(TargetBT, 0)), 3232 S.Context.getFloatTypeSemantics(QualType(SourceBT, 0)))) 3233 return; 3234 } 3235 3236 if (isFromSystemMacro(S, CC)) 3237 return; 3238 3239 DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_float_precision); 3240 } 3241 return; 3242 } 3243 3244 // If the target is integral, always warn. 3245 if ((TargetBT && TargetBT->isInteger())) { 3246 if (isFromSystemMacro(S, CC)) 3247 return; 3248 3249 Expr *InnerE = E->IgnoreParenImpCasts(); 3250 if (FloatingLiteral *FL = dyn_cast<FloatingLiteral>(InnerE)) { 3251 DiagnoseFloatingLiteralImpCast(S, FL, T, CC); 3252 } else { 3253 DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_float_integer); 3254 } 3255 } 3256 3257 return; 3258 } 3259 3260 if (!Source->isIntegerType() || !Target->isIntegerType()) 3261 return; 3262 3263 if ((E->isNullPointerConstant(S.Context, Expr::NPC_ValueDependentIsNotNull) 3264 == Expr::NPCK_GNUNull) && Target->isIntegerType()) { 3265 S.Diag(E->getExprLoc(), diag::warn_impcast_null_pointer_to_integer) 3266 << E->getSourceRange() << clang::SourceRange(CC); 3267 return; 3268 } 3269 3270 IntRange SourceRange = GetExprRange(S.Context, E); 3271 IntRange TargetRange = IntRange::forTargetOfCanonicalType(S.Context, Target); 3272 3273 if (SourceRange.Width > TargetRange.Width) { 3274 // If the source is a constant, use a default-on diagnostic. 3275 // TODO: this should happen for bitfield stores, too. 3276 llvm::APSInt Value(32); 3277 if (E->isIntegerConstantExpr(Value, S.Context)) { 3278 if (isFromSystemMacro(S, CC)) 3279 return; 3280 3281 std::string PrettySourceValue = Value.toString(10); 3282 std::string PrettyTargetValue = PrettyPrintInRange(Value, TargetRange); 3283 3284 S.Diag(E->getExprLoc(), diag::warn_impcast_integer_precision_constant) 3285 << PrettySourceValue << PrettyTargetValue 3286 << E->getType() << T << E->getSourceRange() << clang::SourceRange(CC); 3287 return; 3288 } 3289 3290 // People want to build with -Wshorten-64-to-32 and not -Wconversion. 3291 if (isFromSystemMacro(S, CC)) 3292 return; 3293 3294 if (SourceRange.Width == 64 && TargetRange.Width == 32) 3295 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_integer_64_32); 3296 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_integer_precision); 3297 } 3298 3299 if ((TargetRange.NonNegative && !SourceRange.NonNegative) || 3300 (!TargetRange.NonNegative && SourceRange.NonNegative && 3301 SourceRange.Width == TargetRange.Width)) { 3302 3303 if (isFromSystemMacro(S, CC)) 3304 return; 3305 3306 unsigned DiagID = diag::warn_impcast_integer_sign; 3307 3308 // Traditionally, gcc has warned about this under -Wsign-compare. 3309 // We also want to warn about it in -Wconversion. 3310 // So if -Wconversion is off, use a completely identical diagnostic 3311 // in the sign-compare group. 3312 // The conditional-checking code will 3313 if (ICContext) { 3314 DiagID = diag::warn_impcast_integer_sign_conditional; 3315 *ICContext = true; 3316 } 3317 3318 return DiagnoseImpCast(S, E, T, CC, DiagID); 3319 } 3320 3321 // Diagnose conversions between different enumeration types. 3322 // In C, we pretend that the type of an EnumConstantDecl is its enumeration 3323 // type, to give us better diagnostics. 