SemaChecking.cpp revision bbcb7ea8a062a8f1d5cb504e4518a5d4fbab873a
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/Initialization.h" 16#include "clang/Sema/Sema.h" 17#include "clang/Sema/SemaInternal.h" 18#include "clang/Sema/ScopeInfo.h" 19#include "clang/Analysis/Analyses/FormatString.h" 20#include "clang/AST/ASTContext.h" 21#include "clang/AST/CharUnits.h" 22#include "clang/AST/DeclCXX.h" 23#include "clang/AST/DeclObjC.h" 24#include "clang/AST/ExprCXX.h" 25#include "clang/AST/ExprObjC.h" 26#include "clang/AST/EvaluatedExprVisitor.h" 27#include "clang/AST/DeclObjC.h" 28#include "clang/AST/StmtCXX.h" 29#include "clang/AST/StmtObjC.h" 30#include "clang/Lex/Preprocessor.h" 31#include "llvm/ADT/BitVector.h" 32#include "llvm/ADT/STLExtras.h" 33#include "llvm/Support/raw_ostream.h" 34#include "clang/Basic/TargetBuiltins.h" 35#include "clang/Basic/TargetInfo.h" 36#include "clang/Basic/ConvertUTF.h" 37#include <limits> 38using namespace clang; 39using namespace sema; 40 41SourceLocation Sema::getLocationOfStringLiteralByte(const StringLiteral *SL, 42 unsigned ByteNo) const { 43 return SL->getLocationOfByte(ByteNo, PP.getSourceManager(), 44 PP.getLangOptions(), PP.getTargetInfo()); 45} 46 47 48/// CheckablePrintfAttr - does a function call have a "printf" attribute 49/// and arguments that merit checking? 50bool Sema::CheckablePrintfAttr(const FormatAttr *Format, CallExpr *TheCall) { 51 if (Format->getType() == "printf") return true; 52 if (Format->getType() == "printf0") { 53 // printf0 allows null "format" string; if so don't check format/args 54 unsigned format_idx = Format->getFormatIdx() - 1; 55 // Does the index refer to the implicit object argument? 56 if (isa<CXXMemberCallExpr>(TheCall)) { 57 if (format_idx == 0) 58 return false; 59 --format_idx; 60 } 61 if (format_idx < TheCall->getNumArgs()) { 62 Expr *Format = TheCall->getArg(format_idx)->IgnoreParenCasts(); 63 if (!Format->isNullPointerConstant(Context, 64 Expr::NPC_ValueDependentIsNull)) 65 return true; 66 } 67 } 68 return false; 69} 70 71/// Checks that a call expression's argument count is the desired number. 72/// This is useful when doing custom type-checking. Returns true on error. 73static bool checkArgCount(Sema &S, CallExpr *call, unsigned desiredArgCount) { 74 unsigned argCount = call->getNumArgs(); 75 if (argCount == desiredArgCount) return false; 76 77 if (argCount < desiredArgCount) 78 return S.Diag(call->getLocEnd(), diag::err_typecheck_call_too_few_args) 79 << 0 /*function call*/ << desiredArgCount << argCount 80 << call->getSourceRange(); 81 82 // Highlight all the excess arguments. 83 SourceRange range(call->getArg(desiredArgCount)->getLocStart(), 84 call->getArg(argCount - 1)->getLocEnd()); 85 86 return S.Diag(range.getBegin(), diag::err_typecheck_call_too_many_args) 87 << 0 /*function call*/ << desiredArgCount << argCount 88 << call->getArg(1)->getSourceRange(); 89} 90 91ExprResult 92Sema::CheckBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) { 93 ExprResult TheCallResult(Owned(TheCall)); 94 95 // Find out if any arguments are required to be integer constant expressions. 96 unsigned ICEArguments = 0; 97 ASTContext::GetBuiltinTypeError Error; 98 Context.GetBuiltinType(BuiltinID, Error, &ICEArguments); 99 if (Error != ASTContext::GE_None) 100 ICEArguments = 0; // Don't diagnose previously diagnosed errors. 101 102 // If any arguments are required to be ICE's, check and diagnose. 103 for (unsigned ArgNo = 0; ICEArguments != 0; ++ArgNo) { 104 // Skip arguments not required to be ICE's. 105 if ((ICEArguments & (1 << ArgNo)) == 0) continue; 106 107 llvm::APSInt Result; 108 if (SemaBuiltinConstantArg(TheCall, ArgNo, Result)) 109 return true; 110 ICEArguments &= ~(1 << ArgNo); 111 } 112 113 switch (BuiltinID) { 114 case Builtin::BI__builtin___CFStringMakeConstantString: 115 assert(TheCall->getNumArgs() == 1 && 116 "Wrong # arguments to builtin CFStringMakeConstantString"); 117 if (CheckObjCString(TheCall->getArg(0))) 118 return ExprError(); 119 break; 120 case Builtin::BI__builtin_stdarg_start: 121 case Builtin::BI__builtin_va_start: 122 if (SemaBuiltinVAStart(TheCall)) 123 return ExprError(); 124 break; 125 case Builtin::BI__builtin_isgreater: 126 case Builtin::BI__builtin_isgreaterequal: 127 case Builtin::BI__builtin_isless: 128 case Builtin::BI__builtin_islessequal: 129 case Builtin::BI__builtin_islessgreater: 130 case Builtin::BI__builtin_isunordered: 131 if (SemaBuiltinUnorderedCompare(TheCall)) 132 return ExprError(); 133 break; 134 case Builtin::BI__builtin_fpclassify: 135 if (SemaBuiltinFPClassification(TheCall, 6)) 136 return ExprError(); 137 break; 138 case Builtin::BI__builtin_isfinite: 139 case Builtin::BI__builtin_isinf: 140 case Builtin::BI__builtin_isinf_sign: 141 case Builtin::BI__builtin_isnan: 142 case Builtin::BI__builtin_isnormal: 143 if (SemaBuiltinFPClassification(TheCall, 1)) 144 return ExprError(); 145 break; 146 case Builtin::BI__builtin_shufflevector: 147 return SemaBuiltinShuffleVector(TheCall); 148 // TheCall will be freed by the smart pointer here, but that's fine, since 149 // SemaBuiltinShuffleVector guts it, but then doesn't release it. 150 case Builtin::BI__builtin_prefetch: 151 if (SemaBuiltinPrefetch(TheCall)) 152 return ExprError(); 153 break; 154 case Builtin::BI__builtin_object_size: 155 if (SemaBuiltinObjectSize(TheCall)) 156 return ExprError(); 157 break; 158 case Builtin::BI__builtin_longjmp: 159 if (SemaBuiltinLongjmp(TheCall)) 160 return ExprError(); 161 break; 162 163 case Builtin::BI__builtin_classify_type: 164 if (checkArgCount(*this, TheCall, 1)) return true; 165 TheCall->setType(Context.IntTy); 166 break; 167 case Builtin::BI__builtin_constant_p: 168 if (checkArgCount(*this, TheCall, 1)) return true; 169 TheCall->setType(Context.IntTy); 170 break; 171 case Builtin::BI__sync_fetch_and_add: 172 case Builtin::BI__sync_fetch_and_sub: 173 case Builtin::BI__sync_fetch_and_or: 174 case Builtin::BI__sync_fetch_and_and: 175 case Builtin::BI__sync_fetch_and_xor: 176 case Builtin::BI__sync_add_and_fetch: 177 case Builtin::BI__sync_sub_and_fetch: 178 case Builtin::BI__sync_and_and_fetch: 179 case Builtin::BI__sync_or_and_fetch: 180 case Builtin::BI__sync_xor_and_fetch: 181 case Builtin::BI__sync_val_compare_and_swap: 182 case Builtin::BI__sync_bool_compare_and_swap: 183 case Builtin::BI__sync_lock_test_and_set: 184 case Builtin::BI__sync_lock_release: 185 case Builtin::BI__sync_swap: 186 return SemaBuiltinAtomicOverloaded(move(TheCallResult)); 187 } 188 189 // Since the target specific builtins for each arch overlap, only check those 190 // of the arch we are compiling for. 191 if (BuiltinID >= Builtin::FirstTSBuiltin) { 192 switch (Context.getTargetInfo().getTriple().getArch()) { 193 case llvm::Triple::arm: 194 case llvm::Triple::thumb: 195 if (CheckARMBuiltinFunctionCall(BuiltinID, TheCall)) 196 return ExprError(); 197 break; 198 default: 199 break; 200 } 201 } 202 203 return move(TheCallResult); 204} 205 206// Get the valid immediate range for the specified NEON type code. 207static unsigned RFT(unsigned t, bool shift = false) { 208 bool quad = t & 0x10; 209 210 switch (t & 0x7) { 211 case 0: // i8 212 return shift ? 7 : (8 << (int)quad) - 1; 213 case 1: // i16 214 return shift ? 15 : (4 << (int)quad) - 1; 215 case 2: // i32 216 return shift ? 31 : (2 << (int)quad) - 1; 217 case 3: // i64 218 return shift ? 63 : (1 << (int)quad) - 1; 219 case 4: // f32 220 assert(!shift && "cannot shift float types!"); 221 return (2 << (int)quad) - 1; 222 case 5: // poly8 223 return shift ? 7 : (8 << (int)quad) - 1; 224 case 6: // poly16 225 return shift ? 15 : (4 << (int)quad) - 1; 226 case 7: // float16 227 assert(!shift && "cannot shift float types!"); 228 return (4 << (int)quad) - 1; 229 } 230 return 0; 231} 232 233bool Sema::CheckARMBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) { 234 llvm::APSInt Result; 235 236 unsigned mask = 0; 237 unsigned TV = 0; 238 switch (BuiltinID) { 239#define GET_NEON_OVERLOAD_CHECK 240#include "clang/Basic/arm_neon.inc" 241#undef GET_NEON_OVERLOAD_CHECK 242 } 243 244 // For NEON intrinsics which are overloaded on vector element type, validate 245 // the immediate which specifies which variant to emit. 246 if (mask) { 247 unsigned ArgNo = TheCall->getNumArgs()-1; 248 if (SemaBuiltinConstantArg(TheCall, ArgNo, Result)) 249 return true; 250 251 TV = Result.getLimitedValue(32); 252 if ((TV > 31) || (mask & (1 << TV)) == 0) 253 return Diag(TheCall->getLocStart(), diag::err_invalid_neon_type_code) 254 << TheCall->getArg(ArgNo)->getSourceRange(); 255 } 256 257 // For NEON intrinsics which take an immediate value as part of the 258 // instruction, range check them here. 259 unsigned i = 0, l = 0, u = 0; 260 switch (BuiltinID) { 261 default: return false; 262 case ARM::BI__builtin_arm_ssat: i = 1; l = 1; u = 31; break; 263 case ARM::BI__builtin_arm_usat: i = 1; u = 31; break; 264 case ARM::BI__builtin_arm_vcvtr_f: 265 case ARM::BI__builtin_arm_vcvtr_d: i = 1; u = 1; break; 266#define GET_NEON_IMMEDIATE_CHECK 267#include "clang/Basic/arm_neon.inc" 268#undef GET_NEON_IMMEDIATE_CHECK 269 }; 270 271 // Check that the immediate argument is actually a constant. 272 if (SemaBuiltinConstantArg(TheCall, i, Result)) 273 return true; 274 275 // Range check against the upper/lower values for this isntruction. 276 unsigned Val = Result.getZExtValue(); 277 if (Val < l || Val > (u + l)) 278 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 279 << l << u+l << TheCall->getArg(i)->getSourceRange(); 280 281 // FIXME: VFP Intrinsics should error if VFP not present. 282 return false; 283} 284 285/// CheckFunctionCall - Check a direct function call for various correctness 286/// and safety properties not strictly enforced by the C type system. 287bool Sema::CheckFunctionCall(FunctionDecl *FDecl, CallExpr *TheCall) { 288 // Get the IdentifierInfo* for the called function. 289 IdentifierInfo *FnInfo = FDecl->getIdentifier(); 290 291 // None of the checks below are needed for functions that don't have 292 // simple names (e.g., C++ conversion functions). 293 if (!FnInfo) 294 return false; 295 296 // FIXME: This mechanism should be abstracted to be less fragile and 297 // more efficient. For example, just map function ids to custom 298 // handlers. 299 300 // Printf and scanf checking. 301 for (specific_attr_iterator<FormatAttr> 302 i = FDecl->specific_attr_begin<FormatAttr>(), 303 e = FDecl->specific_attr_end<FormatAttr>(); i != e ; ++i) { 304 305 const FormatAttr *Format = *i; 306 const bool b = Format->getType() == "scanf"; 307 if (b || CheckablePrintfAttr(Format, TheCall)) { 308 bool HasVAListArg = Format->getFirstArg() == 0; 309 CheckPrintfScanfArguments(TheCall, HasVAListArg, 310 Format->getFormatIdx() - 1, 311 HasVAListArg ? 0 : Format->getFirstArg() - 1, 312 !b); 313 } 314 } 315 316 for (specific_attr_iterator<NonNullAttr> 317 i = FDecl->specific_attr_begin<NonNullAttr>(), 318 e = FDecl->specific_attr_end<NonNullAttr>(); i != e; ++i) { 319 CheckNonNullArguments(*i, TheCall->getArgs(), 320 TheCall->getCallee()->getLocStart()); 321 } 322 323 // Builtin handling 324 int CMF = -1; 325 switch (FDecl->getBuiltinID()) { 326 case Builtin::BI__builtin_memset: 327 case Builtin::BI__builtin___memset_chk: 328 case Builtin::BImemset: 329 CMF = CMF_Memset; 330 break; 331 332 case Builtin::BI__builtin_memcpy: 333 case Builtin::BI__builtin___memcpy_chk: 334 case Builtin::BImemcpy: 335 CMF = CMF_Memcpy; 336 break; 337 338 case Builtin::BI__builtin_memmove: 339 case Builtin::BI__builtin___memmove_chk: 340 case Builtin::BImemmove: 341 CMF = CMF_Memmove; 342 break; 343 344 case Builtin::BIstrlcpy: 345 case Builtin::BIstrlcat: 346 CheckStrlcpycatArguments(TheCall, FnInfo); 347 break; 348 349 case Builtin::BI__builtin_memcmp: 350 CMF = CMF_Memcmp; 351 break; 352 353 default: 354 if (FDecl->getLinkage() == ExternalLinkage && 355 (!getLangOptions().CPlusPlus || FDecl->isExternC())) { 356 if (FnInfo->isStr("memset")) 357 CMF = CMF_Memset; 358 else if (FnInfo->isStr("memcpy")) 359 CMF = CMF_Memcpy; 360 else if (FnInfo->isStr("memmove")) 361 CMF = CMF_Memmove; 362 else if (FnInfo->isStr("memcmp")) 363 CMF = CMF_Memcmp; 364 } 365 break; 366 } 367 368 // Memset/memcpy/memmove handling 369 if (CMF != -1) 370 CheckMemaccessArguments(TheCall, CheckedMemoryFunction(CMF), FnInfo); 371 372 return false; 373} 374 375bool Sema::CheckBlockCall(NamedDecl *NDecl, CallExpr *TheCall) { 376 // Printf checking. 377 const FormatAttr *Format = NDecl->getAttr<FormatAttr>(); 378 if (!Format) 379 return false; 380 381 const VarDecl *V = dyn_cast<VarDecl>(NDecl); 382 if (!V) 383 return false; 384 385 QualType Ty = V->getType(); 386 if (!Ty->isBlockPointerType()) 387 return false; 388 389 const bool b = Format->getType() == "scanf"; 390 if (!b && !CheckablePrintfAttr(Format, TheCall)) 391 return false; 392 393 bool HasVAListArg = Format->getFirstArg() == 0; 394 CheckPrintfScanfArguments(TheCall, HasVAListArg, Format->getFormatIdx() - 1, 395 HasVAListArg ? 0 : Format->getFirstArg() - 1, !b); 396 397 return false; 398} 399 400/// checkBuiltinArgument - Given a call to a builtin function, perform 401/// normal type-checking on the given argument, updating the call in 402/// place. This is useful when a builtin function requires custom 403/// type-checking for some of its arguments but not necessarily all of 404/// them. 405/// 406/// Returns true on error. 407static bool checkBuiltinArgument(Sema &S, CallExpr *E, unsigned ArgIndex) { 408 FunctionDecl *Fn = E->getDirectCallee(); 409 assert(Fn && "builtin call without direct callee!"); 410 411 ParmVarDecl *Param = Fn->getParamDecl(ArgIndex); 412 InitializedEntity Entity = 413 InitializedEntity::InitializeParameter(S.Context, Param); 414 415 ExprResult Arg = E->getArg(0); 416 Arg = S.PerformCopyInitialization(Entity, SourceLocation(), Arg); 417 if (Arg.isInvalid()) 418 return true; 419 420 E->setArg(ArgIndex, Arg.take()); 421 return false; 422} 423 424/// SemaBuiltinAtomicOverloaded - We have a call to a function like 425/// __sync_fetch_and_add, which is an overloaded function based on the pointer 426/// type of its first argument. The main ActOnCallExpr routines have already 427/// promoted the types of arguments because all of these calls are prototyped as 428/// void(...). 429/// 430/// This function goes through and does final semantic checking for these 431/// builtins, 432ExprResult 433Sema::SemaBuiltinAtomicOverloaded(ExprResult TheCallResult) { 434 CallExpr *TheCall = (CallExpr *)TheCallResult.get(); 435 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 436 FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl()); 437 438 // Ensure that we have at least one argument to do type inference from. 439 if (TheCall->getNumArgs() < 1) { 440 Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args_at_least) 441 << 0 << 1 << TheCall->getNumArgs() 442 << TheCall->getCallee()->getSourceRange(); 443 return ExprError(); 444 } 445 446 // Inspect the first argument of the atomic builtin. This should always be 447 // a pointer type, whose element is an integral scalar or pointer type. 448 // Because it is a pointer type, we don't have to worry about any implicit 449 // casts here. 450 // FIXME: We don't allow floating point scalars as input. 451 Expr *FirstArg = TheCall->getArg(0); 452 const PointerType *pointerType = FirstArg->getType()->getAs<PointerType>(); 453 if (!pointerType) { 454 Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer) 455 << FirstArg->getType() << FirstArg->getSourceRange(); 456 return ExprError(); 457 } 458 459 QualType ValType = pointerType->getPointeeType(); 460 if (!ValType->isIntegerType() && !ValType->isAnyPointerType() && 461 !ValType->isBlockPointerType()) { 462 Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer_intptr) 463 << FirstArg->getType() << FirstArg->getSourceRange(); 464 return ExprError(); 465 } 466 467 switch (ValType.getObjCLifetime()) { 468 case Qualifiers::OCL_None: 469 case Qualifiers::OCL_ExplicitNone: 470 // okay 471 break; 472 473 case Qualifiers::OCL_Weak: 474 case Qualifiers::OCL_Strong: 475 case Qualifiers::OCL_Autoreleasing: 476 Diag(DRE->getLocStart(), diag::err_arc_atomic_ownership) 477 << ValType << FirstArg->getSourceRange(); 478 return ExprError(); 479 } 480 481 // The majority of builtins return a value, but a few have special return 482 // types, so allow them to override appropriately below. 