3324 QualType SourceType = E->getType(); 3325 if (!S.getLangOptions().CPlusPlus) { 3326 if (DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(E)) 3327 if (EnumConstantDecl *ECD = dyn_cast<EnumConstantDecl>(DRE->getDecl())) { 3328 EnumDecl *Enum = cast<EnumDecl>(ECD->getDeclContext()); 3329 SourceType = S.Context.getTypeDeclType(Enum); 3330 Source = S.Context.getCanonicalType(SourceType).getTypePtr(); 3331 } 3332 } 3333 3334 if (const EnumType *SourceEnum = Source->getAs<EnumType>()) 3335 if (const EnumType *TargetEnum = Target->getAs<EnumType>()) 3336 if ((SourceEnum->getDecl()->getIdentifier() || 3337 SourceEnum->getDecl()->getTypedefNameForAnonDecl()) && 3338 (TargetEnum->getDecl()->getIdentifier() || 3339 TargetEnum->getDecl()->getTypedefNameForAnonDecl()) && 3340 SourceEnum != TargetEnum) { 3341 if (isFromSystemMacro(S, CC)) 3342 return; 3343 3344 return DiagnoseImpCast(S, E, SourceType, T, CC, 3345 diag::warn_impcast_different_enum_types); 3346 } 3347 3348 return; 3349} 3350 3351void CheckConditionalOperator(Sema &S, ConditionalOperator *E, QualType T); 3352 3353void CheckConditionalOperand(Sema &S, Expr *E, QualType T, 3354 SourceLocation CC, bool &ICContext) { 3355 E = E->IgnoreParenImpCasts(); 3356 3357 if (isa<ConditionalOperator>(E)) 3358 return CheckConditionalOperator(S, cast<ConditionalOperator>(E), T); 3359 3360 AnalyzeImplicitConversions(S, E, CC); 3361 if (E->getType() != T) 3362 return CheckImplicitConversion(S, E, T, CC, &ICContext); 3363 return; 3364} 3365 3366void CheckConditionalOperator(Sema &S, ConditionalOperator *E, QualType T) { 3367 SourceLocation CC = E->getQuestionLoc(); 3368 3369 AnalyzeImplicitConversions(S, E->getCond(), CC); 3370 3371 bool Suspicious = false; 3372 CheckConditionalOperand(S, E->getTrueExpr(), T, CC, Suspicious); 3373 CheckConditionalOperand(S, E->getFalseExpr(), T, CC, Suspicious); 3374 3375 // If -Wconversion would have warned about either of the candidates 3376 // for a signedness conversion to the context type... 3377 if (!Suspicious) return; 3378 3379 // ...but it's currently ignored... 3380 if (S.Diags.getDiagnosticLevel(diag::warn_impcast_integer_sign_conditional, 3381 CC)) 3382 return; 3383 3384 // ...then check whether it would have warned about either of the 3385 // candidates for a signedness conversion to the condition type. 3386 if (E->getType() == T) return; 3387 3388 Suspicious = false; 3389 CheckImplicitConversion(S, E->getTrueExpr()->IgnoreParenImpCasts(), 3390 E->getType(), CC, &Suspicious); 3391 if (!Suspicious) 3392 CheckImplicitConversion(S, E->getFalseExpr()->IgnoreParenImpCasts(), 3393 E->getType(), CC, &Suspicious); 3394} 3395 3396/// AnalyzeImplicitConversions - Find and report any interesting 3397/// implicit conversions in the given expression. There are a couple 3398/// of competing diagnostics here, -Wconversion and -Wsign-compare. 3399void AnalyzeImplicitConversions(Sema &S, Expr *OrigE, SourceLocation CC) { 3400 QualType T = OrigE->getType(); 3401 Expr *E = OrigE->IgnoreParenImpCasts(); 3402 3403 // For conditional operators, we analyze the arguments as if they 3404 // were being fed directly into the output. 3405 if (isa<ConditionalOperator>(E)) { 3406 ConditionalOperator *CO = cast<ConditionalOperator>(E); 3407 CheckConditionalOperator(S, CO, T); 3408 return; 3409 } 3410 3411 // Go ahead and check any implicit conversions we might have skipped. 3412 // The non-canonical typecheck is just an optimization; 3413 // CheckImplicitConversion will filter out dead implicit conversions. 3414 if (E->getType() != T) 3415 CheckImplicitConversion(S, E, T, CC); 3416 3417 // Now continue drilling into this expression. 3418 3419 // Skip past explicit casts. 3420 if (isa<ExplicitCastExpr>(E)) { 3421 E = cast<ExplicitCastExpr>(E)->getSubExpr()->IgnoreParenImpCasts(); 3422 return AnalyzeImplicitConversions(S, E, CC); 3423 } 3424 3425 if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) { 3426 // Do a somewhat different check with comparison operators. 3427 if (BO->isComparisonOp()) 3428 return AnalyzeComparison(S, BO); 3429 3430 // And with assignments and compound assignments. 3431 if (BO->isAssignmentOp()) 3432 return AnalyzeAssignment(S, BO); 3433 } 3434 3435 // These break the otherwise-useful invariant below. Fortunately, 3436 // we don't really need to recurse into them, because any internal 3437 // expressions should have been analyzed already when they were 3438 // built into statements. 3439 if (isa<StmtExpr>(E)) return; 3440 3441 // Don't descend into unevaluated contexts. 3442 if (isa<UnaryExprOrTypeTraitExpr>(E)) return; 3443 3444 // Now just recurse over the expression's children. 3445 CC = E->getExprLoc(); 3446 for (Stmt::child_range I = E->children(); I; ++I) 3447 AnalyzeImplicitConversions(S, cast<Expr>(*I), CC); 3448} 3449 3450} // end anonymous namespace 3451 3452/// Diagnoses "dangerous" implicit conversions within the given 3453/// expression (which is a full expression). Implements -Wconversion 3454/// and -Wsign-compare. 3455/// 3456/// \param CC the "context" location of the implicit conversion, i.e. 3457/// the most location of the syntactic entity requiring the implicit 3458/// conversion 3459void Sema::CheckImplicitConversions(Expr *E, SourceLocation CC) { 3460 // Don't diagnose in unevaluated contexts. 3461 if (ExprEvalContexts.back().Context == Sema::Unevaluated) 3462 return; 3463 3464 // Don't diagnose for value- or type-dependent expressions. 3465 if (E->isTypeDependent() || E->isValueDependent()) 3466 return; 3467 3468 // Check for array bounds violations in cases where the check isn't triggered 3469 // elsewhere for other Expr types (like BinaryOperators), e.g. when an 3470 // ArraySubscriptExpr is on the RHS of a variable initialization. 3471 CheckArrayAccess(E); 3472 3473 // This is not the right CC for (e.g.) a variable initialization. 3474 AnalyzeImplicitConversions(*this, E, CC); 3475} 3476 3477void Sema::CheckBitFieldInitialization(SourceLocation InitLoc, 3478 FieldDecl *BitField, 3479 Expr *Init) { 3480 (void) AnalyzeBitFieldAssignment(*this, BitField, Init, InitLoc); 3481} 3482 3483/// CheckParmsForFunctionDef - Check that the parameters of the given 3484/// function are appropriate for the definition of a function. This 3485/// takes care of any checks that cannot be performed on the 3486/// declaration itself, e.g., that the types of each of the function 3487/// parameters are complete. 3488bool Sema::CheckParmsForFunctionDef(ParmVarDecl **P, ParmVarDecl **PEnd, 3489 bool CheckParameterNames) { 3490 bool HasInvalidParm = false; 3491 for (; P != PEnd; ++P) { 3492 ParmVarDecl *Param = *P; 3493 3494 // C99 6.7.5.3p4: the parameters in a parameter type list in a 3495 // function declarator that is part of a function definition of 3496 // that function shall not have incomplete type. 3497 // 3498 // This is also C++ [dcl.fct]p6. 3499 if (!Param->isInvalidDecl() && 3500 RequireCompleteType(Param->getLocation(), Param->getType(), 3501 diag::err_typecheck_decl_incomplete_type)) { 3502 Param->setInvalidDecl(); 3503 HasInvalidParm = true; 3504 } 3505 3506 // C99 6.9.1p5: If the declarator includes a parameter type list, the 3507 // declaration of each parameter shall include an identifier. 