483 QualType ResultType = ValType; 484 485 // We need to figure out which concrete builtin this maps onto. For example, 486 // __sync_fetch_and_add with a 2 byte object turns into 487 // __sync_fetch_and_add_2. 488#define BUILTIN_ROW(x) \ 489 { Builtin::BI##x##_1, Builtin::BI##x##_2, Builtin::BI##x##_4, \ 490 Builtin::BI##x##_8, Builtin::BI##x##_16 } 491 492 static const unsigned BuiltinIndices[][5] = { 493 BUILTIN_ROW(__sync_fetch_and_add), 494 BUILTIN_ROW(__sync_fetch_and_sub), 495 BUILTIN_ROW(__sync_fetch_and_or), 496 BUILTIN_ROW(__sync_fetch_and_and), 497 BUILTIN_ROW(__sync_fetch_and_xor), 498 499 BUILTIN_ROW(__sync_add_and_fetch), 500 BUILTIN_ROW(__sync_sub_and_fetch), 501 BUILTIN_ROW(__sync_and_and_fetch), 502 BUILTIN_ROW(__sync_or_and_fetch), 503 BUILTIN_ROW(__sync_xor_and_fetch), 504 505 BUILTIN_ROW(__sync_val_compare_and_swap), 506 BUILTIN_ROW(__sync_bool_compare_and_swap), 507 BUILTIN_ROW(__sync_lock_test_and_set), 508 BUILTIN_ROW(__sync_lock_release), 509 BUILTIN_ROW(__sync_swap) 510 }; 511#undef BUILTIN_ROW 512 513 // Determine the index of the size. 514 unsigned SizeIndex; 515 switch (Context.getTypeSizeInChars(ValType).getQuantity()) { 516 case 1: SizeIndex = 0; break; 517 case 2: SizeIndex = 1; break; 518 case 4: SizeIndex = 2; break; 519 case 8: SizeIndex = 3; break; 520 case 16: SizeIndex = 4; break; 521 default: 522 Diag(DRE->getLocStart(), diag::err_atomic_builtin_pointer_size) 523 << FirstArg->getType() << FirstArg->getSourceRange(); 524 return ExprError(); 525 } 526 527 // Each of these builtins has one pointer argument, followed by some number of 528 // values (0, 1 or 2) followed by a potentially empty varags list of stuff 529 // that we ignore. Find out which row of BuiltinIndices to read from as well 530 // as the number of fixed args. 531 unsigned BuiltinID = FDecl->getBuiltinID(); 532 unsigned BuiltinIndex, NumFixed = 1; 533 switch (BuiltinID) { 534 default: assert(0 && "Unknown overloaded atomic builtin!"); 535 case Builtin::BI__sync_fetch_and_add: BuiltinIndex = 0; break; 536 case Builtin::BI__sync_fetch_and_sub: BuiltinIndex = 1; break; 537 case Builtin::BI__sync_fetch_and_or: BuiltinIndex = 2; break; 538 case Builtin::BI__sync_fetch_and_and: BuiltinIndex = 3; break; 539 case Builtin::BI__sync_fetch_and_xor: BuiltinIndex = 4; break; 540 541 case Builtin::BI__sync_add_and_fetch: BuiltinIndex = 5; break; 542 case Builtin::BI__sync_sub_and_fetch: BuiltinIndex = 6; break; 543 case Builtin::BI__sync_and_and_fetch: BuiltinIndex = 7; break; 544 case Builtin::BI__sync_or_and_fetch: BuiltinIndex = 8; break; 545 case Builtin::BI__sync_xor_and_fetch: BuiltinIndex = 9; break; 546 547 case Builtin::BI__sync_val_compare_and_swap: 548 BuiltinIndex = 10; 549 NumFixed = 2; 550 break; 551 case Builtin::BI__sync_bool_compare_and_swap: 552 BuiltinIndex = 11; 553 NumFixed = 2; 554 ResultType = Context.BoolTy; 555 break; 556 case Builtin::BI__sync_lock_test_and_set: BuiltinIndex = 12; break; 557 case Builtin::BI__sync_lock_release: 558 BuiltinIndex = 13; 559 NumFixed = 0; 560 ResultType = Context.VoidTy; 561 break; 562 case Builtin::BI__sync_swap: BuiltinIndex = 14; break; 563 } 564 565 // Now that we know how many fixed arguments we expect, first check that we 566 // have at least that many. 567 if (TheCall->getNumArgs() < 1+NumFixed) { 568 Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args_at_least) 569 << 0 << 1+NumFixed << TheCall->getNumArgs() 570 << TheCall->getCallee()->getSourceRange(); 571 return ExprError(); 572 } 573 574 // Get the decl for the concrete builtin from this, we can tell what the 575 // concrete integer type we should convert to is. 576 unsigned NewBuiltinID = BuiltinIndices[BuiltinIndex][SizeIndex]; 577 const char *NewBuiltinName = Context.BuiltinInfo.GetName(NewBuiltinID); 578 IdentifierInfo *NewBuiltinII = PP.getIdentifierInfo(NewBuiltinName); 579 FunctionDecl *NewBuiltinDecl = 580 cast<FunctionDecl>(LazilyCreateBuiltin(NewBuiltinII, NewBuiltinID, 581 TUScope, false, DRE->getLocStart())); 582 583 // The first argument --- the pointer --- has a fixed type; we 584 // deduce the types of the rest of the arguments accordingly. Walk 585 // the remaining arguments, converting them to the deduced value type. 586 for (unsigned i = 0; i != NumFixed; ++i) { 587 ExprResult Arg = TheCall->getArg(i+1); 588 589 // If the argument is an implicit cast, then there was a promotion due to 590 // "...", just remove it now. 591 if (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(Arg.get())) { 592 Arg = ICE->getSubExpr(); 593 ICE->setSubExpr(0); 594 TheCall->setArg(i+1, Arg.get()); 595 } 596 597 // GCC does an implicit conversion to the pointer or integer ValType. This 598 // can fail in some cases (1i -> int**), check for this error case now. 599 CastKind Kind = CK_Invalid; 600 ExprValueKind VK = VK_RValue; 601 CXXCastPath BasePath; 602 Arg = CheckCastTypes(Arg.get()->getLocStart(), Arg.get()->getSourceRange(), 603 ValType, Arg.take(), Kind, VK, BasePath); 604 if (Arg.isInvalid()) 605 return ExprError(); 606 607 // Okay, we have something that *can* be converted to the right type. Check 608 // to see if there is a potentially weird extension going on here. This can 609 // happen when you do an atomic operation on something like an char* and 610 // pass in 42. The 42 gets converted to char. This is even more strange 611 // for things like 45.123 -> char, etc. 612 // FIXME: Do this check. 613 Arg = ImpCastExprToType(Arg.take(), ValType, Kind, VK, &BasePath); 614 TheCall->setArg(i+1, Arg.get()); 615 } 616 617 ASTContext& Context = this->getASTContext(); 618 619 // Create a new DeclRefExpr to refer to the new decl. 620 DeclRefExpr* NewDRE = DeclRefExpr::Create( 621 Context, 622 DRE->getQualifierLoc(), 623 NewBuiltinDecl, 624 DRE->getLocation(), 625 NewBuiltinDecl->getType(), 626 DRE->getValueKind()); 627 628 // Set the callee in the CallExpr. 629 // FIXME: This leaks the original parens and implicit casts. 630 ExprResult PromotedCall = UsualUnaryConversions(NewDRE); 631 if (PromotedCall.isInvalid()) 632 return ExprError(); 633 TheCall->setCallee(PromotedCall.take()); 634 635 // Change the result type of the call to match the original value type. This 636 // is arbitrary, but the codegen for these builtins ins design to handle it 637 // gracefully. 638 TheCall->setType(ResultType); 639 640 return move(TheCallResult); 641} 642 643 644/// CheckObjCString - Checks that the argument to the builtin 645/// CFString constructor is correct 646/// Note: It might also make sense to do the UTF-16 conversion here (would 647/// simplify the backend). 648bool Sema::CheckObjCString(Expr *Arg) { 649 Arg = Arg->IgnoreParenCasts(); 650 StringLiteral *Literal = dyn_cast<StringLiteral>(Arg); 651 652 if (!Literal || !Literal->isAscii()) { 653 Diag(Arg->getLocStart(), diag::err_cfstring_literal_not_string_constant) 654 << Arg->getSourceRange(); 655 return true; 656 } 657 658 if (Literal->containsNonAsciiOrNull()) { 659 StringRef String = Literal->getString(); 660 unsigned NumBytes = String.size(); 661 SmallVector<UTF16, 128> ToBuf(NumBytes); 662 const UTF8 *FromPtr = (UTF8 *)String.data(); 663 UTF16 *ToPtr = &ToBuf[0]; 664 665 ConversionResult Result = ConvertUTF8toUTF16(&FromPtr, FromPtr + NumBytes, 666 &ToPtr, ToPtr + NumBytes, 667 strictConversion); 668 // Check for conversion failure. 669 if (Result != conversionOK) 670 Diag(Arg->getLocStart(), 671 diag::warn_cfstring_truncated) << Arg->getSourceRange(); 672 } 673 return false; 674} 675 676/// SemaBuiltinVAStart - Check the arguments to __builtin_va_start for validity. 677/// Emit an error and return true on failure, return false on success. 678bool Sema::SemaBuiltinVAStart(CallExpr *TheCall) { 679 Expr *Fn = TheCall->getCallee(); 680 if (TheCall->getNumArgs() > 2) { 681 Diag(TheCall->getArg(2)->getLocStart(), 682 diag::err_typecheck_call_too_many_args) 683 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 684 << Fn->getSourceRange() 685 << SourceRange(TheCall->getArg(2)->getLocStart(), 686 (*(TheCall->arg_end()-1))->getLocEnd()); 687 return true; 688 } 689 690 if (TheCall->getNumArgs() < 2) { 691 return Diag(TheCall->getLocEnd(), 692 diag::err_typecheck_call_too_few_args_at_least) 693 << 0 /*function call*/ << 2 << TheCall->getNumArgs(); 694 } 695 696 // Type-check the first argument normally. 697 if (checkBuiltinArgument(*this, TheCall, 0)) 698 return true; 699 700 // Determine whether the current function is variadic or not. 701 BlockScopeInfo *CurBlock = getCurBlock(); 702 bool isVariadic; 703 if (CurBlock) 704 isVariadic = CurBlock->TheDecl->isVariadic(); 705 else if (FunctionDecl *FD = getCurFunctionDecl()) 706 isVariadic = FD->isVariadic(); 707 else 708 isVariadic = getCurMethodDecl()->isVariadic(); 709 710 if (!isVariadic) { 711 Diag(Fn->getLocStart(), diag::err_va_start_used_in_non_variadic_function); 712 return true; 713 } 714 715 // Verify that the second argument to the builtin is the last argument of the 716 // current function or method. 717 bool SecondArgIsLastNamedArgument = false; 718 const Expr *Arg = TheCall->getArg(1)->IgnoreParenCasts(); 719 720 if (const DeclRefExpr *DR = dyn_cast<DeclRefExpr>(Arg)) { 721 if (const ParmVarDecl *PV = dyn_cast<ParmVarDecl>(DR->getDecl())) { 722 // FIXME: This isn't correct for methods (results in bogus warning). 723 // Get the last formal in the current function. 724 const ParmVarDecl *LastArg; 725 if (CurBlock) 726 LastArg = *(CurBlock->TheDecl->param_end()-1); 727 else if (FunctionDecl *FD = getCurFunctionDecl()) 728 LastArg = *(FD->param_end()-1); 729 else 730 LastArg = *(getCurMethodDecl()->param_end()-1); 731 SecondArgIsLastNamedArgument = PV == LastArg; 732 } 733 } 734 735 if (!SecondArgIsLastNamedArgument) 736 Diag(TheCall->getArg(1)->getLocStart(), 737 diag::warn_second_parameter_of_va_start_not_last_named_argument); 738 return false; 739} 740 741/// SemaBuiltinUnorderedCompare - Handle functions like __builtin_isgreater and 742/// friends. This is declared to take (...), so we have to check everything. 743bool Sema::SemaBuiltinUnorderedCompare(CallExpr *TheCall) { 744 if (TheCall->getNumArgs() < 2) 745 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 746 << 0 << 2 << TheCall->getNumArgs()/*function call*/; 747 if (TheCall->getNumArgs() > 2) 748 return Diag(TheCall->getArg(2)->getLocStart(), 749 diag::err_typecheck_call_too_many_args) 750 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 751 << SourceRange(TheCall->getArg(2)->getLocStart(), 752 (*(TheCall->arg_end()-1))->getLocEnd()); 753 754 ExprResult OrigArg0 = TheCall->getArg(0); 755 ExprResult OrigArg1 = TheCall->getArg(1); 756 757 // Do standard promotions between the two arguments, returning their common 758 // type. 759 QualType Res = UsualArithmeticConversions(OrigArg0, OrigArg1, false); 760 if (OrigArg0.isInvalid() || OrigArg1.isInvalid()) 761 return true; 762 763 // Make sure any conversions are pushed back into the call; this is 764 // type safe since unordered compare builtins are declared as "_Bool 765 // foo(...)". 766 TheCall->setArg(0, OrigArg0.get()); 767 TheCall->setArg(1, OrigArg1.get()); 768 769 if (OrigArg0.get()->isTypeDependent() || OrigArg1.get()->isTypeDependent()) 770 return false; 771 772 // If the common type isn't a real floating type, then the arguments were 773 // invalid for this operation. 774 if (!Res->isRealFloatingType()) 775 return Diag(OrigArg0.get()->getLocStart(), 776 diag::err_typecheck_call_invalid_ordered_compare) 777 << OrigArg0.get()->getType() << OrigArg1.get()->getType() 778 << SourceRange(OrigArg0.get()->getLocStart(), OrigArg1.get()->getLocEnd()); 779 780 return false; 781} 782 783/// SemaBuiltinSemaBuiltinFPClassification - Handle functions like 784/// __builtin_isnan and friends. This is declared to take (...), so we have 785/// to check everything. We expect the last argument to be a floating point 786/// value. 787bool Sema::SemaBuiltinFPClassification(CallExpr *TheCall, unsigned NumArgs) { 788 if (TheCall->getNumArgs() < NumArgs) 789 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 790 << 0 << NumArgs << TheCall->getNumArgs()/*function call*/; 791 if (TheCall->getNumArgs() > NumArgs) 792 return Diag(TheCall->getArg(NumArgs)->getLocStart(), 793 diag::err_typecheck_call_too_many_args) 794 << 0 /*function call*/ << NumArgs << TheCall->getNumArgs() 795 << SourceRange(TheCall->getArg(NumArgs)->getLocStart(), 796 (*(TheCall->arg_end()-1))->getLocEnd()); 797 798 Expr *OrigArg = TheCall->getArg(NumArgs-1); 799 800 if (OrigArg->isTypeDependent()) 801 return false; 802 803 // This operation requires a non-_Complex floating-point number. 804 if (!OrigArg->getType()->isRealFloatingType()) 805 return Diag(OrigArg->getLocStart(), 806 diag::err_typecheck_call_invalid_unary_fp) 807 << OrigArg->getType() << OrigArg->getSourceRange(); 808 809 // If this is an implicit conversion from float -> double, remove it. 810 if (ImplicitCastExpr *Cast = dyn_cast<ImplicitCastExpr>(OrigArg)) { 811 Expr *CastArg = Cast->getSubExpr(); 812 if (CastArg->getType()->isSpecificBuiltinType(BuiltinType::Float)) { 813 assert(Cast->getType()->isSpecificBuiltinType(BuiltinType::Double) && 814 "promotion from float to double is the only expected cast here"); 815 Cast->setSubExpr(0); 816 TheCall->setArg(NumArgs-1, CastArg); 817 OrigArg = CastArg; 818 } 819 } 820 821 return false; 822} 823 824/// SemaBuiltinShuffleVector - Handle __builtin_shufflevector. 825// This is declared to take (...), so we have to check everything. 826ExprResult Sema::SemaBuiltinShuffleVector(CallExpr *TheCall) { 827 if (TheCall->getNumArgs() < 2) 828 return ExprError(Diag(TheCall->getLocEnd(), 829 diag::err_typecheck_call_too_few_args_at_least) 830 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 831 << TheCall->getSourceRange()); 832 833 // Determine which of the following types of shufflevector we're checking: 834 // 1) unary, vector mask: (lhs, mask) 835 // 2) binary, vector mask: (lhs, rhs, mask) 836 // 3) binary, scalar mask: (lhs, rhs, index, ..., index) 837 QualType resType = TheCall->getArg(0)->getType(); 838 unsigned numElements = 0; 839 840 if (!TheCall->getArg(0)->isTypeDependent() && 841 !TheCall->getArg(1)->isTypeDependent()) { 842 QualType LHSType = TheCall->getArg(0)->getType(); 843 QualType RHSType = TheCall->getArg(1)->getType(); 844 845 if (!LHSType->isVectorType() || !RHSType->isVectorType()) { 846 Diag(TheCall->getLocStart(), diag::err_shufflevector_non_vector) 847 << SourceRange(TheCall->getArg(0)->getLocStart(), 848 TheCall->getArg(1)->getLocEnd()); 849 return ExprError(); 850 } 851 852 numElements = LHSType->getAs<VectorType>()->getNumElements(); 853 unsigned numResElements = TheCall->getNumArgs() - 2; 854 855 // Check to see if we have a call with 2 vector arguments, the unary shuffle 856 // with mask. If so, verify that RHS is an integer vector type with the 857 // same number of elts as lhs. 858 if (TheCall->getNumArgs() == 2) { 859 if (!RHSType->hasIntegerRepresentation() || 860 RHSType->getAs<VectorType>()->getNumElements() != numElements) 861 Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector) 862 << SourceRange(TheCall->getArg(1)->getLocStart(), 863 TheCall->getArg(1)->getLocEnd()); 864 numResElements = numElements; 865 } 866 else if (!Context.hasSameUnqualifiedType(LHSType, RHSType)) { 867 Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector) 868 << SourceRange(TheCall->getArg(0)->getLocStart(), 869 TheCall->getArg(1)->getLocEnd()); 870 return ExprError(); 871 } else if (numElements != numResElements) { 872 QualType eltType = LHSType->getAs<VectorType>()->getElementType(); 873 resType = Context.getVectorType(eltType, numResElements, 874 VectorType::GenericVector); 875 } 876 } 877 878 for (unsigned i = 2; i < TheCall->getNumArgs(); i++) { 879 if (TheCall->getArg(i)->isTypeDependent() || 880 TheCall->getArg(i)->isValueDependent()) 881 continue; 882 883 llvm::APSInt Result(32); 884 if (!TheCall->getArg(i)->isIntegerConstantExpr(Result, Context)) 885 return ExprError(Diag(TheCall->getLocStart(), 886 diag::err_shufflevector_nonconstant_argument) 887 << TheCall->getArg(i)->getSourceRange()); 888 889 if (Result.getActiveBits() > 64 || Result.getZExtValue() >= numElements*2) 890 return ExprError(Diag(TheCall->getLocStart(), 891 diag::err_shufflevector_argument_too_large) 892 << TheCall->getArg(i)->getSourceRange()); 893 } 894 895 SmallVector<Expr*, 32> exprs; 896 897 for (unsigned i = 0, e = TheCall->getNumArgs(); i != e; i++) { 898 exprs.push_back(TheCall->getArg(i)); 899 TheCall->setArg(i, 0); 900 } 901 902 return Owned(new (Context) ShuffleVectorExpr(Context, exprs.begin(), 903 exprs.size(), resType, 904 TheCall->getCallee()->getLocStart(), 905 TheCall->getRParenLoc())); 906} 907 908/// SemaBuiltinPrefetch - Handle __builtin_prefetch. 