3508 if (CheckParameterNames && 3509 Param->getIdentifier() == 0 && 3510 !Param->isImplicit() && 3511 !getLangOptions().CPlusPlus) 3512 Diag(Param->getLocation(), diag::err_parameter_name_omitted); 3513 3514 // C99 6.7.5.3p12: 3515 // If the function declarator is not part of a definition of that 3516 // function, parameters may have incomplete type and may use the [*] 3517 // notation in their sequences of declarator specifiers to specify 3518 // variable length array types. 3519 QualType PType = Param->getOriginalType(); 3520 if (const ArrayType *AT = Context.getAsArrayType(PType)) { 3521 if (AT->getSizeModifier() == ArrayType::Star) { 3522 // FIXME: This diagnosic should point the the '[*]' if source-location 3523 // information is added for it. 3524 Diag(Param->getLocation(), diag::err_array_star_in_function_definition); 3525 } 3526 } 3527 } 3528 3529 return HasInvalidParm; 3530} 3531 3532/// CheckCastAlign - Implements -Wcast-align, which warns when a 3533/// pointer cast increases the alignment requirements. 3534void Sema::CheckCastAlign(Expr *Op, QualType T, SourceRange TRange) { 3535 // This is actually a lot of work to potentially be doing on every 3536 // cast; don't do it if we're ignoring -Wcast_align (as is the default). 3537 if (getDiagnostics().getDiagnosticLevel(diag::warn_cast_align, 3538 TRange.getBegin()) 3539 == Diagnostic::Ignored) 3540 return; 3541 3542 // Ignore dependent types. 3543 if (T->isDependentType() || Op->getType()->isDependentType()) 3544 return; 3545 3546 // Require that the destination be a pointer type. 3547 const PointerType *DestPtr = T->getAs<PointerType>(); 3548 if (!DestPtr) return; 3549 3550 // If the destination has alignment 1, we're done. 3551 QualType DestPointee = DestPtr->getPointeeType(); 3552 if (DestPointee->isIncompleteType()) return; 3553 CharUnits DestAlign = Context.getTypeAlignInChars(DestPointee); 3554 if (DestAlign.isOne()) return; 3555 3556 // Require that the source be a pointer type. 3557 const PointerType *SrcPtr = Op->getType()->getAs<PointerType>(); 3558 if (!SrcPtr) return; 3559 QualType SrcPointee = SrcPtr->getPointeeType(); 3560 3561 // Whitelist casts from cv void*. We already implicitly 3562 // whitelisted casts to cv void*, since they have alignment 1. 3563 // Also whitelist casts involving incomplete types, which implicitly 3564 // includes 'void'. 3565 if (SrcPointee->isIncompleteType()) return; 3566 3567 CharUnits SrcAlign = Context.getTypeAlignInChars(SrcPointee); 3568 if (SrcAlign >= DestAlign) return; 3569 3570 Diag(TRange.getBegin(), diag::warn_cast_align) 3571 << Op->getType() << T 3572 << static_cast<unsigned>(SrcAlign.getQuantity()) 3573 << static_cast<unsigned>(DestAlign.getQuantity()) 3574 << TRange << Op->getSourceRange(); 3575} 3576 3577static const Type* getElementType(const Expr *BaseExpr) { 3578 const Type* EltType = BaseExpr->getType().getTypePtr(); 3579 if (EltType->isAnyPointerType()) 3580 return EltType->getPointeeType().getTypePtr(); 3581 else if (EltType->isArrayType()) 3582 return EltType->getBaseElementTypeUnsafe(); 3583 return EltType; 3584} 3585 3586/// \brief Check whether this array fits the idiom of a size-one tail padded 3587/// array member of a struct. 3588/// 3589/// We avoid emitting out-of-bounds access warnings for such arrays as they are 3590/// commonly used to emulate flexible arrays in C89 code. 3591static bool IsTailPaddedMemberArray(Sema &S, llvm::APInt Size, 3592 const NamedDecl *ND) { 3593 if (Size != 1 || !ND) return false; 3594 3595 const FieldDecl *FD = dyn_cast<FieldDecl>(ND); 3596 if (!FD) return false; 3597 3598 // Don't consider sizes resulting from macro expansions or template argument 3599 // substitution to form C89 tail-padded arrays. 3600 ConstantArrayTypeLoc TL = 3601 cast<ConstantArrayTypeLoc>(FD->getTypeSourceInfo()->getTypeLoc()); 3602 const Expr *SizeExpr = dyn_cast<IntegerLiteral>(TL.getSizeExpr()); 3603 if (!SizeExpr || SizeExpr->getExprLoc().isMacroID()) 3604 return false; 3605 3606 const RecordDecl *RD = dyn_cast<RecordDecl>(FD->getDeclContext()); 3607 if (!RD || !RD->isStruct()) 3608 return false; 3609 3610 // See if this is the last field decl in the record. 3611 const Decl *D = FD; 3612 while ((D = D->getNextDeclInContext())) 3613 if (isa<FieldDecl>(D)) 3614 return false; 3615 return true; 3616} 3617 3618void Sema::CheckArrayAccess(const Expr *BaseExpr, const Expr *IndexExpr, 3619 bool isSubscript, bool AllowOnePastEnd) { 3620 const Type* EffectiveType = getElementType(BaseExpr); 3621 BaseExpr = BaseExpr->IgnoreParenCasts(); 3622 IndexExpr = IndexExpr->IgnoreParenCasts(); 3623 3624 const ConstantArrayType *ArrayTy = 3625 Context.getAsConstantArrayType(BaseExpr->getType()); 3626 if (!ArrayTy) 3627 return; 3628 3629 if (IndexExpr->isValueDependent()) 3630 return; 3631 llvm::APSInt index; 3632 if (!IndexExpr->isIntegerConstantExpr(index, Context)) 3633 return; 3634 3635 const NamedDecl *ND = NULL; 3636 if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(BaseExpr)) 3637 ND = dyn_cast<NamedDecl>(DRE->getDecl()); 3638 if (const MemberExpr *ME = dyn_cast<MemberExpr>(BaseExpr)) 3639 ND = dyn_cast<NamedDecl>(ME->getMemberDecl()); 3640 3641 if (index.isUnsigned() || !index.isNegative()) { 3642 llvm::APInt size = ArrayTy->getSize(); 3643 if (!size.isStrictlyPositive()) 3644 return; 3645 3646 const Type* BaseType = getElementType(BaseExpr); 3647 if (!isSubscript && BaseType != EffectiveType) { 3648 // Make sure we're comparing apples to apples when comparing index to size 3649 uint64_t ptrarith_typesize = Context.getTypeSize(EffectiveType); 3650 uint64_t array_typesize = Context.getTypeSize(BaseType); 3651 // Handle ptrarith_typesize being zero, such as when casting to void* 3652 if (!ptrarith_typesize) ptrarith_typesize = 1; 3653 if (ptrarith_typesize != array_typesize) { 3654 // There's a cast to a different size type involved 3655 uint64_t ratio = array_typesize / ptrarith_typesize; 3656 // TODO: Be smarter about handling cases where array_typesize is not a 3657 // multiple of ptrarith_typesize 3658 if (ptrarith_typesize * ratio == array_typesize) 3659 size *= llvm::APInt(size.getBitWidth(), ratio); 3660 } 3661 } 3662 3663 if (size.getBitWidth() > index.getBitWidth()) 3664 index = index.sext(size.getBitWidth()); 3665 else if (size.getBitWidth() < index.getBitWidth()) 3666 size = size.sext(index.getBitWidth()); 3667 3668 // For array subscripting the index must be less than size, but for pointer 3669 // arithmetic also allow the index (offset) to be equal to size since 3670 // computing the next address after the end of the array is legal and 3671 // commonly done e.g. in C++ iterators and range-based for loops. 