909// This is declared to take (const void*, ...) and can take two 910// optional constant int args. 911bool Sema::SemaBuiltinPrefetch(CallExpr *TheCall) { 912 unsigned NumArgs = TheCall->getNumArgs(); 913 914 if (NumArgs > 3) 915 return Diag(TheCall->getLocEnd(), 916 diag::err_typecheck_call_too_many_args_at_most) 917 << 0 /*function call*/ << 3 << NumArgs 918 << TheCall->getSourceRange(); 919 920 // Argument 0 is checked for us and the remaining arguments must be 921 // constant integers. 922 for (unsigned i = 1; i != NumArgs; ++i) { 923 Expr *Arg = TheCall->getArg(i); 924 925 llvm::APSInt Result; 926 if (SemaBuiltinConstantArg(TheCall, i, Result)) 927 return true; 928 929 // FIXME: gcc issues a warning and rewrites these to 0. These 930 // seems especially odd for the third argument since the default 931 // is 3. 932 if (i == 1) { 933 if (Result.getLimitedValue() > 1) 934 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 935 << "0" << "1" << Arg->getSourceRange(); 936 } else { 937 if (Result.getLimitedValue() > 3) 938 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 939 << "0" << "3" << Arg->getSourceRange(); 940 } 941 } 942 943 return false; 944} 945 946/// SemaBuiltinConstantArg - Handle a check if argument ArgNum of CallExpr 947/// TheCall is a constant expression. 948bool Sema::SemaBuiltinConstantArg(CallExpr *TheCall, int ArgNum, 949 llvm::APSInt &Result) { 950 Expr *Arg = TheCall->getArg(ArgNum); 951 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 952 FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl()); 953 954 if (Arg->isTypeDependent() || Arg->isValueDependent()) return false; 955 956 if (!Arg->isIntegerConstantExpr(Result, Context)) 957 return Diag(TheCall->getLocStart(), diag::err_constant_integer_arg_type) 958 << FDecl->getDeclName() << Arg->getSourceRange(); 959 960 return false; 961} 962 963/// SemaBuiltinObjectSize - Handle __builtin_object_size(void *ptr, 964/// int type). This simply type checks that type is one of the defined 965/// constants (0-3). 966// For compatibility check 0-3, llvm only handles 0 and 2. 967bool Sema::SemaBuiltinObjectSize(CallExpr *TheCall) { 968 llvm::APSInt Result; 969 970 // Check constant-ness first. 971 if (SemaBuiltinConstantArg(TheCall, 1, Result)) 972 return true; 973 974 Expr *Arg = TheCall->getArg(1); 975 if (Result.getSExtValue() < 0 || Result.getSExtValue() > 3) { 976 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 977 << "0" << "3" << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 978 } 979 980 return false; 981} 982 983/// SemaBuiltinLongjmp - Handle __builtin_longjmp(void *env[5], int val). 984/// This checks that val is a constant 1. 985bool Sema::SemaBuiltinLongjmp(CallExpr *TheCall) { 986 Expr *Arg = TheCall->getArg(1); 987 llvm::APSInt Result; 988 989 // TODO: This is less than ideal. Overload this to take a value. 990 if (SemaBuiltinConstantArg(TheCall, 1, Result)) 991 return true; 992 993 if (Result != 1) 994 return Diag(TheCall->getLocStart(), diag::err_builtin_longjmp_invalid_val) 995 << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 996 997 return false; 998} 999 1000// Handle i > 1 ? "x" : "y", recursively. 1001bool Sema::SemaCheckStringLiteral(const Expr *E, const CallExpr *TheCall, 1002 bool HasVAListArg, 1003 unsigned format_idx, unsigned firstDataArg, 1004 bool isPrintf) { 1005 tryAgain: 1006 if (E->isTypeDependent() || E->isValueDependent()) 1007 return false; 1008 1009 E = E->IgnoreParens(); 1010 1011 switch (E->getStmtClass()) { 1012 case Stmt::BinaryConditionalOperatorClass: 1013 case Stmt::ConditionalOperatorClass: { 1014 const AbstractConditionalOperator *C = cast<AbstractConditionalOperator>(E); 1015 return SemaCheckStringLiteral(C->getTrueExpr(), TheCall, HasVAListArg, 1016 format_idx, firstDataArg, isPrintf) 1017 && SemaCheckStringLiteral(C->getFalseExpr(), TheCall, HasVAListArg, 1018 format_idx, firstDataArg, isPrintf); 1019 } 1020 1021 case Stmt::IntegerLiteralClass: 1022 // Technically -Wformat-nonliteral does not warn about this case. 1023 // The behavior of printf and friends in this case is implementation 1024 // dependent. Ideally if the format string cannot be null then 1025 // it should have a 'nonnull' attribute in the function prototype. 1026 return true; 1027 1028 case Stmt::ImplicitCastExprClass: { 1029 E = cast<ImplicitCastExpr>(E)->getSubExpr(); 1030 goto tryAgain; 1031 } 1032 1033 case Stmt::OpaqueValueExprClass: 1034 if (const Expr *src = cast<OpaqueValueExpr>(E)->getSourceExpr()) { 1035 E = src; 1036 goto tryAgain; 1037 } 1038 return false; 1039 1040 case Stmt::PredefinedExprClass: 1041 // While __func__, etc., are technically not string literals, they 1042 // cannot contain format specifiers and thus are not a security 1043 // liability. 1044 return true; 1045 1046 case Stmt::DeclRefExprClass: { 1047 const DeclRefExpr *DR = cast<DeclRefExpr>(E); 1048 1049 // As an exception, do not flag errors for variables binding to 1050 // const string literals. 1051 if (const VarDecl *VD = dyn_cast<VarDecl>(DR->getDecl())) { 1052 bool isConstant = false; 1053 QualType T = DR->getType(); 1054 1055 if (const ArrayType *AT = Context.getAsArrayType(T)) { 1056 isConstant = AT->getElementType().isConstant(Context); 1057 } else if (const PointerType *PT = T->getAs<PointerType>()) { 1058 isConstant = T.isConstant(Context) && 1059 PT->getPointeeType().isConstant(Context); 1060 } 1061 1062 if (isConstant) { 1063 if (const Expr *Init = VD->getAnyInitializer()) 1064 return SemaCheckStringLiteral(Init, TheCall, 1065 HasVAListArg, format_idx, firstDataArg, 1066 isPrintf); 1067 } 1068 1069 // For vprintf* functions (i.e., HasVAListArg==true), we add a 1070 // special check to see if the format string is a function parameter 1071 // of the function calling the printf function. If the function 1072 // has an attribute indicating it is a printf-like function, then we 1073 // should suppress warnings concerning non-literals being used in a call 1074 // to a vprintf function. For example: 1075 // 1076 // void 1077 // logmessage(char const *fmt __attribute__ (format (printf, 1, 2)), ...){ 1078 // va_list ap; 1079 // va_start(ap, fmt); 1080 // vprintf(fmt, ap); // Do NOT emit a warning about "fmt". 1081 // ... 1082 // 1083 // 1084 // FIXME: We don't have full attribute support yet, so just check to see 1085 // if the argument is a DeclRefExpr that references a parameter. We'll 1086 // add proper support for checking the attribute later. 1087 if (HasVAListArg) 1088 if (isa<ParmVarDecl>(VD)) 1089 return true; 1090 } 1091 1092 return false; 1093 } 1094 1095 case Stmt::CallExprClass: { 1096 const CallExpr *CE = cast<CallExpr>(E); 1097 if (const ImplicitCastExpr *ICE 1098 = dyn_cast<ImplicitCastExpr>(CE->getCallee())) { 1099 if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(ICE->getSubExpr())) { 1100 if (const FunctionDecl *FD = dyn_cast<FunctionDecl>(DRE->getDecl())) { 1101 if (const FormatArgAttr *FA = FD->getAttr<FormatArgAttr>()) { 1102 unsigned ArgIndex = FA->getFormatIdx(); 1103 const Expr *Arg = CE->getArg(ArgIndex - 1); 1104 1105 return SemaCheckStringLiteral(Arg, TheCall, HasVAListArg, 1106 format_idx, firstDataArg, isPrintf); 1107 } 1108 } 1109 } 1110 } 1111 1112 return false; 1113 } 1114 case Stmt::ObjCStringLiteralClass: 1115 case Stmt::StringLiteralClass: { 1116 const StringLiteral *StrE = NULL; 1117 1118 if (const ObjCStringLiteral *ObjCFExpr = dyn_cast<ObjCStringLiteral>(E)) 1119 StrE = ObjCFExpr->getString(); 1120 else 1121 StrE = cast<StringLiteral>(E); 1122 1123 if (StrE) { 1124 CheckFormatString(StrE, E, TheCall, HasVAListArg, format_idx, 1125 firstDataArg, isPrintf); 1126 return true; 1127 } 1128 1129 return false; 1130 } 1131 1132 default: 1133 return false; 1134 } 1135} 1136 1137void 1138Sema::CheckNonNullArguments(const NonNullAttr *NonNull, 1139 const Expr * const *ExprArgs, 1140 SourceLocation CallSiteLoc) { 1141 for (NonNullAttr::args_iterator i = NonNull->args_begin(), 1142 e = NonNull->args_end(); 1143 i != e; ++i) { 1144 const Expr *ArgExpr = ExprArgs[*i]; 1145 if (ArgExpr->isNullPointerConstant(Context, 1146 Expr::NPC_ValueDependentIsNotNull)) 1147 Diag(CallSiteLoc, diag::warn_null_arg) << ArgExpr->getSourceRange(); 1148 } 1149} 1150 1151/// CheckPrintfScanfArguments - Check calls to printf and scanf (and similar 1152/// functions) for correct use of format strings. 1153void 1154Sema::CheckPrintfScanfArguments(const CallExpr *TheCall, bool HasVAListArg, 1155 unsigned format_idx, unsigned firstDataArg, 1156 bool isPrintf) { 1157 1158 const Expr *Fn = TheCall->getCallee(); 1159 1160 // The way the format attribute works in GCC, the implicit this argument 1161 // of member functions is counted. However, it doesn't appear in our own 1162 // lists, so decrement format_idx in that case. 1163 if (isa<CXXMemberCallExpr>(TheCall)) { 1164 const CXXMethodDecl *method_decl = 1165 dyn_cast<CXXMethodDecl>(TheCall->getCalleeDecl()); 1166 if (method_decl && method_decl->isInstance()) { 1167 // Catch a format attribute mistakenly referring to the object argument. 1168 if (format_idx == 0) 1169 return; 1170 --format_idx; 1171 if(firstDataArg != 0) 1172 --firstDataArg; 1173 } 1174 } 1175 1176 // CHECK: printf/scanf-like function is called with no format string. 1177 if (format_idx >= TheCall->getNumArgs()) { 1178 Diag(TheCall->getRParenLoc(), diag::warn_missing_format_string) 1179 << Fn->getSourceRange(); 1180 return; 1181 } 1182 1183 const Expr *OrigFormatExpr = TheCall->getArg(format_idx)->IgnoreParenCasts(); 1184 1185 // CHECK: format string is not a string literal. 1186 // 1187 // Dynamically generated format strings are difficult to 1188 // automatically vet at compile time. Requiring that format strings 1189 // are string literals: (1) permits the checking of format strings by 1190 // the compiler and thereby (2) can practically remove the source of 1191 // many format string exploits. 1192 1193 // Format string can be either ObjC string (e.g. @"%d") or 1194 // C string (e.g. "%d") 1195 // ObjC string uses the same format specifiers as C string, so we can use 1196 // the same format string checking logic for both ObjC and C strings. 1197 if (SemaCheckStringLiteral(OrigFormatExpr, TheCall, HasVAListArg, format_idx, 1198 firstDataArg, isPrintf)) 1199 return; // Literal format string found, check done! 1200 1201 // If there are no arguments specified, warn with -Wformat-security, otherwise 1202 // warn only with -Wformat-nonliteral. 1203 if (TheCall->getNumArgs() == format_idx+1) 1204 Diag(TheCall->getArg(format_idx)->getLocStart(), 1205 diag::warn_format_nonliteral_noargs) 1206 << OrigFormatExpr->getSourceRange(); 1207 else 1208 Diag(TheCall->getArg(format_idx)->getLocStart(), 1209 diag::warn_format_nonliteral) 1210 << OrigFormatExpr->getSourceRange(); 1211} 1212 1213namespace { 1214class CheckFormatHandler : public analyze_format_string::FormatStringHandler { 1215protected: 1216 Sema &S; 1217 const StringLiteral *FExpr; 1218 const Expr *OrigFormatExpr; 1219 const unsigned FirstDataArg; 1220 const unsigned NumDataArgs; 1221 const bool IsObjCLiteral; 1222 const char *Beg; // Start of format string. 1223 const bool HasVAListArg; 1224 const CallExpr *TheCall; 1225 unsigned FormatIdx; 1226 llvm::BitVector CoveredArgs; 1227 bool usesPositionalArgs; 1228 bool atFirstArg; 1229public: 1230 CheckFormatHandler(Sema &s, const StringLiteral *fexpr, 1231 const Expr *origFormatExpr, unsigned firstDataArg, 1232 unsigned numDataArgs, bool isObjCLiteral, 1233 const char *beg, bool hasVAListArg, 1234 const CallExpr *theCall, unsigned formatIdx) 1235 : S(s), FExpr(fexpr), OrigFormatExpr(origFormatExpr), 1236 FirstDataArg(firstDataArg), 1237 NumDataArgs(numDataArgs), 1238 IsObjCLiteral(isObjCLiteral), Beg(beg), 1239 HasVAListArg(hasVAListArg), 1240 TheCall(theCall), FormatIdx(formatIdx), 1241 usesPositionalArgs(false), atFirstArg(true) { 1242 CoveredArgs.resize(numDataArgs); 1243 CoveredArgs.reset(); 1244 } 1245 1246 void DoneProcessing(); 1247 1248 void HandleIncompleteSpecifier(const char *startSpecifier, 1249 unsigned specifierLen); 1250 1251 virtual void HandleInvalidPosition(const char *startSpecifier, 1252 unsigned specifierLen, 1253 analyze_format_string::PositionContext p); 1254 1255 virtual void HandleZeroPosition(const char *startPos, unsigned posLen); 1256 1257 void HandleNullChar(const char *nullCharacter); 1258 1259protected: 1260 bool HandleInvalidConversionSpecifier(unsigned argIndex, SourceLocation Loc, 1261 const char *startSpec, 1262 unsigned specifierLen, 1263 const char *csStart, unsigned csLen); 1264 1265 SourceRange getFormatStringRange(); 1266 CharSourceRange getSpecifierRange(const char *startSpecifier, 1267 unsigned specifierLen); 1268 SourceLocation getLocationOfByte(const char *x); 1269 1270 const Expr *getDataArg(unsigned i) const; 1271 1272 bool CheckNumArgs(const analyze_format_string::FormatSpecifier &FS, 1273 const analyze_format_string::ConversionSpecifier &CS, 1274 const char *startSpecifier, unsigned specifierLen, 1275 unsigned argIndex); 1276}; 1277} 1278 1279SourceRange CheckFormatHandler::getFormatStringRange() { 1280 return OrigFormatExpr->getSourceRange(); 1281} 1282 1283CharSourceRange CheckFormatHandler:: 1284getSpecifierRange(const char *startSpecifier, unsigned specifierLen) { 1285 SourceLocation Start = getLocationOfByte(startSpecifier); 1286 SourceLocation End = getLocationOfByte(startSpecifier + specifierLen - 1); 1287 1288 // Advance the end SourceLocation by one due to half-open ranges. 1289 End = End.getFileLocWithOffset(1); 1290 1291 return CharSourceRange::getCharRange(Start, End); 1292} 1293 1294SourceLocation CheckFormatHandler::getLocationOfByte(const char *x) { 1295 return S.getLocationOfStringLiteralByte(FExpr, x - Beg); 1296} 1297 1298void CheckFormatHandler::HandleIncompleteSpecifier(const char *startSpecifier, 1299 unsigned specifierLen){ 1300 SourceLocation Loc = getLocationOfByte(startSpecifier); 1301 S.Diag(Loc, diag::warn_printf_incomplete_specifier) 1302 << getSpecifierRange(startSpecifier, specifierLen); 1303} 1304 1305void 1306CheckFormatHandler::HandleInvalidPosition(const char *startPos, unsigned posLen, 1307 analyze_format_string::PositionContext p) { 1308 SourceLocation Loc = getLocationOfByte(startPos); 1309 S.Diag(Loc, diag::warn_format_invalid_positional_specifier) 1310 << (unsigned) p << getSpecifierRange(startPos, posLen); 1311} 1312 1313void CheckFormatHandler::HandleZeroPosition(const char *startPos, 1314 unsigned posLen) { 1315 SourceLocation Loc = getLocationOfByte(startPos); 1316 S.Diag(Loc, diag::warn_format_zero_positional_specifier) 1317 << getSpecifierRange(startPos, posLen); 1318} 1319 1320void CheckFormatHandler::HandleNullChar(const char *nullCharacter) { 1321 if (!IsObjCLiteral) { 1322 // The presence of a null character is likely an error. 1323 S.Diag(getLocationOfByte(nullCharacter), 1324 diag::warn_printf_format_string_contains_null_char) 1325 << getFormatStringRange(); 1326 } 1327} 1328 1329const Expr *CheckFormatHandler::getDataArg(unsigned i) const { 1330 return TheCall->getArg(FirstDataArg + i); 1331} 1332 1333void CheckFormatHandler::DoneProcessing() { 1334 // Does the number of data arguments exceed the number of 1335 // format conversions in the format string? 1336 if (!HasVAListArg) { 1337 // Find any arguments that weren't covered. 1338 CoveredArgs.flip(); 1339 signed notCoveredArg = CoveredArgs.find_first(); 1340 if (notCoveredArg >= 0) { 1341 assert((unsigned)notCoveredArg < NumDataArgs); 1342 S.Diag(getDataArg((unsigned) notCoveredArg)->getLocStart(), 1343 diag::warn_printf_data_arg_not_used) 1344 << getFormatStringRange(); 1345 } 1346 } 1347} 1348 1349bool 1350CheckFormatHandler::HandleInvalidConversionSpecifier(unsigned argIndex, 1351 SourceLocation Loc, 1352 const char *startSpec, 1353 unsigned specifierLen, 1354 const char *csStart, 1355 unsigned csLen) { 1356 1357 bool keepGoing = true; 1358 if (argIndex < NumDataArgs) { 1359 // Consider the argument coverered, even though the specifier doesn't 1360 // make sense. 