3672 if (AllowOnePastEnd ? index.sle(size) : index.slt(size)) 3673 return; 3674 3675 // Also don't warn for arrays of size 1 which are members of some 3676 // structure. These are often used to approximate flexible arrays in C89 3677 // code. 3678 if (IsTailPaddedMemberArray(*this, size, ND)) 3679 return; 3680 3681 unsigned DiagID = diag::warn_ptr_arith_exceeds_bounds; 3682 if (isSubscript) 3683 DiagID = diag::warn_array_index_exceeds_bounds; 3684 3685 DiagRuntimeBehavior(BaseExpr->getLocStart(), BaseExpr, 3686 PDiag(DiagID) << index.toString(10, true) 3687 << size.toString(10, true) 3688 << (unsigned)size.getLimitedValue(~0U) 3689 << IndexExpr->getSourceRange()); 3690 } else { 3691 unsigned DiagID = diag::warn_array_index_precedes_bounds; 3692 if (!isSubscript) { 3693 DiagID = diag::warn_ptr_arith_precedes_bounds; 3694 if (index.isNegative()) index = -index; 3695 } 3696 3697 DiagRuntimeBehavior(BaseExpr->getLocStart(), BaseExpr, 3698 PDiag(DiagID) << index.toString(10, true) 3699 << IndexExpr->getSourceRange()); 3700 } 3701 3702 if (ND) 3703 DiagRuntimeBehavior(ND->getLocStart(), BaseExpr, 3704 PDiag(diag::note_array_index_out_of_bounds) 3705 << ND->getDeclName()); 3706} 3707 3708void Sema::CheckArrayAccess(const Expr *expr) { 3709 int AllowOnePastEnd = 0; 3710 while (expr) { 3711 expr = expr->IgnoreParenImpCasts(); 3712 switch (expr->getStmtClass()) { 3713 case Stmt::ArraySubscriptExprClass: { 3714 const ArraySubscriptExpr *ASE = cast<ArraySubscriptExpr>(expr); 3715 CheckArrayAccess(ASE->getBase(), ASE->getIdx(), true, 3716 AllowOnePastEnd > 0); 3717 return; 3718 } 3719 case Stmt::UnaryOperatorClass: { 3720 // Only unwrap the * and & unary operators 3721 const UnaryOperator *UO = cast<UnaryOperator>(expr); 3722 expr = UO->getSubExpr(); 3723 switch (UO->getOpcode()) { 3724 case UO_AddrOf: 3725 AllowOnePastEnd++; 3726 break; 3727 case UO_Deref: 3728 AllowOnePastEnd--; 3729 break; 3730 default: 3731 return; 3732 } 3733 break; 3734 } 3735 case Stmt::ConditionalOperatorClass: { 3736 const ConditionalOperator *cond = cast<ConditionalOperator>(expr); 3737 if (const Expr *lhs = cond->getLHS()) 3738 CheckArrayAccess(lhs); 3739 if (const Expr *rhs = cond->getRHS()) 3740 CheckArrayAccess(rhs); 3741 return; 3742 } 3743 default: 3744 return; 3745 } 3746 } 3747} 3748 3749//===--- CHECK: Objective-C retain cycles ----------------------------------// 3750 3751namespace { 3752 struct RetainCycleOwner { 3753 RetainCycleOwner() : Variable(0), Indirect(false) {} 3754 VarDecl *Variable; 3755 SourceRange Range; 3756 SourceLocation Loc; 3757 bool Indirect; 3758 3759 void setLocsFrom(Expr *e) { 3760 Loc = e->getExprLoc(); 3761 Range = e->getSourceRange(); 3762 } 3763 }; 3764} 3765 3766/// Consider whether capturing the given variable can possibly lead to 3767/// a retain cycle. 3768static bool considerVariable(VarDecl *var, Expr *ref, RetainCycleOwner &owner) { 3769 // In ARC, it's captured strongly iff the variable has __strong 3770 // lifetime. In MRR, it's captured strongly if the variable is 3771 // __block and has an appropriate type. 3772 if (var->getType().getObjCLifetime() != Qualifiers::OCL_Strong) 3773 return false; 3774 3775 owner.Variable = var; 3776 owner.setLocsFrom(ref); 3777 return true; 3778} 3779 3780static bool findRetainCycleOwner(Expr *e, RetainCycleOwner &owner) { 3781 while (true) { 3782 e = e->IgnoreParens(); 3783 if (CastExpr *cast = dyn_cast<CastExpr>(e)) { 3784 switch (cast->getCastKind()) { 3785 case CK_BitCast: 3786 case CK_LValueBitCast: 3787 case CK_LValueToRValue: 3788 case CK_ObjCReclaimReturnedObject: 3789 e = cast->getSubExpr(); 3790 continue; 3791 3792 case CK_GetObjCProperty: { 3793 // Bail out if this isn't a strong explicit property. 3794 const ObjCPropertyRefExpr *pre = cast->getSubExpr()->getObjCProperty(); 3795 if (pre->isImplicitProperty()) return false; 3796 ObjCPropertyDecl *property = pre->getExplicitProperty(); 3797 if (!