1361 CoveredArgs.set(argIndex); 1362 } 1363 else { 1364 // If argIndex exceeds the number of data arguments we 1365 // don't issue a warning because that is just a cascade of warnings (and 1366 // they may have intended '%%' anyway). We don't want to continue processing 1367 // the format string after this point, however, as we will like just get 1368 // gibberish when trying to match arguments. 1369 keepGoing = false; 1370 } 1371 1372 S.Diag(Loc, diag::warn_format_invalid_conversion) 1373 << StringRef(csStart, csLen) 1374 << getSpecifierRange(startSpec, specifierLen); 1375 1376 return keepGoing; 1377} 1378 1379bool 1380CheckFormatHandler::CheckNumArgs( 1381 const analyze_format_string::FormatSpecifier &FS, 1382 const analyze_format_string::ConversionSpecifier &CS, 1383 const char *startSpecifier, unsigned specifierLen, unsigned argIndex) { 1384 1385 if (argIndex >= NumDataArgs) { 1386 if (FS.usesPositionalArg()) { 1387 S.Diag(getLocationOfByte(CS.getStart()), 1388 diag::warn_printf_positional_arg_exceeds_data_args) 1389 << (argIndex+1) << NumDataArgs 1390 << getSpecifierRange(startSpecifier, specifierLen); 1391 } 1392 else { 1393 S.Diag(getLocationOfByte(CS.getStart()), 1394 diag::warn_printf_insufficient_data_args) 1395 << getSpecifierRange(startSpecifier, specifierLen); 1396 } 1397 1398 return false; 1399 } 1400 return true; 1401} 1402 1403//===--- CHECK: Printf format string checking ------------------------------===// 1404 1405namespace { 1406class CheckPrintfHandler : public CheckFormatHandler { 1407public: 1408 CheckPrintfHandler(Sema &s, const StringLiteral *fexpr, 1409 const Expr *origFormatExpr, unsigned firstDataArg, 1410 unsigned numDataArgs, bool isObjCLiteral, 1411 const char *beg, bool hasVAListArg, 1412 const CallExpr *theCall, unsigned formatIdx) 1413 : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg, 1414 numDataArgs, isObjCLiteral, beg, hasVAListArg, 1415 theCall, formatIdx) {} 1416 1417 1418 bool HandleInvalidPrintfConversionSpecifier( 1419 const analyze_printf::PrintfSpecifier &FS, 1420 const char *startSpecifier, 1421 unsigned specifierLen); 1422 1423 bool HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier &FS, 1424 const char *startSpecifier, 1425 unsigned specifierLen); 1426 1427 bool HandleAmount(const analyze_format_string::OptionalAmount &Amt, unsigned k, 1428 const char *startSpecifier, unsigned specifierLen); 1429 void HandleInvalidAmount(const analyze_printf::PrintfSpecifier &FS, 1430 const analyze_printf::OptionalAmount &Amt, 1431 unsigned type, 1432 const char *startSpecifier, unsigned specifierLen); 1433 void HandleFlag(const analyze_printf::PrintfSpecifier &FS, 1434 const analyze_printf::OptionalFlag &flag, 1435 const char *startSpecifier, unsigned specifierLen); 1436 void HandleIgnoredFlag(const analyze_printf::PrintfSpecifier &FS, 1437 const analyze_printf::OptionalFlag &ignoredFlag, 1438 const analyze_printf::OptionalFlag &flag, 1439 const char *startSpecifier, unsigned specifierLen); 1440}; 1441} 1442 1443bool CheckPrintfHandler::HandleInvalidPrintfConversionSpecifier( 1444 const analyze_printf::PrintfSpecifier &FS, 1445 const char *startSpecifier, 1446 unsigned specifierLen) { 1447 const analyze_printf::PrintfConversionSpecifier &CS = 1448 FS.getConversionSpecifier(); 1449 1450 return HandleInvalidConversionSpecifier(FS.getArgIndex(), 1451 getLocationOfByte(CS.getStart()), 1452 startSpecifier, specifierLen, 1453 CS.getStart(), CS.getLength()); 1454} 1455 1456bool CheckPrintfHandler::HandleAmount( 1457 const analyze_format_string::OptionalAmount &Amt, 1458 unsigned k, const char *startSpecifier, 1459 unsigned specifierLen) { 1460 1461 if (Amt.hasDataArgument()) { 1462 if (!HasVAListArg) { 1463 unsigned argIndex = Amt.getArgIndex(); 1464 if (argIndex >= NumDataArgs) { 1465 S.Diag(getLocationOfByte(Amt.getStart()), 1466 diag::warn_printf_asterisk_missing_arg) 1467 << k << getSpecifierRange(startSpecifier, specifierLen); 1468 // Don't do any more checking. We will just emit 1469 // spurious errors. 1470 return false; 1471 } 1472 1473 // Type check the data argument. It should be an 'int'. 1474 // Although not in conformance with C99, we also allow the argument to be 1475 // an 'unsigned int' as that is a reasonably safe case. GCC also 1476 // doesn't emit a warning for that case. 1477 CoveredArgs.set(argIndex); 1478 const Expr *Arg = getDataArg(argIndex); 1479 QualType T = Arg->getType(); 1480 1481 const analyze_printf::ArgTypeResult &ATR = Amt.getArgType(S.Context); 1482 assert(ATR.isValid()); 1483 1484 if (!ATR.matchesType(S.Context, T)) { 1485 S.Diag(getLocationOfByte(Amt.getStart()), 1486 diag::warn_printf_asterisk_wrong_type) 1487 << k 1488 << ATR.getRepresentativeType(S.Context) << T 1489 << getSpecifierRange(startSpecifier, specifierLen) 1490 << Arg->getSourceRange(); 1491 // Don't do any more checking. We will just emit 1492 // spurious errors. 1493 return false; 1494 } 1495 } 1496 } 1497 return true; 1498} 1499 1500void CheckPrintfHandler::HandleInvalidAmount( 1501 const analyze_printf::PrintfSpecifier &FS, 1502 const analyze_printf::OptionalAmount &Amt, 1503 unsigned type, 1504 const char *startSpecifier, 1505 unsigned specifierLen) { 1506 const analyze_printf::PrintfConversionSpecifier &CS = 1507 FS.getConversionSpecifier(); 1508 switch (Amt.getHowSpecified()) { 1509 case analyze_printf::OptionalAmount::Constant: 1510 S.Diag(getLocationOfByte(Amt.getStart()), 1511 diag::warn_printf_nonsensical_optional_amount) 1512 << type 1513 << CS.toString() 1514 << getSpecifierRange(startSpecifier, specifierLen) 1515 << FixItHint::CreateRemoval(getSpecifierRange(Amt.getStart(), 1516 Amt.getConstantLength())); 1517 break; 1518 1519 default: 1520 S.Diag(getLocationOfByte(Amt.getStart()), 1521 diag::warn_printf_nonsensical_optional_amount) 1522 << type 1523 << CS.toString() 1524 << getSpecifierRange(startSpecifier, specifierLen); 1525 break; 1526 } 1527} 1528 1529void CheckPrintfHandler::HandleFlag(const analyze_printf::PrintfSpecifier &FS, 1530 const analyze_printf::OptionalFlag &flag, 1531 const char *startSpecifier, 1532 unsigned specifierLen) { 1533 // Warn about pointless flag with a fixit removal. 1534 const analyze_printf::PrintfConversionSpecifier &CS = 1535 FS.getConversionSpecifier(); 1536 S.Diag(getLocationOfByte(flag.getPosition()), 1537 diag::warn_printf_nonsensical_flag) 1538 << flag.toString() << CS.toString() 1539 << getSpecifierRange(startSpecifier, specifierLen) 1540 << FixItHint::CreateRemoval(getSpecifierRange(flag.getPosition(), 1)); 1541} 1542 1543void CheckPrintfHandler::HandleIgnoredFlag( 1544 const analyze_printf::PrintfSpecifier &FS, 1545 const analyze_printf::OptionalFlag &ignoredFlag, 1546 const analyze_printf::OptionalFlag &flag, 1547 const char *startSpecifier, 1548 unsigned specifierLen) { 1549 // Warn about ignored flag with a fixit removal. 1550 S.Diag(getLocationOfByte(ignoredFlag.getPosition()), 1551 diag::warn_printf_ignored_flag) 1552 << ignoredFlag.toString() << flag.toString() 1553 << getSpecifierRange(startSpecifier, specifierLen) 1554 << FixItHint::CreateRemoval(getSpecifierRange( 1555 ignoredFlag.getPosition(), 1)); 1556} 1557 1558bool 1559CheckPrintfHandler::HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier 1560 &FS, 1561 const char *startSpecifier, 1562 unsigned specifierLen) { 1563 1564 using namespace analyze_format_string; 1565 using namespace analyze_printf; 1566 const PrintfConversionSpecifier &CS = FS.getConversionSpecifier(); 1567 1568 if (FS.consumesDataArgument()) { 1569 if (atFirstArg) { 1570 atFirstArg = false; 1571 usesPositionalArgs = FS.usesPositionalArg(); 1572 } 1573 else if (usesPositionalArgs != FS.usesPositionalArg()) { 1574 // Cannot mix-and-match positional and non-positional arguments. 1575 S.Diag(getLocationOfByte(CS.getStart()), 1576 diag::warn_format_mix_positional_nonpositional_args) 1577 << getSpecifierRange(startSpecifier, specifierLen); 1578 return false; 1579 } 1580 } 1581 1582 // First check if the field width, precision, and conversion specifier 1583 // have matching data arguments. 1584 if (!HandleAmount(FS.getFieldWidth(), /* field width */ 0, 1585 startSpecifier, specifierLen)) { 1586 return false; 1587 } 1588 1589 if (!HandleAmount(FS.getPrecision(), /* precision */ 1, 1590 startSpecifier, specifierLen)) { 1591 return false; 1592 } 1593 1594 if (!CS.consumesDataArgument()) { 1595 // FIXME: Technically specifying a precision or field width here 1596 // makes no sense. Worth issuing a warning at some point. 1597 return true; 1598 } 1599 1600 // Consume the argument. 1601 unsigned argIndex = FS.getArgIndex(); 1602 if (argIndex < NumDataArgs) { 1603 // The check to see if the argIndex is valid will come later. 1604 // We set the bit here because we may exit early from this 1605 // function if we encounter some other error. 1606 CoveredArgs.set(argIndex); 1607 } 1608 1609 // Check for using an Objective-C specific conversion specifier 1610 // in a non-ObjC literal. 1611 if (!IsObjCLiteral && CS.isObjCArg()) { 1612 return HandleInvalidPrintfConversionSpecifier(FS, startSpecifier, 1613 specifierLen); 1614 } 1615 1616 // Check for invalid use of field width 1617 if (!FS.hasValidFieldWidth()) { 1618 HandleInvalidAmount(FS, FS.getFieldWidth(), /* field width */ 0, 1619 startSpecifier, specifierLen); 1620 } 1621 1622 // Check for invalid use of precision 1623 if (!FS.hasValidPrecision()) { 1624 HandleInvalidAmount(FS, FS.getPrecision(), /* precision */ 1, 1625 startSpecifier, specifierLen); 1626 } 1627 1628 // Check each flag does not conflict with any other component. 1629 if (!FS.hasValidThousandsGroupingPrefix()) 1630 HandleFlag(FS, FS.hasThousandsGrouping(), startSpecifier, specifierLen); 1631 if (!FS.hasValidLeadingZeros()) 1632 HandleFlag(FS, FS.hasLeadingZeros(), startSpecifier, specifierLen); 1633 if (!FS.hasValidPlusPrefix()) 1634 HandleFlag(FS, FS.hasPlusPrefix(), startSpecifier, specifierLen); 1635 if (!FS.hasValidSpacePrefix()) 1636 HandleFlag(FS, FS.hasSpacePrefix(), startSpecifier, specifierLen); 1637 if (!FS.hasValidAlternativeForm()) 1638 HandleFlag(FS, FS.hasAlternativeForm(), startSpecifier, specifierLen); 1639 if (!FS.hasValidLeftJustified()) 1640 HandleFlag(FS, FS.isLeftJustified(), startSpecifier, specifierLen); 1641 1642 // Check that flags are not ignored by another flag 1643 if (FS.hasSpacePrefix() && FS.hasPlusPrefix()) // ' ' ignored by '+' 1644 HandleIgnoredFlag(FS, FS.hasSpacePrefix(), FS.hasPlusPrefix(), 1645 startSpecifier, specifierLen); 1646 if (FS.hasLeadingZeros() && FS.isLeftJustified()) // '0' ignored by '-' 1647 HandleIgnoredFlag(FS, FS.hasLeadingZeros(), FS.isLeftJustified(), 1648 startSpecifier, specifierLen); 1649 1650 // Check the length modifier is valid with the given conversion specifier. 1651 const LengthModifier &LM = FS.getLengthModifier(); 1652 if (!FS.hasValidLengthModifier()) 1653 S.Diag(getLocationOfByte(LM.getStart()), 1654 diag::warn_format_nonsensical_length) 1655 << LM.toString() << CS.toString() 1656 << getSpecifierRange(startSpecifier, specifierLen) 1657 << FixItHint::CreateRemoval(getSpecifierRange(LM.getStart(), 1658 LM.getLength())); 1659 1660 // Are we using '%n'? 1661 if (CS.getKind() == ConversionSpecifier::nArg) { 1662 // Issue a warning about this being a possible security issue. 1663 S.Diag(getLocationOfByte(CS.getStart()), diag::warn_printf_write_back) 1664 << getSpecifierRange(startSpecifier, specifierLen); 1665 // Continue checking the other format specifiers. 1666 return true; 1667 } 1668 1669 // The remaining checks depend on the data arguments. 1670 if (HasVAListArg) 1671 return true; 1672 1673 if (!CheckNumArgs(FS, CS, startSpecifier, specifierLen, argIndex)) 1674 return false; 1675 1676 // Now type check the data expression that matches the 1677 // format specifier. 1678 const Expr *Ex = getDataArg(argIndex); 1679 const analyze_printf::ArgTypeResult &ATR = FS.getArgType(S.Context); 1680 if (ATR.isValid() && !ATR.matchesType(S.Context, Ex->getType())) { 1681 // Check if we didn't match because of an implicit cast from a 'char' 1682 // or 'short' to an 'int'. This is done because printf is a varargs 1683 // function. 1684 if (const ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(Ex)) 1685 if (ICE->getType() == S.Context.IntTy) { 1686 // All further checking is done on the subexpression. 1687 Ex = ICE->getSubExpr(); 1688 if (ATR.matchesType(S.Context, Ex->getType())) 1689 return true; 1690 } 1691 1692 // We may be able to offer a FixItHint if it is a supported type. 1693 PrintfSpecifier fixedFS = FS; 1694 bool success = fixedFS.fixType(Ex->getType()); 1695 1696 if (success) { 1697 // Get the fix string from the fixed format specifier 1698 llvm::SmallString<128> buf; 1699 llvm::raw_svector_ostream os(buf); 1700 fixedFS.toString(os); 1701 1702 // FIXME: getRepresentativeType() perhaps should return a string 1703 // instead of a QualType to better handle when the representative 1704 // type is 'wint_t' (which is defined in the system headers). 1705 S.Diag(getLocationOfByte(CS.getStart()), 1706 diag::warn_printf_conversion_argument_type_mismatch) 1707 << ATR.getRepresentativeType(S.Context) << Ex->getType() 1708 << getSpecifierRange(startSpecifier, specifierLen) 1709 << Ex->getSourceRange() 1710 << FixItHint::CreateReplacement( 1711 getSpecifierRange(startSpecifier, specifierLen), 1712 os.str()); 1713 } 1714 else { 1715 S.Diag(getLocationOfByte(CS.getStart()), 1716 diag::warn_printf_conversion_argument_type_mismatch) 1717 << ATR.getRepresentativeType(S.Context) << Ex->getType() 1718 << getSpecifierRange(startSpecifier, specifierLen) 1719 << Ex->getSourceRange(); 1720 } 1721 } 1722 1723 return true; 1724} 1725 1726//===--- CHECK: Scanf format string checking ------------------------------===// 1727 1728namespace { 1729class CheckScanfHandler : public CheckFormatHandler { 1730public: 1731 CheckScanfHandler(Sema &s, const StringLiteral *fexpr, 1732 const Expr *origFormatExpr, unsigned firstDataArg, 1733 unsigned numDataArgs, bool isObjCLiteral, 1734 const char *beg, bool hasVAListArg, 1735 const CallExpr *theCall, unsigned formatIdx) 1736 : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg, 1737 numDataArgs, isObjCLiteral, beg, hasVAListArg, 1738 theCall, formatIdx) {} 1739 1740 bool HandleScanfSpecifier(const analyze_scanf::ScanfSpecifier &FS, 1741 const char *startSpecifier, 1742 unsigned specifierLen); 1743 1744 bool HandleInvalidScanfConversionSpecifier( 1745 const analyze_scanf::ScanfSpecifier &FS, 1746 const char *startSpecifier, 1747 unsigned specifierLen); 1748 1749 void HandleIncompleteScanList(const char *start, const char *end); 1750}; 1751} 1752 1753void CheckScanfHandler::HandleIncompleteScanList(const char *start, 1754 const char *end) { 1755 S.Diag(getLocationOfByte(end), diag::warn_scanf_scanlist_incomplete) 1756 << getSpecifierRange(start, end - start); 1757} 1758 1759bool CheckScanfHandler::HandleInvalidScanfConversionSpecifier( 1760 const analyze_scanf::ScanfSpecifier &FS, 1761 const char *startSpecifier, 1762 unsigned specifierLen) { 1763 1764 const analyze_scanf::ScanfConversionSpecifier &CS = 1765 FS.getConversionSpecifier(); 1766 1767 return HandleInvalidConversionSpecifier(FS.getArgIndex(), 1768 getLocationOfByte(CS.getStart()), 1769 startSpecifier, specifierLen, 1770 CS.getStart(), CS.getLength()); 1771} 1772 1773bool CheckScanfHandler::HandleScanfSpecifier( 1774 const analyze_scanf::ScanfSpecifier &FS, 1775 const char *startSpecifier, 1776 unsigned specifierLen) { 1777 1778 using namespace analyze_scanf; 1779 using namespace analyze_format_string; 1780 1781 const ScanfConversionSpecifier &CS = FS.getConversionSpecifier(); 1782 1783 // Handle case where '%' and '*' don't consume an argument. These shouldn't 1784 // be used to decide if we are using positional arguments consistently. 1785 if (FS.consumesDataArgument()) { 1786 if (atFirstArg) { 1787 atFirstArg = false; 1788 usesPositionalArgs = FS.usesPositionalArg(); 1789 } 1790 else if (usesPositionalArgs != FS.usesPositionalArg()) { 1791 // Cannot mix-and-match positional and non-positional arguments. 