(property->getPropertyAttributes() & 3798 (ObjCPropertyDecl::OBJC_PR_retain | 3799 ObjCPropertyDecl::OBJC_PR_copy | 3800 ObjCPropertyDecl::OBJC_PR_strong)) && 3801 !(property->getPropertyIvarDecl() && 3802 property->getPropertyIvarDecl()->getType() 3803 .getObjCLifetime() == Qualifiers::OCL_Strong)) 3804 return false; 3805 3806 owner.Indirect = true; 3807 e = const_cast<Expr*>(pre->getBase()); 3808 continue; 3809 } 3810 3811 default: 3812 return false; 3813 } 3814 } 3815 3816 if (ObjCIvarRefExpr *ref = dyn_cast<ObjCIvarRefExpr>(e)) { 3817 ObjCIvarDecl *ivar = ref->getDecl(); 3818 if (ivar->getType().getObjCLifetime() != Qualifiers::OCL_Strong) 3819 return false; 3820 3821 // Try to find a retain cycle in the base. 3822 if (!findRetainCycleOwner(ref->getBase(), owner)) 3823 return false; 3824 3825 if (ref->isFreeIvar()) owner.setLocsFrom(ref); 3826 owner.Indirect = true; 3827 return true; 3828 } 3829 3830 if (DeclRefExpr *ref = dyn_cast<DeclRefExpr>(e)) { 3831 VarDecl *var = dyn_cast<VarDecl>(ref->getDecl()); 3832 if (!var) return false; 3833 return considerVariable(var, ref, owner); 3834 } 3835 3836 if (BlockDeclRefExpr *ref = dyn_cast<BlockDeclRefExpr>(e)) { 3837 owner.Variable = ref->getDecl(); 3838 owner.setLocsFrom(ref); 3839 return true; 3840 } 3841 3842 if (MemberExpr *member = dyn_cast<MemberExpr>(e)) { 3843 if (member->isArrow()) return false; 3844 3845 // Don't count this as an indirect ownership. 3846 e = member->getBase(); 3847 continue; 3848 } 3849 3850 // Array ivars? 3851 3852 return false; 3853 } 3854} 3855 3856namespace { 3857 struct FindCaptureVisitor : EvaluatedExprVisitor<FindCaptureVisitor> { 3858 FindCaptureVisitor(ASTContext &Context, VarDecl *variable) 3859 : EvaluatedExprVisitor<FindCaptureVisitor>(Context), 3860 Variable(variable), Capturer(0) {} 3861 3862 VarDecl *Variable; 3863 Expr *Capturer; 3864 3865 void VisitDeclRefExpr(DeclRefExpr *ref) { 3866 if (ref->getDecl() == Variable && !Capturer) 3867 Capturer = ref; 3868 } 3869 3870 void VisitBlockDeclRefExpr(BlockDeclRefExpr *ref) { 3871 if (ref->getDecl() == Variable && !Capturer) 3872 Capturer = ref; 3873 } 3874 3875 void VisitObjCIvarRefExpr(ObjCIvarRefExpr *ref) { 3876 if (Capturer) return; 3877 Visit(ref->getBase()); 3878 if (Capturer && ref->isFreeIvar()) 3879 Capturer = ref; 3880 } 3881 3882 void VisitBlockExpr(BlockExpr *block) { 3883 // Look inside nested blocks 3884 if (block->getBlockDecl()->capturesVariable(Variable)) 3885 Visit(block->getBlockDecl()->getBody()); 3886 } 3887 }; 3888} 3889 3890/// Check whether the given argument is a block which captures a 3891/// variable. 3892static Expr *findCapturingExpr(Sema &S, Expr *e, RetainCycleOwner &owner) { 3893 assert(owner.Variable && owner.Loc.isValid()); 3894 3895 e = e->IgnoreParenCasts(); 3896 BlockExpr *block = dyn_cast<BlockExpr>(e); 3897 if (!block || !block->getBlockDecl()->capturesVariable(owner.Variable)) 3898 return 0; 3899 3900 FindCaptureVisitor visitor(S.Context, owner.Variable); 3901 visitor.Visit(block->getBlockDecl()->getBody()); 3902 return visitor.Capturer; 3903} 3904 3905static void diagnoseRetainCycle(Sema &S, Expr *capturer, 3906 RetainCycleOwner &owner) { 3907 assert(capturer); 3908 assert(owner.Variable && owner.Loc.isValid()); 3909 3910 S.Diag(capturer->getExprLoc(), diag::warn_arc_retain_cycle) 3911 << owner.Variable << capturer->getSourceRange(); 3912 S.Diag(owner.Loc, diag::note_arc_retain_cycle_owner) 3913 << owner.Indirect << owner.Range; 3914} 3915 3916/// Check for a keyword selector that starts with the word 'add' or 3917/// 'set'. 