1792 S.Diag(getLocationOfByte(CS.getStart()), 1793 diag::warn_format_mix_positional_nonpositional_args) 1794 << getSpecifierRange(startSpecifier, specifierLen); 1795 return false; 1796 } 1797 } 1798 1799 // Check if the field with is non-zero. 1800 const OptionalAmount &Amt = FS.getFieldWidth(); 1801 if (Amt.getHowSpecified() == OptionalAmount::Constant) { 1802 if (Amt.getConstantAmount() == 0) { 1803 const CharSourceRange &R = getSpecifierRange(Amt.getStart(), 1804 Amt.getConstantLength()); 1805 S.Diag(getLocationOfByte(Amt.getStart()), 1806 diag::warn_scanf_nonzero_width) 1807 << R << FixItHint::CreateRemoval(R); 1808 } 1809 } 1810 1811 if (!FS.consumesDataArgument()) { 1812 // FIXME: Technically specifying a precision or field width here 1813 // makes no sense. Worth issuing a warning at some point. 1814 return true; 1815 } 1816 1817 // Consume the argument. 1818 unsigned argIndex = FS.getArgIndex(); 1819 if (argIndex < NumDataArgs) { 1820 // The check to see if the argIndex is valid will come later. 1821 // We set the bit here because we may exit early from this 1822 // function if we encounter some other error. 1823 CoveredArgs.set(argIndex); 1824 } 1825 1826 // Check the length modifier is valid with the given conversion specifier. 1827 const LengthModifier &LM = FS.getLengthModifier(); 1828 if (!FS.hasValidLengthModifier()) { 1829 S.Diag(getLocationOfByte(LM.getStart()), 1830 diag::warn_format_nonsensical_length) 1831 << LM.toString() << CS.toString() 1832 << getSpecifierRange(startSpecifier, specifierLen) 1833 << FixItHint::CreateRemoval(getSpecifierRange(LM.getStart(), 1834 LM.getLength())); 1835 } 1836 1837 // The remaining checks depend on the data arguments. 1838 if (HasVAListArg) 1839 return true; 1840 1841 if (!CheckNumArgs(FS, CS, startSpecifier, specifierLen, argIndex)) 1842 return false; 1843 1844 // FIXME: Check that the argument type matches the format specifier. 1845 1846 return true; 1847} 1848 1849void Sema::CheckFormatString(const StringLiteral *FExpr, 1850 const Expr *OrigFormatExpr, 1851 const CallExpr *TheCall, bool HasVAListArg, 1852 unsigned format_idx, unsigned firstDataArg, 1853 bool isPrintf) { 1854 1855 // CHECK: is the format string a wide literal? 1856 if (!FExpr->isAscii()) { 1857 Diag(FExpr->getLocStart(), 1858 diag::warn_format_string_is_wide_literal) 1859 << OrigFormatExpr->getSourceRange(); 1860 return; 1861 } 1862 1863 // Str - The format string. NOTE: this is NOT null-terminated! 1864 StringRef StrRef = FExpr->getString(); 1865 const char *Str = StrRef.data(); 1866 unsigned StrLen = StrRef.size(); 1867 1868 // CHECK: empty format string? 1869 if (StrLen == 0) { 1870 Diag(FExpr->getLocStart(), diag::warn_empty_format_string) 1871 << OrigFormatExpr->getSourceRange(); 1872 return; 1873 } 1874 1875 if (isPrintf) { 1876 CheckPrintfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg, 1877 TheCall->getNumArgs() - firstDataArg, 1878 isa<ObjCStringLiteral>(OrigFormatExpr), Str, 1879 HasVAListArg, TheCall, format_idx); 1880 1881 if (!analyze_format_string::ParsePrintfString(H, Str, Str + StrLen)) 1882 H.DoneProcessing(); 1883 } 1884 else { 1885 CheckScanfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg, 1886 TheCall->getNumArgs() - firstDataArg, 1887 isa<ObjCStringLiteral>(OrigFormatExpr), Str, 1888 HasVAListArg, TheCall, format_idx); 1889 1890 if (!analyze_format_string::ParseScanfString(H, Str, Str + StrLen)) 1891 H.DoneProcessing(); 1892 } 1893} 1894 1895//===--- CHECK: Standard memory functions ---------------------------------===// 1896 1897/// \brief Determine whether the given type is a dynamic class type (e.g., 1898/// whether it has a vtable). 1899static bool isDynamicClassType(QualType T) { 1900 if (CXXRecordDecl *Record = T->getAsCXXRecordDecl()) 1901 if (CXXRecordDecl *Definition = Record->getDefinition()) 1902 if (Definition->isDynamicClass()) 1903 return true; 1904 1905 return false; 1906} 1907 1908/// \brief If E is a sizeof expression, returns its argument expression, 1909/// otherwise returns NULL. 1910static const Expr *getSizeOfExprArg(const Expr* E) { 1911 if (const UnaryExprOrTypeTraitExpr *SizeOf = 1912 dyn_cast<UnaryExprOrTypeTraitExpr>(E)) 1913 if (SizeOf->getKind() == clang::UETT_SizeOf && !SizeOf->isArgumentType()) 1914 return SizeOf->getArgumentExpr()->IgnoreParenImpCasts(); 1915 1916 return 0; 1917} 1918 1919/// \brief If E is a sizeof expression, returns its argument type. 1920static QualType getSizeOfArgType(const Expr* E) { 1921 if (const UnaryExprOrTypeTraitExpr *SizeOf = 1922 dyn_cast<UnaryExprOrTypeTraitExpr>(E)) 1923 if (SizeOf->getKind() == clang::UETT_SizeOf) 1924 return SizeOf->getTypeOfArgument(); 1925 1926 return QualType(); 1927} 1928 1929/// \brief Check for dangerous or invalid arguments to memset(). 1930/// 1931/// This issues warnings on known problematic, dangerous or unspecified 1932/// arguments to the standard 'memset', 'memcpy', 'memmove', and 'memcmp' 1933/// function calls. 1934/// 1935/// \param Call The call expression to diagnose. 1936void Sema::CheckMemaccessArguments(const CallExpr *Call, 1937 CheckedMemoryFunction CMF, 1938 IdentifierInfo *FnName) { 1939 // It is possible to have a non-standard definition of memset. Validate 1940 // we have enough arguments, and if not, abort further checking. 1941 if (Call->getNumArgs() < 3) 1942 return; 1943 1944 unsigned LastArg = (CMF == CMF_Memset? 1 : 2); 1945 const Expr *LenExpr = Call->getArg(2)->IgnoreParenImpCasts(); 1946 1947 // We have special checking when the length is a sizeof expression. 1948 QualType SizeOfArgTy = getSizeOfArgType(LenExpr); 1949 const Expr *SizeOfArg = getSizeOfExprArg(LenExpr); 1950 llvm::FoldingSetNodeID SizeOfArgID; 1951 1952 for (unsigned ArgIdx = 0; ArgIdx != LastArg; ++ArgIdx) { 1953 const Expr *Dest = Call->getArg(ArgIdx)->IgnoreParenImpCasts(); 1954 SourceRange ArgRange = Call->getArg(ArgIdx)->getSourceRange(); 1955 1956 QualType DestTy = Dest->getType(); 1957 if (const PointerType *DestPtrTy = DestTy->getAs<PointerType>()) { 1958 QualType PointeeTy = DestPtrTy->getPointeeType(); 1959 1960 // Never warn about void type pointers. This can be used to suppress 1961 // false positives. 1962 if (PointeeTy->isVoidType()) 1963 continue; 1964 1965 // Catch "memset(p, 0, sizeof(p))" -- needs to be sizeof(*p). Do this by 1966 // actually comparing the expressions for equality. Because computing the 1967 // expression IDs can be expensive, we only do this if the diagnostic is 1968 // enabled. 1969 if (SizeOfArg && 1970 Diags.getDiagnosticLevel(diag::warn_sizeof_pointer_expr_memaccess, 1971 SizeOfArg->getExprLoc())) { 1972 // We only compute IDs for expressions if the warning is enabled, and 1973 // cache the sizeof arg's ID. 1974 if (SizeOfArgID == llvm::FoldingSetNodeID()) 1975 SizeOfArg->Profile(SizeOfArgID, Context, true); 1976 llvm::FoldingSetNodeID DestID; 1977 Dest->Profile(DestID, Context, true); 1978 if (DestID == SizeOfArgID) { 1979 unsigned ActionIdx = 0; // Default is to suggest dereferencing. 1980 if (const UnaryOperator *UnaryOp = dyn_cast<UnaryOperator>(Dest)) 1981 if (UnaryOp->getOpcode() == UO_AddrOf) 1982 ActionIdx = 1; // If its an address-of operator, just remove it. 1983 if (Context.getTypeSize(PointeeTy) == Context.getCharWidth()) 1984 ActionIdx = 2; // If the pointee's size is sizeof(char), 1985 // suggest an explicit length. 1986 DiagRuntimeBehavior(SizeOfArg->getExprLoc(), Dest, 1987 PDiag(diag::warn_sizeof_pointer_expr_memaccess) 1988 << FnName << ArgIdx << ActionIdx 1989 << Dest->getSourceRange() 1990 << SizeOfArg->getSourceRange()); 1991 break; 1992 } 1993 } 1994 1995 // Also check for cases where the sizeof argument is the exact same 1996 // type as the memory argument, and where it points to a user-defined 1997 // record type. 1998 if (SizeOfArgTy != QualType()) { 1999 if (PointeeTy->isRecordType() && 2000 Context.typesAreCompatible(SizeOfArgTy, DestTy)) { 2001 DiagRuntimeBehavior(LenExpr->getExprLoc(), Dest, 2002 PDiag(diag::warn_sizeof_pointer_type_memaccess) 2003 << FnName << SizeOfArgTy << ArgIdx 2004 << PointeeTy << Dest->getSourceRange() 2005 << LenExpr->getSourceRange()); 2006 break; 2007 } 2008 } 2009 2010 // Always complain about dynamic classes. 2011 if (isDynamicClassType(PointeeTy)) 2012 DiagRuntimeBehavior( 2013 Dest->getExprLoc(), Dest, 2014 PDiag(diag::warn_dyn_class_memaccess) 2015 << (CMF == CMF_Memcmp ? ArgIdx + 2 : ArgIdx) << FnName << PointeeTy 2016 // "overwritten" if we're warning about the destination for any call 2017 // but memcmp; otherwise a verb appropriate to the call. 2018 << (ArgIdx == 0 && CMF != CMF_Memcmp ? 0 : (unsigned)CMF) 2019 << Call->getCallee()->getSourceRange()); 2020 else if (PointeeTy.hasNonTrivialObjCLifetime() && CMF != CMF_Memset) 2021 DiagRuntimeBehavior( 2022 Dest->getExprLoc(), Dest, 2023 PDiag(diag::warn_arc_object_memaccess) 2024 << ArgIdx << FnName << PointeeTy 2025 << Call->getCallee()->getSourceRange()); 2026 else 2027 continue; 2028 2029 DiagRuntimeBehavior( 2030 Dest->getExprLoc(), Dest, 2031 PDiag(diag::note_bad_memaccess_silence) 2032 << FixItHint::CreateInsertion(ArgRange.getBegin(), "(void*)")); 2033 break; 2034 } 2035 } 2036} 2037 2038// A little helper routine: ignore addition and subtraction of integer literals. 2039// This intentionally does not ignore all integer constant expressions because 2040// we don't want to remove sizeof(). 2041static const Expr *ignoreLiteralAdditions(const Expr *Ex, ASTContext &Ctx) { 2042 Ex = Ex->IgnoreParenCasts(); 2043 2044 for (;;) { 2045 const BinaryOperator * BO = dyn_cast<BinaryOperator>(Ex); 2046 if (!BO || !BO->isAdditiveOp()) 2047 break; 2048 2049 const Expr *RHS = BO->getRHS()->IgnoreParenCasts(); 2050 const Expr *LHS = BO->getLHS()->IgnoreParenCasts(); 2051 2052 if (isa<IntegerLiteral>(RHS)) 2053 Ex = LHS; 2054 else if (isa<IntegerLiteral>(LHS)) 2055 Ex = RHS; 2056 else 2057 break; 2058 } 2059 2060 return Ex; 2061} 2062 2063// Warn if the user has made the 'size' argument to strlcpy or strlcat 2064// be the size of the source, instead of the destination. 2065void Sema::CheckStrlcpycatArguments(const CallExpr *Call, 2066 IdentifierInfo *FnName) { 2067 2068 // Don't crash if the user has the wrong number of arguments 2069 if (Call->getNumArgs() != 3) 2070 return; 2071 2072 const Expr *SrcArg = ignoreLiteralAdditions(Call->getArg(1), Context); 2073 const Expr *SizeArg = ignoreLiteralAdditions(Call->getArg(2), Context); 2074 const Expr *CompareWithSrc = NULL; 2075 2076 // Look for 'strlcpy(dst, x, sizeof(x))' 2077 if (const Expr *Ex = getSizeOfExprArg(SizeArg)) 2078 CompareWithSrc = Ex; 2079 else { 2080 // Look for 'strlcpy(dst, x, strlen(x))' 2081 if (const CallExpr *SizeCall = dyn_cast<CallExpr>(SizeArg)) { 2082 if (SizeCall->isBuiltinCall(Context) == Builtin::BIstrlen 2083 && SizeCall->getNumArgs() == 1) 2084 CompareWithSrc = ignoreLiteralAdditions(SizeCall->getArg(0), Context); 2085 } 2086 } 2087 2088 if (!CompareWithSrc) 2089 return; 2090 2091 // Determine if the argument to sizeof/strlen is equal to the source 2092 // argument. In principle there's all kinds of things you could do 2093 // here, for instance creating an == expression and evaluating it with 2094 // EvaluateAsBooleanCondition, but this uses a more direct technique: 2095 const DeclRefExpr *SrcArgDRE = dyn_cast<DeclRefExpr>(SrcArg); 2096 if (!SrcArgDRE) 2097 return; 2098 2099 const DeclRefExpr *CompareWithSrcDRE = dyn_cast<DeclRefExpr>(CompareWithSrc); 2100 if (!CompareWithSrcDRE || 2101 SrcArgDRE->getDecl() != CompareWithSrcDRE->getDecl()) 2102 return; 2103 2104 const Expr *OriginalSizeArg = Call->getArg(2); 2105 Diag(CompareWithSrcDRE->getLocStart(), diag::warn_strlcpycat_wrong_size) 2106 << OriginalSizeArg->getSourceRange() << FnName; 2107 2108 // Output a FIXIT hint if the destination is an array (rather than a 2109 // pointer to an array). This could be enhanced to handle some 2110 // pointers if we know the actual size, like if DstArg is 'array+2' 2111 // we could say 'sizeof(array)-2'. 2112 const Expr *DstArg = Call->getArg(0)->IgnoreParenImpCasts(); 2113 QualType DstArgTy = DstArg->getType(); 2114 2115 // Only handle constant-sized or VLAs, but not flexible members. 2116 if (const ConstantArrayType *CAT = Context.getAsConstantArrayType(DstArgTy)) { 2117 // Only issue the FIXIT for arrays of size > 1. 2118 if (CAT->getSize().getSExtValue() <= 1) 2119 return; 2120 } else if (!DstArgTy->isVariableArrayType()) { 2121 return; 2122 } 2123 2124 llvm::SmallString<128> sizeString; 2125 llvm::raw_svector_ostream OS(sizeString); 2126 OS << "sizeof("; 2127 DstArg->printPretty(OS, Context, 0, Context.PrintingPolicy); 2128 OS << ")"; 2129 2130 Diag(OriginalSizeArg->getLocStart(), diag::note_strlcpycat_wrong_size) 2131 << FixItHint::CreateReplacement(OriginalSizeArg->getSourceRange(), 2132 OS.str()); 2133} 2134 2135//===--- CHECK: Return Address of Stack Variable --------------------------===// 2136 2137static Expr *EvalVal(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars); 2138static Expr *EvalAddr(Expr* E, SmallVectorImpl<DeclRefExpr *> &refVars); 2139 2140/// CheckReturnStackAddr - Check if a return statement returns the address 2141/// of a stack variable. 2142void 2143Sema::CheckReturnStackAddr(Expr *RetValExp, QualType lhsType, 2144 SourceLocation ReturnLoc) { 2145 2146 Expr *stackE = 0; 2147 SmallVector<DeclRefExpr *, 8> refVars; 2148 2149 // Perform checking for returned stack addresses, local blocks, 2150 // label addresses or references to temporaries. 2151 if (lhsType->isPointerType() || 2152 (!getLangOptions().ObjCAutoRefCount && lhsType->isBlockPointerType())) { 2153 stackE = EvalAddr(RetValExp, refVars); 2154 } else if (lhsType->isReferenceType()) { 2155 stackE = EvalVal(RetValExp, refVars); 2156 } 2157 2158 if (stackE == 0) 2159 return; // Nothing suspicious was found. 2160 2161 SourceLocation diagLoc; 2162 SourceRange diagRange; 2163 if (refVars.empty()) { 2164 diagLoc = stackE->getLocStart(); 2165 diagRange = stackE->getSourceRange(); 2166 } else { 2167 // We followed through a reference variable. 'stackE' contains the 2168 // problematic expression but we will warn at the return statement pointing 2169 // at the reference variable. We will later display the "trail" of 2170 // reference variables using notes. 2171 diagLoc = refVars[0]->getLocStart(); 2172 diagRange = refVars[0]->getSourceRange(); 2173 } 2174 2175 if (DeclRefExpr *DR = dyn_cast<DeclRefExpr>(stackE)) { //address of local var. 2176 Diag(diagLoc, lhsType->isReferenceType() ? diag::warn_ret_stack_ref 2177 : diag::warn_ret_stack_addr) 2178 << DR->getDecl()->getDeclName() << diagRange; 2179 } else if (isa<BlockExpr>(stackE)) { // local block. 2180 Diag(diagLoc, diag::err_ret_local_block) << diagRange; 2181 } else if (isa<AddrLabelExpr>(stackE)) { // address of label. 2182 Diag(diagLoc, diag::warn_ret_addr_label) << diagRange; 2183 } else { // local temporary. 2184 Diag(diagLoc, lhsType->isReferenceType() ? diag::warn_ret_local_temp_ref 2185 : diag::warn_ret_local_temp_addr) 2186 << diagRange; 2187 } 2188 2189 // Display the "trail" of reference variables that we followed until we 2190 // found the problematic expression using notes. 2191 for (unsigned i = 0, e = refVars.size(); i != e; ++i) { 2192 VarDecl *VD = cast<VarDecl>(refVars[i]->getDecl()); 2193 // If this var binds to another reference var, show the range of the next 2194 // var, otherwise the var binds to the problematic expression, in which case 2195 // show the range of the expression. 2196 SourceRange range = (i < e-1) ? refVars[i+1]->getSourceRange() 2197 : stackE->getSourceRange(); 2198 Diag(VD->getLocation(), diag::note_ref_var_local_bind) 2199 << VD->getDeclName() << range; 2200 } 2201} 2202 2203/// EvalAddr - EvalAddr and EvalVal are mutually recursive functions that 2204/// check if the expression in a return statement evaluates to an address 2205/// to a location on the stack, a local block, an address of a label, or a 2206/// reference to local temporary. The recursion is used to traverse the 2207/// AST of the return expression, with recursion backtracking when we 2208/// encounter a subexpression that (1) clearly does not lead to one of the 2209/// above problematic expressions (2) is something we cannot determine leads to 2210/// a problematic expression based on such local checking. 2211/// 2212/// Both EvalAddr and EvalVal follow through reference variables to evaluate 2213/// the expression that they point to. Such variables are added to the 2214/// 'refVars' vector so that we know what the reference variable "trail" was. 2215/// 2216/// EvalAddr processes expressions that are pointers that are used as 2217/// references (and not L-values). EvalVal handles all other values. 2218/// At the base case of the recursion is a check for the above problematic 2219/// expressions. 2220/// 2221/// This implementation handles: 2222/// 2223/// * pointer-to-pointer casts 2224/// * implicit conversions from array references to pointers 2225/// * taking the address of fields 2226/// * arbitrary interplay between "&" and "*" operators 2227/// * pointer arithmetic from an address of a stack variable 2228/// * taking the address of an array element where the array is on the stack 2229static Expr *EvalAddr(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars) { 2230 if (E->isTypeDependent()) 2231 return NULL; 2232 2233 // We should only be called for evaluating pointer expressions. 2234 assert((E->getType()->isAnyPointerType() || 2235 E->getType()->isBlockPointerType() || 2236 E->getType()->isObjCQualifiedIdType()) && 2237 "EvalAddr only works on pointers"); 2238 2239 E = E->IgnoreParens(); 2240 2241 // Our "symbolic interpreter" is just a dispatch off the currently 2242 // viewed AST node. We then recursively traverse the AST by calling 2243 // EvalAddr and EvalVal appropriately. 2244 switch (E->getStmtClass()) { 2245 case Stmt::DeclRefExprClass: { 2246 DeclRefExpr *DR = cast<DeclRefExpr>(E); 2247 2248 if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl())) 2249 // If this is a reference variable, follow through to the expression that 2250 // it points to. 2251 if (V->hasLocalStorage() && 2252 V->getType()->isReferenceType() && V->hasInit()) { 2253 // Add the reference variable to the "trail". 2254 refVars.push_back(DR); 2255 return EvalAddr(V->getInit(), refVars); 2256 } 2257 2258 return NULL; 2259 } 2260 2261 case Stmt::UnaryOperatorClass: { 2262 // The only unary operator that make sense to handle here 2263 // is AddrOf. All others don't make sense as pointers. 2264 UnaryOperator *U = cast<UnaryOperator>(E); 2265 2266 if (U->getOpcode() == UO_AddrOf) 2267 return EvalVal(U->getSubExpr(), refVars); 2268 else 2269 return NULL; 2270 } 2271 2272 case Stmt::BinaryOperatorClass: { 2273 // Handle pointer arithmetic. All other binary operators are not valid 2274 // in this context. 2275 BinaryOperator *B = cast<BinaryOperator>(E); 2276 BinaryOperatorKind op = B->getOpcode(); 2277 2278 if (op != BO_Add && op != BO_Sub) 2279 return NULL; 2280 2281 Expr *Base = B->getLHS(); 2282 2283 // Determine which argument is the real pointer base. It could be 2284 // the RHS argument instead of the LHS. 2285 if (!Base->getType()->isPointerType()) Base = B->getRHS(); 2286 2287 assert (Base->getType()->isPointerType()); 2288 return EvalAddr(Base, refVars); 2289 } 2290 2291 // For conditional operators we need to see if either the LHS or RHS are 2292 // valid DeclRefExpr*s. If one of them is valid, we return it. 2293 case Stmt::ConditionalOperatorClass: { 2294 ConditionalOperator *C = cast<ConditionalOperator>(E); 2295 2296 // Handle the GNU extension for missing LHS. 2297 if (Expr *lhsExpr = C->getLHS()) { 2298 // In C++, we can have a throw-expression, which has 'void' type. 2299 if (!lhsExpr->getType()->isVoidType()) 2300 if (Expr* LHS = EvalAddr(lhsExpr, refVars)) 2301 return LHS; 2302 } 2303 2304 // In C++, we can have a throw-expression, which has 'void' type. 2305 if (C->getRHS()->getType()->isVoidType()) 2306 return NULL; 2307 2308 return EvalAddr(C->getRHS(), refVars); 2309 } 2310 2311 case Stmt::BlockExprClass: 2312 if (cast<BlockExpr>(E)->getBlockDecl()->hasCaptures()) 2313 return E; // local block. 2314 return NULL; 2315 2316 case Stmt::AddrLabelExprClass: 2317 return E; // address of label. 2318 2319 // For casts, we need to handle conversions from arrays to 2320 // pointer values, and pointer-to-pointer conversions. 2321 case Stmt::ImplicitCastExprClass: 2322 case Stmt::CStyleCastExprClass: 2323 case Stmt::CXXFunctionalCastExprClass: 2324 case Stmt::ObjCBridgedCastExprClass: { 2325 Expr* SubExpr = cast<CastExpr>(E)->getSubExpr(); 2326 QualType T = SubExpr->getType(); 2327 2328 if (SubExpr->getType()->isPointerType() || 2329 SubExpr->getType()->isBlockPointerType() || 2330 SubExpr->getType()->isObjCQualifiedIdType()) 2331 return EvalAddr(SubExpr, refVars); 2332 else if (T->isArrayType()) 2333 return EvalVal(SubExpr, refVars); 2334 else 2335 return 0; 2336 } 2337 2338 // C++ casts. For dynamic casts, static casts, and const casts, we 2339 // are always converting from a pointer-to-pointer, so we just blow 2340 // through the cast. In the case the dynamic cast doesn't fail (and 2341 // return NULL), we take the conservative route and report cases 2342 // where we return the address of a stack variable. For Reinterpre 2343 // FIXME: The comment about is wrong; we're not always converting 2344 // from pointer to pointer. I'm guessing that this code should also 2345 // handle references to objects. 2346 case Stmt::CXXStaticCastExprClass: 2347 case Stmt::CXXDynamicCastExprClass: 2348 case Stmt::CXXConstCastExprClass: 2349 case Stmt::CXXReinterpretCastExprClass: { 2350 Expr *S = cast<CXXNamedCastExpr>(E)->getSubExpr(); 2351 if (S->getType()->isPointerType() || S->getType()->isBlockPointerType()) 2352 return EvalAddr(S, refVars); 2353 else 2354 return NULL; 2355 } 2356 2357 case Stmt::MaterializeTemporaryExprClass: 2358 if (Expr *Result = EvalAddr( 2359 cast<MaterializeTemporaryExpr>(E)->GetTemporaryExpr(), 2360 refVars)) 2361 return Result; 2362 2363 return E; 2364 2365 // Everything else: we simply don't reason about them. 2366 default: 2367 return NULL; 2368 } 2369} 2370 2371 2372/// EvalVal - This function is complements EvalAddr in the mutual recursion. 2373/// See the comments for EvalAddr for more details. 2374static Expr *EvalVal(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars) { 2375do { 2376 // We should only be called for evaluating non-pointer expressions, or 2377 // expressions with a pointer type that are not used as references but instead 2378 // are l-values (e.g., DeclRefExpr with a pointer type). 2379 2380 // Our "symbolic interpreter" is just a dispatch off the currently 2381 // viewed AST node. We then recursively traverse the AST by calling 2382 // EvalAddr and EvalVal appropriately. 2383 2384 E = E->IgnoreParens(); 2385 switch (E->getStmtClass()) { 2386 case Stmt::ImplicitCastExprClass: { 2387 ImplicitCastExpr *IE = cast<ImplicitCastExpr>(E); 2388 if (IE->getValueKind() == VK_LValue) { 2389 E = IE->getSubExpr(); 2390 continue; 2391 } 2392 return NULL; 2393 } 2394 2395 case Stmt::DeclRefExprClass: { 2396 // When we hit a DeclRefExpr we are looking at code that refers to a 2397 // variable's name. If it's not a reference variable we check if it has 2398 // local storage within the function, and if so, return the expression. 2399 DeclRefExpr *DR = cast<DeclRefExpr>(E); 2400 2401 if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl())) 2402 if (V->hasLocalStorage()) { 2403 if (!V->getType()->isReferenceType()) 2404 return DR; 2405 2406 // Reference variable, follow through to the expression that 2407 // it points to. 2408 if (V->hasInit()) { 2409 // Add the reference variable to the "trail". 2410 refVars.push_back(DR); 2411 return EvalVal(V->getInit(), refVars); 2412 } 2413 } 2414 2415 return NULL; 2416 } 2417 2418 case Stmt::UnaryOperatorClass: { 2419 // The only unary operator that make sense to handle here 2420 // is Deref. All others don't resolve to a "name." This includes 2421 // handling all sorts of rvalues passed to a unary operator. 2422 UnaryOperator *U = cast<UnaryOperator>(E); 2423 2424 if (U->getOpcode() == UO_Deref) 2425 return EvalAddr(U->getSubExpr(), refVars); 2426 2427 return NULL; 2428 } 2429 2430 case Stmt::ArraySubscriptExprClass: { 2431 // Array subscripts are potential references to data on the stack. We 2432 // retrieve the DeclRefExpr* for the array variable if it indeed 2433 // has local storage. 2434 return EvalAddr(cast<ArraySubscriptExpr>(E)->getBase(), refVars); 2435 } 2436 2437 case Stmt::ConditionalOperatorClass: { 2438 // For conditional operators we need to see if either the LHS or RHS are 2439 // non-NULL Expr's. If one is non-NULL, we return it. 2440 ConditionalOperator *C = cast<ConditionalOperator>(E); 2441 2442 // Handle the GNU extension for missing LHS. 2443 if (Expr *lhsExpr = C->getLHS()) 2444 if (Expr *LHS = EvalVal(lhsExpr, refVars)) 2445 return LHS; 2446 2447 return EvalVal(C->getRHS(), refVars); 2448 } 2449 2450 // Accesses to members are potential references to data on the stack. 2451 case Stmt::MemberExprClass: { 2452 MemberExpr *M = cast<MemberExpr>(E); 2453 2454 // Check for indirect access. We only want direct field accesses. 2455 if (M->isArrow()) 2456 return NULL; 2457 2458 // Check whether the member type is itself a reference, in which case 2459 // we're not going to refer to the member, but to what the member refers to. 2460 if (M->getMemberDecl()->getType()->isReferenceType()) 2461 return NULL; 2462 2463 return EvalVal(M->getBase(), refVars); 2464 } 2465 2466 case Stmt::MaterializeTemporaryExprClass: 2467 if (Expr *Result = EvalVal( 2468 cast<MaterializeTemporaryExpr>(E)->GetTemporaryExpr(), 2469 refVars)) 2470 return Result; 2471 2472 return E; 2473 2474 default: 2475 // Check that we don't return or take the address of a reference to a 2476 // temporary. This is only useful in C++. 2477 if (!E->isTypeDependent() && E->isRValue()) 2478 return E; 2479 2480 // Everything else: we simply don't reason about them. 2481 return NULL; 2482 } 2483} while (true); 2484} 2485 2486//===--- CHECK: Floating-Point comparisons (-Wfloat-equal) ---------------===// 2487 2488/// Check for comparisons of floating point operands using != and ==. 2489/// Issue a warning if these are no self-comparisons, as they are not likely 2490/// to do what the programmer intended. 2491void Sema::CheckFloatComparison(SourceLocation loc, Expr* lex, Expr *rex) { 2492 bool EmitWarning = true; 2493 2494 Expr* LeftExprSansParen = lex->IgnoreParenImpCasts(); 2495 Expr* RightExprSansParen = rex->IgnoreParenImpCasts(); 2496 2497 // Special case: check for x == x (which is OK). 2498 // Do not emit warnings for such cases. 2499 if (DeclRefExpr* DRL = dyn_cast<DeclRefExpr>(LeftExprSansParen)) 2500 if (DeclRefExpr* DRR = dyn_cast<DeclRefExpr>(RightExprSansParen)) 2501 if (DRL->getDecl() == DRR->getDecl()) 2502 EmitWarning = false; 2503 2504 2505 // Special case: check for comparisons against literals that can be exactly 2506 // represented by APFloat. In such cases, do not emit a warning. This 2507 // is a heuristic: often comparison against such literals are used to 2508 // detect if a value in a variable has not changed. This clearly can 2509 // lead to false negatives. 2510 if (EmitWarning) { 2511 if (FloatingLiteral* FLL = dyn_cast<FloatingLiteral>(LeftExprSansParen)) { 2512 if (FLL->isExact()) 2513 EmitWarning = false; 2514 } else 2515 if (FloatingLiteral* FLR = dyn_cast<FloatingLiteral>(RightExprSansParen)){ 2516 if (FLR->isExact()) 2517 EmitWarning = false; 2518 } 2519 } 2520 2521 // Check for comparisons with builtin types. 2522 if (EmitWarning) 2523 if (CallExpr* CL = dyn_cast<CallExpr>(LeftExprSansParen)) 2524 if (CL->isBuiltinCall(Context)) 2525 EmitWarning = false; 2526 2527 if (EmitWarning) 2528 if (CallExpr* CR = dyn_cast<CallExpr>(RightExprSansParen)) 2529 if (CR->isBuiltinCall(Context)) 2530 EmitWarning = false; 2531 2532 // Emit the diagnostic. 2533 if (EmitWarning) 2534 Diag(loc, diag::warn_floatingpoint_eq) 2535 << lex->getSourceRange() << rex->getSourceRange(); 2536} 2537 2538//===--- CHECK: Integer mixed-sign comparisons (-Wsign-compare) --------===// 2539//===--- CHECK: Lossy implicit conversions (-Wconversion) --------------===// 2540 2541namespace { 2542 2543/// Structure recording the 'active' range of an integer-valued 2544/// expression. 2545struct IntRange { 2546 /// The number of bits active in the int. 2547 unsigned Width; 2548 2549 /// True if the int is known not to have negative values. 2550 bool NonNegative; 2551 2552 IntRange(unsigned Width, bool NonNegative) 2553 : Width(Width), NonNegative(NonNegative) 2554 {} 2555 2556 /// Returns the range of the bool type. 2557 static IntRange forBoolType() { 2558 return IntRange(1, true); 2559 } 2560 2561 /// Returns the range of an opaque value of the given integral type. 2562 static IntRange forValueOfType(ASTContext &C, QualType T) { 2563 return forValueOfCanonicalType(C, 2564 T->getCanonicalTypeInternal().getTypePtr()); 2565 } 2566 2567 /// Returns the range of an opaque value of a canonical integral type. 2568 static IntRange forValueOfCanonicalType(ASTContext &C, const Type *T) { 2569 assert(T->isCanonicalUnqualified()); 2570 2571 if (const VectorType *VT = dyn_cast<VectorType>(T)) 2572 T = VT->getElementType().getTypePtr(); 2573 if (const ComplexType *CT = dyn_cast<ComplexType>(T)) 2574 T = CT->getElementType().getTypePtr(); 2575 2576 // For enum types, use the known bit width of the enumerators. 2577 if (const EnumType *ET = dyn_cast<EnumType>(T)) { 2578 EnumDecl *Enum = ET->getDecl(); 2579 if (!Enum->isDefinition()) 2580 return IntRange(C.getIntWidth(QualType(T, 0)), false); 2581 2582 unsigned NumPositive = Enum->getNumPositiveBits(); 2583 unsigned NumNegative = Enum->getNumNegativeBits(); 2584 2585 return IntRange(std::max(NumPositive, NumNegative), NumNegative == 0); 2586 } 2587 2588 const BuiltinType *BT = cast<BuiltinType>(T); 2589 assert(BT->isInteger()); 2590 2591 return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger()); 2592 } 2593 2594 /// Returns the "target" range of a canonical integral type, i.e. 2595 /// the range of values expressible in the type. 2596 /// 2597 /// This matches forValueOfCanonicalType except that enums have the 2598 /// full range of their type, not the range of their enumerators. 2599 static IntRange forTargetOfCanonicalType(ASTContext &C, const Type *T) { 2600 assert(T->isCanonicalUnqualified()); 2601 2602 if (const VectorType *VT = dyn_cast<VectorType>(T)) 2603 T = VT->getElementType().getTypePtr(); 2604 if (const ComplexType *CT = dyn_cast<ComplexType>(T)) 2605 T = CT->getElementType().getTypePtr(); 2606 if (const EnumType *ET = dyn_cast<EnumType>(T)) 2607 T = C.getCanonicalType(ET->getDecl()->getIntegerType()).getTypePtr(); 2608 2609 const BuiltinType *BT = cast<BuiltinType>(T); 2610 assert(BT->isInteger()); 2611 2612 return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger()); 2613 } 2614 2615 /// Returns the supremum of two ranges: i.e. their conservative merge. 2616 static IntRange join(IntRange L, IntRange R) { 2617 return IntRange(std::max(L.Width, R.Width), 2618 L.NonNegative && R.NonNegative); 2619 } 2620 2621 /// Returns the infinum of two ranges: i.e. their aggressive merge. 2622 static IntRange meet(IntRange L, IntRange R) { 2623 return IntRange(std::min(L.Width, R.Width), 2624 L.NonNegative || R.NonNegative); 2625 } 2626}; 2627 2628IntRange GetValueRange(ASTContext &C, llvm::APSInt &value, unsigned MaxWidth) { 2629 if (value.isSigned() && value.isNegative()) 2630 return IntRange(value.getMinSignedBits(), false); 2631 2632 if (value.getBitWidth() > MaxWidth) 2633 value = value.trunc(MaxWidth); 2634 2635 // isNonNegative() just checks the sign bit without considering 2636 // signedness. 2637 return IntRange(value.getActiveBits(), true); 2638} 2639 2640IntRange GetValueRange(ASTContext &C, APValue &result, QualType Ty, 2641 unsigned MaxWidth) { 2642 if (result.isInt()) 2643 return GetValueRange(C, result.getInt(), MaxWidth); 2644 2645 if (result.isVector()) { 2646 IntRange R = GetValueRange(C, result.getVectorElt(0), Ty, MaxWidth); 2647 for (unsigned i = 1, e = result.getVectorLength(); i != e; ++i) { 2648 IntRange El = GetValueRange(C, result.getVectorElt(i), Ty, MaxWidth); 2649 R = IntRange::join(R, El); 2650 } 2651 return R; 2652 } 2653 2654 if (result.isComplexInt()) { 2655 IntRange R = GetValueRange(C, result.getComplexIntReal(), MaxWidth); 2656 IntRange I = GetValueRange(C, result.getComplexIntImag(), MaxWidth); 2657 return IntRange::join(R, I); 2658 } 2659 2660 // This can happen with lossless casts to intptr_t of "based" lvalues. 