3918static bool isSetterLikeSelector(Selector sel) { 3919 if (sel.isUnarySelector()) return false; 3920 3921 StringRef str = sel.getNameForSlot(0); 3922 while (!str.empty() && str.front() == '_') str = str.substr(1); 3923 if (str.startswith("set") || str.startswith("add")) 3924 str = str.substr(3); 3925 else 3926 return false; 3927 3928 if (str.empty()) return true; 3929 return !islower(str.front()); 3930} 3931 3932/// Check a message send to see if it's likely to cause a retain cycle. 3933void Sema::checkRetainCycles(ObjCMessageExpr *msg) { 3934 // Only check instance methods whose selector looks like a setter. 3935 if (!msg->isInstanceMessage() || !isSetterLikeSelector(msg->getSelector())) 3936 return; 3937 3938 // Try to find a variable that the receiver is strongly owned by. 3939 RetainCycleOwner owner; 3940 if (msg->getReceiverKind() == ObjCMessageExpr::Instance) { 3941 if (!findRetainCycleOwner(msg->getInstanceReceiver(), owner)) 3942 return; 3943 } else { 3944 assert(msg->getReceiverKind() == ObjCMessageExpr::SuperInstance); 3945 owner.Variable = getCurMethodDecl()->getSelfDecl(); 3946 owner.Loc = msg->getSuperLoc(); 3947 owner.Range = msg->getSuperLoc(); 3948 } 3949 3950 // Check whether the receiver is captured by any of the arguments. 3951 for (unsigned i = 0, e = msg->getNumArgs(); i != e; ++i) 3952 if (Expr *capturer = findCapturingExpr(*this, msg->getArg(i), owner)) 3953 return diagnoseRetainCycle(*this, capturer, owner); 3954} 3955 3956/// Check a property assign to see if it's likely to cause a retain cycle. 3957void Sema::checkRetainCycles(Expr *receiver, Expr *argument) { 3958 RetainCycleOwner owner; 3959 if (!findRetainCycleOwner(receiver, owner)) 3960 return; 3961 3962 if (Expr *capturer = findCapturingExpr(*this, argument, owner)) 3963 diagnoseRetainCycle(*this, capturer, owner); 3964} 3965 3966bool Sema::checkUnsafeAssigns(SourceLocation Loc, 3967 QualType LHS, Expr *RHS) { 3968 Qualifiers::ObjCLifetime LT = LHS.getObjCLifetime(); 3969 if (LT != Qualifiers::OCL_Weak && LT != Qualifiers::OCL_ExplicitNone) 3970 return false; 3971 // strip off any implicit cast added to get to the one arc-specific 3972 while (ImplicitCastExpr *cast = dyn_cast<ImplicitCastExpr>(RHS)) { 3973 if (cast->getCastKind() == CK_ObjCConsumeObject) { 3974 Diag(Loc, diag::warn_arc_retained_assign) 3975 << (LT == Qualifiers::OCL_ExplicitNone) 3976 << RHS->getSourceRange(); 3977 return true; 3978 } 3979 RHS = cast->getSubExpr(); 3980 } 3981 return false; 3982} 3983 3984void Sema::checkUnsafeExprAssigns(SourceLocation Loc, 3985 Expr *LHS, Expr *RHS) { 3986 QualType LHSType = LHS->getType(); 3987 if (checkUnsafeAssigns(Loc, LHSType, RHS)) 3988 return; 3989 Qualifiers::ObjCLifetime LT = LHSType.getObjCLifetime(); 3990 // FIXME. Check for other life times. 3991 if (LT != Qualifiers::OCL_None) 3992 return; 3993 3994 if (ObjCPropertyRefExpr *PRE = dyn_cast<ObjCPropertyRefExpr>(LHS)) { 3995 if (PRE->isImplicitProperty()) 3996 return; 3997 const ObjCPropertyDecl *PD = PRE->getExplicitProperty(); 3998 if (!PD) 3999 return; 4000 4001 unsigned Attributes = PD->getPropertyAttributes(); 4002 if (Attributes & ObjCPropertyDecl::OBJC_PR_assign) 4003 while (ImplicitCastExpr *cast = dyn_cast<ImplicitCastExpr>(RHS)) { 4004 if (cast->getCastKind() == CK_ObjCConsumeObject) { 4005 Diag(Loc, diag::warn_arc_retained_property_assign) 4006 << RHS->getSourceRange(); 4007 return; 4008 } 4009 RHS = cast->getSubExpr(); 4010 } 4011 } 4012} 4013