2661 // Assume it might use arbitrary bits. 2662 // FIXME: The only reason we need to pass the type in here is to get 2663 // the sign right on this one case. It would be nice if APValue 2664 // preserved this. 2665 assert(result.isLValue()); 2666 return IntRange(MaxWidth, Ty->isUnsignedIntegerOrEnumerationType()); 2667} 2668 2669/// Pseudo-evaluate the given integer expression, estimating the 2670/// range of values it might take. 2671/// 2672/// \param MaxWidth - the width to which the value will be truncated 2673IntRange GetExprRange(ASTContext &C, Expr *E, unsigned MaxWidth) { 2674 E = E->IgnoreParens(); 2675 2676 // Try a full evaluation first. 2677 Expr::EvalResult result; 2678 if (E->Evaluate(result, C)) 2679 return GetValueRange(C, result.Val, E->getType(), MaxWidth); 2680 2681 // I think we only want to look through implicit casts here; if the 2682 // user has an explicit widening cast, we should treat the value as 2683 // being of the new, wider type. 2684 if (ImplicitCastExpr *CE = dyn_cast<ImplicitCastExpr>(E)) { 2685 if (CE->getCastKind() == CK_NoOp) 2686 return GetExprRange(C, CE->getSubExpr(), MaxWidth); 2687 2688 IntRange OutputTypeRange = IntRange::forValueOfType(C, CE->getType()); 2689 2690 bool isIntegerCast = (CE->getCastKind() == CK_IntegralCast); 2691 2692 // Assume that non-integer casts can span the full range of the type. 2693 if (!isIntegerCast) 2694 return OutputTypeRange; 2695 2696 IntRange SubRange 2697 = GetExprRange(C, CE->getSubExpr(), 2698 std::min(MaxWidth, OutputTypeRange.Width)); 2699 2700 // Bail out if the subexpr's range is as wide as the cast type. 2701 if (SubRange.Width >= OutputTypeRange.Width) 2702 return OutputTypeRange; 2703 2704 // Otherwise, we take the smaller width, and we're non-negative if 2705 // either the output type or the subexpr is. 2706 return IntRange(SubRange.Width, 2707 SubRange.NonNegative || OutputTypeRange.NonNegative); 2708 } 2709 2710 if (ConditionalOperator *CO = dyn_cast<ConditionalOperator>(E)) { 2711 // If we can fold the condition, just take that operand. 2712 bool CondResult; 2713 if (CO->getCond()->EvaluateAsBooleanCondition(CondResult, C)) 2714 return GetExprRange(C, CondResult ? CO->getTrueExpr() 2715 : CO->getFalseExpr(), 2716 MaxWidth); 2717 2718 // Otherwise, conservatively merge. 2719 IntRange L = GetExprRange(C, CO->getTrueExpr(), MaxWidth); 2720 IntRange R = GetExprRange(C, CO->getFalseExpr(), MaxWidth); 2721 return IntRange::join(L, R); 2722 } 2723 2724 if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) { 2725 switch (BO->getOpcode()) { 2726 2727 // Boolean-valued operations are single-bit and positive. 2728 case BO_LAnd: 2729 case BO_LOr: 2730 case BO_LT: 2731 case BO_GT: 2732 case BO_LE: 2733 case BO_GE: 2734 case BO_EQ: 2735 case BO_NE: 2736 return IntRange::forBoolType(); 2737 2738 // The type of the assignments is the type of the LHS, so the RHS 2739 // is not necessarily the same type. 2740 case BO_MulAssign: 2741 case BO_DivAssign: 2742 case BO_RemAssign: 2743 case BO_AddAssign: 2744 case BO_SubAssign: 2745 case BO_XorAssign: 2746 case BO_OrAssign: 2747 // TODO: bitfields? 2748 return IntRange::forValueOfType(C, E->getType()); 2749 2750 // Simple assignments just pass through the RHS, which will have 2751 // been coerced to the LHS type. 2752 case BO_Assign: 2753 // TODO: bitfields? 2754 return GetExprRange(C, BO->getRHS(), MaxWidth); 2755 2756 // Operations with opaque sources are black-listed. 2757 case BO_PtrMemD: 2758 case BO_PtrMemI: 2759 return IntRange::forValueOfType(C, E->getType()); 2760 2761 // Bitwise-and uses the *infinum* of the two source ranges. 2762 case BO_And: 2763 case BO_AndAssign: 2764 return IntRange::meet(GetExprRange(C, BO->getLHS(), MaxWidth), 2765 GetExprRange(C, BO->getRHS(), MaxWidth)); 2766 2767 // Left shift gets black-listed based on a judgement call. 2768 case BO_Shl: 2769 // ...except that we want to treat '1 << (blah)' as logically 2770 // positive. It's an important idiom. 2771 if (IntegerLiteral *I 2772 = dyn_cast<IntegerLiteral>(BO->getLHS()->IgnoreParenCasts())) { 2773 if (I->getValue() == 1) { 2774 IntRange R = IntRange::forValueOfType(C, E->getType()); 2775 return IntRange(R.Width, /*NonNegative*/ true); 2776 } 2777 } 2778 // fallthrough 2779 2780 case BO_ShlAssign: 2781 return IntRange::forValueOfType(C, E->getType()); 2782 2783 // Right shift by a constant can narrow its left argument. 2784 case BO_Shr: 2785 case BO_ShrAssign: { 2786 IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth); 2787 2788 // If the shift amount is a positive constant, drop the width by 2789 // that much. 2790 llvm::APSInt shift; 2791 if (BO->getRHS()->isIntegerConstantExpr(shift, C) && 2792 shift.isNonNegative()) { 2793 unsigned zext = shift.getZExtValue(); 2794 if (zext >= L.Width) 2795 L.Width = (L.NonNegative ? 0 : 1); 2796 else 2797 L.Width -= zext; 2798 } 2799 2800 return L; 2801 } 2802 2803 // Comma acts as its right operand. 2804 case BO_Comma: 2805 return GetExprRange(C, BO->getRHS(), MaxWidth); 2806 2807 // Black-list pointer subtractions. 2808 case BO_Sub: 2809 if (BO->getLHS()->getType()->isPointerType()) 2810 return IntRange::forValueOfType(C, E->getType()); 2811 break; 2812 2813 // The width of a division result is mostly determined by the size 2814 // of the LHS. 2815 case BO_Div: { 2816 // Don't 'pre-truncate' the operands. 2817 unsigned opWidth = C.getIntWidth(E->getType()); 2818 IntRange L = GetExprRange(C, BO->getLHS(), opWidth); 2819 2820 // If the divisor is constant, use that. 2821 llvm::APSInt divisor; 2822 if (BO->getRHS()->isIntegerConstantExpr(divisor, C)) { 2823 unsigned log2 = divisor.logBase2(); // floor(log_2(divisor)) 2824 if (log2 >= L.Width) 2825 L.Width = (L.NonNegative ? 0 : 1); 2826 else 2827 L.Width = std::min(L.Width - log2, MaxWidth); 2828 return L; 2829 } 2830 2831 // Otherwise, just use the LHS's width. 2832 IntRange R = GetExprRange(C, BO->getRHS(), opWidth); 2833 return IntRange(L.Width, L.NonNegative && R.NonNegative); 2834 } 2835 2836 // The result of a remainder can't be larger than the result of 2837 // either side. 2838 case BO_Rem: { 2839 // Don't 'pre-truncate' the operands. 2840 unsigned opWidth = C.getIntWidth(E->getType()); 2841 IntRange L = GetExprRange(C, BO->getLHS(), opWidth); 2842 IntRange R = GetExprRange(C, BO->getRHS(), opWidth); 2843 2844 IntRange meet = IntRange::meet(L, R); 2845 meet.Width = std::min(meet.Width, MaxWidth); 2846 return meet; 2847 } 2848 2849 // The default behavior is okay for these. 2850 case BO_Mul: 2851 case BO_Add: 2852 case BO_Xor: 2853 case BO_Or: 2854 break; 2855 } 2856 2857 // The default case is to treat the operation as if it were closed 2858 // on the narrowest type that encompasses both operands. 2859 IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth); 2860 IntRange R = GetExprRange(C, BO->getRHS(), MaxWidth); 2861 return IntRange::join(L, R); 2862 } 2863 2864 if (UnaryOperator *UO = dyn_cast<UnaryOperator>(E)) { 2865 switch (UO->getOpcode()) { 2866 // Boolean-valued operations are white-listed. 2867 case UO_LNot: 2868 return IntRange::forBoolType(); 2869 2870 // Operations with opaque sources are black-listed. 2871 case UO_Deref: 2872 case UO_AddrOf: // should be impossible 2873 return IntRange::forValueOfType(C, E->getType()); 2874 2875 default: 2876 return GetExprRange(C, UO->getSubExpr(), MaxWidth); 2877 } 2878 } 2879 2880 if (dyn_cast<OffsetOfExpr>(E)) { 2881 IntRange::forValueOfType(C, E->getType()); 2882 } 2883 2884 FieldDecl *BitField = E->getBitField(); 2885 if (BitField) { 2886 llvm::APSInt BitWidthAP = BitField->getBitWidth()->EvaluateAsInt(C); 2887 unsigned BitWidth = BitWidthAP.getZExtValue(); 2888 2889 return IntRange(BitWidth, 2890 BitField->getType()->isUnsignedIntegerOrEnumerationType()); 2891 } 2892 2893 return IntRange::forValueOfType(C, E->getType()); 2894} 2895 2896IntRange GetExprRange(ASTContext &C, Expr *E) { 2897 return GetExprRange(C, E, C.getIntWidth(E->getType())); 2898} 2899 2900/// Checks whether the given value, which currently has the given 2901/// source semantics, has the same value when coerced through the 2902/// target semantics. 2903bool IsSameFloatAfterCast(const llvm::APFloat &value, 2904 const llvm::fltSemantics &Src, 2905 const llvm::fltSemantics &Tgt) { 2906 llvm::APFloat truncated = value; 2907 2908 bool ignored; 2909 truncated.convert(Src, llvm::APFloat::rmNearestTiesToEven, &ignored); 2910 truncated.convert(Tgt, llvm::APFloat::rmNearestTiesToEven, &ignored); 2911 2912 return truncated.bitwiseIsEqual(value); 2913} 2914 2915/// Checks whether the given value, which currently has the given 2916/// source semantics, has the same value when coerced through the 2917/// target semantics. 2918/// 2919/// The value might be a vector of floats (or a complex number). 2920bool IsSameFloatAfterCast(const APValue &value, 2921 const llvm::fltSemantics &Src, 2922 const llvm::fltSemantics &Tgt) { 2923 if (value.isFloat()) 2924 return IsSameFloatAfterCast(value.getFloat(), Src, Tgt); 2925 2926 if (value.isVector()) { 2927 for (unsigned i = 0, e = value.getVectorLength(); i != e; ++i) 2928 if (!IsSameFloatAfterCast(value.getVectorElt(i), Src, Tgt)) 2929 return false; 2930 return true; 2931 } 2932 2933 assert(value.isComplexFloat()); 2934 return (IsSameFloatAfterCast(value.getComplexFloatReal(), Src, Tgt) && 2935 IsSameFloatAfterCast(value.getComplexFloatImag(), Src, Tgt)); 2936} 2937 2938void AnalyzeImplicitConversions(Sema &S, Expr *E, SourceLocation CC); 2939 2940static bool IsZero(Sema &S, Expr *E) { 2941 // Suppress cases where we are comparing against an enum constant. 2942 if (const DeclRefExpr *DR = 2943 dyn_cast<DeclRefExpr>(E->IgnoreParenImpCasts())) 2944 if (isa<EnumConstantDecl>(DR->getDecl())) 2945 return false; 2946 2947 // Suppress cases where the '0' value is expanded from a macro. 2948 if (E->getLocStart().isMacroID()) 2949 return false; 2950 2951 llvm::APSInt Value; 2952 return E->isIntegerConstantExpr(Value, S.Context) && Value == 0; 2953} 2954 2955static bool HasEnumType(Expr *E) { 2956 // Strip off implicit integral promotions. 2957 while (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(E)) { 2958 if (ICE->getCastKind() != CK_IntegralCast && 2959 ICE->getCastKind() != CK_NoOp) 2960 break; 2961 E = ICE->getSubExpr(); 2962 } 2963 2964 return E->getType()->isEnumeralType(); 2965} 2966 2967void CheckTrivialUnsignedComparison(Sema &S, BinaryOperator *E) { 2968 BinaryOperatorKind op = E->getOpcode(); 2969 if (E->isValueDependent()) 2970 return; 2971 2972 if (op == BO_LT && IsZero(S, E->getRHS())) { 2973 S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison) 2974 << "< 0" << "false" << HasEnumType(E->getLHS()) 2975 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 2976 } else if (op == BO_GE && IsZero(S, E->getRHS())) { 2977 S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison) 2978 << ">= 0" << "true" << HasEnumType(E->getLHS()) 2979 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 2980 } else if (op == BO_GT && IsZero(S, E->getLHS())) { 2981 S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison) 2982 << "0 >" << "false" << HasEnumType(E->getRHS()) 2983 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 2984 } else if (op == BO_LE && IsZero(S, E->getLHS())) { 2985 S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison) 2986 << "0 <=" << "true" << HasEnumType(E->getRHS()) 2987 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 2988 } 2989} 2990 2991/// Analyze the operands of the given comparison. Implements the 2992/// fallback case from AnalyzeComparison. 2993void AnalyzeImpConvsInComparison(Sema &S, BinaryOperator *E) { 2994 AnalyzeImplicitConversions(S, E->getLHS(), E->getOperatorLoc()); 2995 AnalyzeImplicitConversions(S, E->getRHS(), E->getOperatorLoc()); 2996} 2997 2998/// \brief Implements -Wsign-compare. 2999/// 3000/// \param lex the left-hand expression 3001/// \param rex the right-hand expression 3002/// \param OpLoc the location of the joining operator 3003/// \param BinOpc binary opcode or 0 3004void AnalyzeComparison(Sema &S, BinaryOperator *E) { 3005 // The type the comparison is being performed in. 3006 QualType T = E->getLHS()->getType(); 3007 assert(S.Context.hasSameUnqualifiedType(T, E->getRHS()->getType()) 3008 && "comparison with mismatched types"); 3009 3010 // We don't do anything special if this isn't an unsigned integral 3011 // comparison: we're only interested in integral comparisons, and 3012 // signed comparisons only happen in cases we don't care to warn about. 3013 // 3014 // We also don't care about value-dependent expressions or expressions 3015 // whose result is a constant. 3016 if (!T->hasUnsignedIntegerRepresentation() 3017 || E->isValueDependent() || E->isIntegerConstantExpr(S.Context)) 3018 return AnalyzeImpConvsInComparison(S, E); 3019 3020 Expr *lex = E->getLHS()->IgnoreParenImpCasts(); 3021 Expr *rex = E->getRHS()->IgnoreParenImpCasts(); 3022 3023 // Check to see if one of the (unmodified) operands is of different 3024 // signedness. 3025 Expr *signedOperand, *unsignedOperand; 3026 if (lex->getType()->hasSignedIntegerRepresentation()) { 3027 assert(!rex->getType()->hasSignedIntegerRepresentation() && 3028 "unsigned comparison between two signed integer expressions?"); 3029 signedOperand = lex; 3030 unsignedOperand = rex; 3031 } else if (rex->getType()->hasSignedIntegerRepresentation()) { 3032 signedOperand = rex; 3033 unsignedOperand = lex; 3034 } else { 3035 CheckTrivialUnsignedComparison(S, E); 3036 return AnalyzeImpConvsInComparison(S, E); 3037 } 3038 3039 // Otherwise, calculate the effective range of the signed operand. 3040 IntRange signedRange = GetExprRange(S.Context, signedOperand); 3041 3042 // Go ahead and analyze implicit conversions in the operands. Note 3043 // that we skip the implicit conversions on both sides. 3044 AnalyzeImplicitConversions(S, lex, E->getOperatorLoc()); 3045 AnalyzeImplicitConversions(S, rex, E->getOperatorLoc()); 3046 3047 // If the signed range is non-negative, -Wsign-compare won't fire, 3048 // but we should still check for comparisons which are always true 3049 // or false. 3050 if (signedRange.NonNegative) 3051 return CheckTrivialUnsignedComparison(S, E); 3052 3053 // For (in)equality comparisons, if the unsigned operand is a 3054 // constant which cannot collide with a overflowed signed operand, 3055 // then reinterpreting the signed operand as unsigned will not 3056 // change the result of the comparison. 3057 if (E->isEqualityOp()) { 3058 unsigned comparisonWidth = S.Context.getIntWidth(T); 3059 IntRange unsignedRange = GetExprRange(S.Context, unsignedOperand); 3060 3061 // We should never be unable to prove that the unsigned operand is 3062 // non-negative. 3063 assert(unsignedRange.NonNegative && "unsigned range includes negative?"); 3064 3065 if (unsignedRange.Width < comparisonWidth) 3066 return; 3067 } 3068 3069 S.Diag(E->getOperatorLoc(), diag::warn_mixed_sign_comparison) 3070 << lex->getType() << rex->getType() 3071 << lex->getSourceRange() << rex->getSourceRange(); 3072} 3073 3074/// Analyzes an attempt to assign the given value to a bitfield. 3075/// 3076/// Returns true if there was something fishy about the attempt. 3077bool AnalyzeBitFieldAssignment(Sema &S, FieldDecl *Bitfield, Expr *Init, 3078 SourceLocation InitLoc) { 3079 assert(Bitfield->isBitField()); 3080 if (Bitfield->isInvalidDecl()) 3081 return false; 3082 3083 // White-list bool bitfields. 3084 if (Bitfield->getType()->isBooleanType()) 3085 return false; 3086 3087 // Ignore value- or type-dependent expressions. 3088 if (Bitfield->getBitWidth()->isValueDependent() || 3089 Bitfield->getBitWidth()->isTypeDependent() || 3090 Init->isValueDependent() || 3091 Init->isTypeDependent()) 3092 return false; 3093 3094 Expr *OriginalInit = Init->IgnoreParenImpCasts(); 3095 3096 llvm::APSInt Width(32); 3097 Expr::EvalResult InitValue; 3098 if (!Bitfield->getBitWidth()->isIntegerConstantExpr(Width, S.Context) || 3099 !OriginalInit->Evaluate(InitValue, S.Context) || 3100 !InitValue.Val.isInt()) 3101 return false; 3102 3103 const llvm::APSInt &Value = InitValue.Val.getInt(); 3104 unsigned OriginalWidth = Value.getBitWidth(); 3105 unsigned FieldWidth = Width.getZExtValue(); 3106 3107 if (OriginalWidth <= FieldWidth) 3108 return false; 3109 3110 llvm::APSInt TruncatedValue = Value.trunc(FieldWidth); 3111 3112 // It's fairly common to write values into signed bitfields 3113 // that, if sign-extended, would end up becoming a different 3114 // value. We don't want to warn about that. 3115 if (Value.isSigned() && Value.isNegative()) 3116 TruncatedValue = TruncatedValue.sext(OriginalWidth); 3117 else 3118 TruncatedValue = TruncatedValue.zext(OriginalWidth); 3119 3120 if (Value == TruncatedValue) 3121 return false; 3122 3123 std::string PrettyValue = Value.toString(10); 3124 std::string PrettyTrunc = TruncatedValue.toString(10); 3125 3126 S.Diag(InitLoc, diag::warn_impcast_bitfield_precision_constant) 3127 << PrettyValue << PrettyTrunc << OriginalInit->getType() 3128 << Init->getSourceRange(); 3129 3130 return true; 3131} 3132 3133/// Analyze the given simple or compound assignment for warning-worthy 3134/// operations. 3135void AnalyzeAssignment(Sema &S, BinaryOperator *E) { 3136 // Just recurse on the LHS. 3137 AnalyzeImplicitConversions(S, E->getLHS(), E->getOperatorLoc()); 3138 3139 // We want to recurse on the RHS as normal unless we're assigning to 3140 // a bitfield. 3141 if (FieldDecl *Bitfield = E->getLHS()->getBitField()) { 3142 if (AnalyzeBitFieldAssignment(S, Bitfield, E->getRHS(), 3143 E->getOperatorLoc())) { 3144 // Recurse, ignoring any implicit conversions on the RHS. 3145 return AnalyzeImplicitConversions(S, E->getRHS()->IgnoreParenImpCasts(), 3146 E->getOperatorLoc()); 3147 } 3148 } 3149 3150 AnalyzeImplicitConversions(S, E->getRHS(), E->getOperatorLoc()); 3151} 3152 3153/// Diagnose an implicit cast; purely a helper for CheckImplicitConversion. 3154void DiagnoseImpCast(Sema &S, Expr *E, QualType SourceType, QualType T, 3155 SourceLocation CContext, unsigned diag) { 3156 S.Diag(E->getExprLoc(), diag) 3157 << SourceType << T << E->getSourceRange() << SourceRange(CContext); 3158} 3159 3160/// Diagnose an implicit cast; purely a helper for CheckImplicitConversion. 3161void DiagnoseImpCast(Sema &S, Expr *E, QualType T, SourceLocation CContext, 3162 unsigned diag) { 3163 DiagnoseImpCast(S, E, E->getType(), T, CContext, diag); 3164} 3165 3166/// Diagnose an implicit cast from a literal expression. Also attemps to supply 3167/// fixit hints when the cast wouldn't lose information to simply write the 3168/// expression with the expected type. 3169void DiagnoseFloatingLiteralImpCast(Sema &S, FloatingLiteral *FL, QualType T, 3170 SourceLocation CContext) { 3171 // Emit the primary warning first, then try to emit a fixit hint note if 3172 // reasonable. 3173 S.Diag(FL->getExprLoc(), diag::warn_impcast_literal_float_to_integer) 3174 << FL->getType() << T << FL->getSourceRange() << SourceRange(CContext); 3175 3176 const llvm::APFloat &Value = FL->getValue(); 3177 3178 // Don't attempt to fix PPC double double literals. 3179 if (&Value.getSemantics() == &llvm::APFloat::PPCDoubleDouble) 3180 return; 3181 3182 // Try to convert this exactly to an integer. 3183 bool isExact = false; 3184 llvm::APSInt IntegerValue(S.Context.getIntWidth(T), 3185 T->hasUnsignedIntegerRepresentation()); 3186 if (Value.convertToInteger(IntegerValue, 3187 llvm::APFloat::rmTowardZero, &isExact) 3188 != llvm::APFloat::opOK || !isExact) 3189 return; 3190 3191 std::string LiteralValue = IntegerValue.toString(10); 3192 S.Diag(FL->getExprLoc(), diag::note_fix_integral_float_as_integer) 3193 << FixItHint::CreateReplacement(FL->getSourceRange(), LiteralValue); 3194} 3195 3196std::string PrettyPrintInRange(const llvm::APSInt &Value, IntRange Range) { 3197 if (!Range.Width) return "0"; 3198 3199 llvm::APSInt ValueInRange = Value; 3200 ValueInRange.setIsSigned(!Range.NonNegative); 3201 ValueInRange = ValueInRange.trunc(Range.Width); 3202 return ValueInRange.toString(10); 3203} 3204 3205static bool isFromSystemMacro(Sema &S, SourceLocation loc) { 3206 SourceManager &smgr = S.Context.getSourceManager(); 3207 return loc.isMacroID() && smgr.isInSystemHeader(smgr.getSpellingLoc(loc)); 3208} 3209 3210void CheckImplicitConversion(Sema &S, Expr *E, QualType T, 3211 SourceLocation CC, bool *ICContext = 0) { 3212 if (E->isTypeDependent() || E->isValueDependent()) return; 3213 3214 const Type *Source = S.Context.getCanonicalType(E->getType()).getTypePtr(); 3215 const Type *Target = S.Context.getCanonicalType(T).getTypePtr(); 3216 if (Source == Target) return; 3217 if (Target->isDependentType()) return; 3218 3219 // If the conversion context location is invalid don't complain. We also 3220 // don't want to emit a warning if the issue occurs from the expansion of 3221 // a system macro. The problem is that 'getSpellingLoc()' is slow, so we 3222 // delay this check as long as possible. Once we detect we are in that 3223 // scenario, we just return. 3224 if (CC.isInvalid()) 3225 return; 3226 3227 // Never diagnose implicit casts to bool. 3228 if (Target->isSpecificBuiltinType(BuiltinType::Bool)) 3229 return; 3230 3231 // Strip vector types. 3232 if (isa<VectorType>(Source)) { 3233 if (!isa<VectorType>(Target)) { 3234 if (isFromSystemMacro(S, CC)) 3235 return; 3236 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_vector_scalar); 3237 } 3238 3239 // If the vector cast is cast between two vectors of the same size, it is 3240 // a bitcast, not a conversion. 3241 if (S.Context.getTypeSize(Source) == S.Context.getTypeSize(Target)) 3242 return; 3243 3244 Source = cast<VectorType>(Source)->getElementType().getTypePtr(); 3245 Target = cast<VectorType>(Target)->getElementType().getTypePtr(); 3246 } 3247 3248 // Strip complex types. 3249 if (isa<ComplexType>(Source)) { 3250 if (!isa<ComplexType>(Target)) { 3251 if (isFromSystemMacro(S, CC)) 3252 return; 3253 3254 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_complex_scalar); 3255 } 3256 3257 Source = cast<ComplexType>(Source)->getElementType().getTypePtr(); 3258 Target = cast<ComplexType>(Target)->getElementType().getTypePtr(); 3259 } 3260 3261 const BuiltinType *SourceBT = dyn_cast<BuiltinType>(Source); 3262 const BuiltinType *TargetBT = dyn_cast<BuiltinType>(Target); 3263 3264 // If the source is floating point... 3265 if (SourceBT && SourceBT->isFloatingPoint()) { 3266 // ...and the target is floating point... 3267 if (TargetBT && TargetBT->isFloatingPoint()) { 3268 // ...then warn if we're dropping FP rank. 3269 3270 // Builtin FP kinds are ordered by increasing FP rank. 3271 if (SourceBT->getKind() > TargetBT->getKind()) { 3272 // Don't warn about float constants that are precisely 3273 // representable in the target type. 3274 Expr::EvalResult result; 3275 if (E->Evaluate(result, S.Context)) { 3276 // Value might be a float, a float vector, or a float complex. 3277 if (IsSameFloatAfterCast(result.Val, 3278 S.Context.getFloatTypeSemantics(QualType(TargetBT, 0)), 3279 S.Context.getFloatTypeSemantics(QualType(SourceBT, 0)))) 3280 return; 3281 } 3282 3283 if (isFromSystemMacro(S, CC)) 3284 return; 3285 3286 DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_float_precision); 3287 } 3288 return; 3289 } 3290 3291 // If the target is integral, always warn. 3292 if ((TargetBT && TargetBT->isInteger())) { 3293 if (isFromSystemMacro(S, CC)) 3294 return; 3295 3296 Expr *InnerE = E->IgnoreParenImpCasts(); 3297 // We also want to warn on, e.g., "int i = -1.234" 3298 if (UnaryOperator *UOp = dyn_cast<UnaryOperator>(InnerE)) 3299 if (UOp->getOpcode() == UO_Minus || UOp->getOpcode() == UO_Plus) 3300 InnerE = UOp->getSubExpr()->IgnoreParenImpCasts(); 3301 3302 if (FloatingLiteral *FL = dyn_cast<FloatingLiteral>(InnerE)) { 3303 DiagnoseFloatingLiteralImpCast(S, FL, T, CC); 3304 } else { 3305 DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_float_integer); 3306 } 3307 } 3308 3309 return; 3310 } 3311 3312 if (!Source->isIntegerType() || !Target->isIntegerType()) 3313 return; 3314 3315 if ((E->isNullPointerConstant(S.Context, Expr::NPC_ValueDependentIsNotNull) 3316 == Expr::NPCK_GNUNull) && Target->isIntegerType()) { 3317 S.Diag(E->getExprLoc(), diag::warn_impcast_null_pointer_to_integer) 3318 << E->getSourceRange() << clang::SourceRange(CC); 3319 return; 3320 } 3321 3322 IntRange SourceRange = GetExprRange(S.Context, E); 3323 IntRange TargetRange = IntRange::forTargetOfCanonicalType(S.Context, Target); 3324 3325 if (SourceRange.Width > TargetRange.Width) { 3326 // If the source is a constant, use a default-on diagnostic. 3327 // TODO: this should happen for bitfield stores, too. 3328 llvm::APSInt Value(32); 3329 if (E->isIntegerConstantExpr(Value, S.Context)) { 3330 if (isFromSystemMacro(S, CC)) 3331 return; 3332 3333 std::string PrettySourceValue = Value.toString(10); 3334 std::string PrettyTargetValue = PrettyPrintInRange(Value, TargetRange); 3335 3336 S.Diag(E->getExprLoc(), diag::warn_impcast_integer_precision_constant) 3337 << PrettySourceValue << PrettyTargetValue 3338 << E->getType() << T << E->getSourceRange() << clang::SourceRange(CC); 3339 return; 3340 } 3341 3342 // People want to build with -Wshorten-64-to-32 and not -Wconversion. 3343 if (isFromSystemMacro(S, CC)) 3344 return; 3345 3346 if (SourceRange.Width == 64 && TargetRange.Width == 32) 3347 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_integer_64_32); 3348 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_integer_precision); 3349 } 3350 3351 if ((TargetRange.NonNegative && !SourceRange.NonNegative) || 3352 (!TargetRange.NonNegative && SourceRange.NonNegative && 3353 SourceRange.Width == TargetRange.Width)) { 3354 3355 if (isFromSystemMacro(S, CC)) 3356 return; 3357 3358 unsigned DiagID = diag::warn_impcast_integer_sign; 3359 3360 // Traditionally, gcc has warned about this under -Wsign-compare. 3361 // We also want to warn about it in -Wconversion. 3362 // So if -Wconversion is off, use a completely identical diagnostic 3363 // in the sign-compare group. 3364 // The conditional-checking code will 3365 if (ICContext) { 3366 DiagID = diag::warn_impcast_integer_sign_conditional; 3367 *ICContext = true; 3368 } 3369 3370 return DiagnoseImpCast(S, E, T, CC, DiagID); 3371 } 3372 3373 // Diagnose conversions between different enumeration types. 3374 // In C, we pretend that the type of an EnumConstantDecl is its enumeration 3375 // type, to give us better diagnostics. 3376 QualType SourceType = E->getType(); 3377 if (!S.getLangOptions().CPlusPlus) { 3378 if (DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(E)) 3379 if (EnumConstantDecl *ECD = dyn_cast<EnumConstantDecl>(DRE->getDecl())) { 3380 EnumDecl *Enum = cast<EnumDecl>(ECD->getDeclContext()); 3381 SourceType = S.Context.getTypeDeclType(Enum); 3382 Source = S.Context.getCanonicalType(SourceType).getTypePtr(); 3383 } 3384 } 3385 3386 if (const EnumType *SourceEnum = Source->getAs<EnumType>()) 3387 if (const EnumType *TargetEnum = Target->getAs<EnumType>()) 3388 if ((SourceEnum->getDecl()->getIdentifier() || 3389 SourceEnum->getDecl()->getTypedefNameForAnonDecl()) && 3390 (TargetEnum->getDecl()->getIdentifier() || 3391 TargetEnum->getDecl()->getTypedefNameForAnonDecl()) && 3392 SourceEnum != TargetEnum) { 3393 if (isFromSystemMacro(S, CC)) 3394 return; 3395 3396 return DiagnoseImpCast(S, E, SourceType, T, CC, 3397 diag::warn_impcast_different_enum_types); 3398 } 3399 3400 return; 3401} 3402 3403void CheckConditionalOperator(Sema &S, ConditionalOperator *E, QualType T); 3404 3405void CheckConditionalOperand(Sema &S, Expr *E, QualType T, 3406 SourceLocation CC, bool &ICContext) { 3407 E = E->IgnoreParenImpCasts(); 3408 3409 if (isa<ConditionalOperator>(E)) 3410 return CheckConditionalOperator(S, cast<ConditionalOperator>(E), T); 3411 3412 AnalyzeImplicitConversions(S, E, CC); 3413 if (E->getType() != T) 3414 return CheckImplicitConversion(S, E, T, CC, &ICContext); 3415 return; 3416} 3417 3418void CheckConditionalOperator(Sema &S, ConditionalOperator *E, QualType T) { 3419 SourceLocation CC = E->getQuestionLoc(); 3420 3421 AnalyzeImplicitConversions(S, E->getCond(), CC); 3422 3423 bool Suspicious = false; 3424 CheckConditionalOperand(S, E->getTrueExpr(), T, CC, Suspicious); 3425 CheckConditionalOperand(S, E->getFalseExpr(), T, CC, Suspicious); 3426 3427 // If -Wconversion would have warned about either of the candidates 3428 // for a signedness conversion to the context type... 3429 if (!Suspicious) return; 3430 3431 // ...but it's currently ignored... 3432 if (S.Diags.getDiagnosticLevel(diag::warn_impcast_integer_sign_conditional, 3433 CC)) 3434 return; 3435 3436 // ...then check whether it would have warned about either of the 3437 // candidates for a signedness conversion to the condition type. 3438 if (E->getType() == T) return; 3439 3440 Suspicious = false; 3441 CheckImplicitConversion(S, E->getTrueExpr()->IgnoreParenImpCasts(), 3442 E->getType(), CC, &Suspicious); 3443 if (!Suspicious) 3444 CheckImplicitConversion(S, E->getFalseExpr()->IgnoreParenImpCasts(), 3445 E->getType(), CC, &Suspicious); 3446} 3447 3448/// AnalyzeImplicitConversions - Find and report any interesting 3449/// implicit conversions in the given expression. There are a couple 3450/// of competing diagnostics here, -Wconversion and -Wsign-compare. 3451void AnalyzeImplicitConversions(Sema &S, Expr *OrigE, SourceLocation CC) { 3452 QualType T = OrigE->getType(); 3453 Expr *E = OrigE->IgnoreParenImpCasts(); 3454 3455 // For conditional operators, we analyze the arguments as if they 3456 // were being fed directly into the output. 3457 if (isa<ConditionalOperator>(E)) { 3458 ConditionalOperator *CO = cast<ConditionalOperator>(E); 3459 CheckConditionalOperator(S, CO, T); 3460 return; 3461 } 3462 3463 // Go ahead and check any implicit conversions we might have skipped. 3464 // The non-canonical typecheck is just an optimization; 3465 // CheckImplicitConversion will filter out dead implicit conversions. 3466 if (E->getType() != T) 3467 CheckImplicitConversion(S, E, T, CC); 3468 3469 // Now continue drilling into this expression. 3470 3471 // Skip past explicit casts. 3472 if (isa<ExplicitCastExpr>(E)) { 3473 E = cast<ExplicitCastExpr>(E)->getSubExpr()->IgnoreParenImpCasts(); 3474 return AnalyzeImplicitConversions(S, E, CC); 3475 } 3476 3477 if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) { 3478 // Do a somewhat different check with comparison operators. 3479 if (BO->isComparisonOp()) 3480 return AnalyzeComparison(S, BO); 3481 3482 // And with assignments and compound assignments. 3483 if (BO->isAssignmentOp()) 3484 return AnalyzeAssignment(S, BO); 3485 } 3486 3487 // These break the otherwise-useful invariant below. Fortunately, 3488 // we don't really need to recurse into them, because any internal 3489 // expressions should have been analyzed already when they were 3490 // built into statements. 3491 if (isa<StmtExpr>(E)) return; 3492 3493 // Don't descend into unevaluated contexts. 3494 if (isa<UnaryExprOrTypeTraitExpr>(E)) return; 3495 3496 // Now just recurse over the expression's children. 3497 CC = E->getExprLoc(); 3498 for (Stmt::child_range I = E->children(); I; ++I) 3499 AnalyzeImplicitConversions(S, cast<Expr>(*I), CC); 3500} 3501 3502} // end anonymous namespace 3503 3504/// Diagnoses "dangerous" implicit conversions within the given 3505/// expression (which is a full expression). Implements -Wconversion 3506/// and -Wsign-compare. 3507/// 3508/// \param CC the "context" location of the implicit conversion, i.e. 3509/// the most location of the syntactic entity requiring the implicit 3510/// conversion 3511void Sema::CheckImplicitConversions(Expr *E, SourceLocation CC) { 3512 // Don't diagnose in unevaluated contexts. 3513 if (ExprEvalContexts.back().Context == Sema::Unevaluated) 3514 return; 3515 3516 // Don't diagnose for value- or type-dependent expressions. 3517 if (E->isTypeDependent() || E->isValueDependent()) 3518 return; 3519 3520 // Check for array bounds violations in cases where the check isn't triggered 3521 // elsewhere for other Expr types (like BinaryOperators), e.g. when an 3522 // ArraySubscriptExpr is on the RHS of a variable initialization. 3523 CheckArrayAccess(E); 3524 3525 // This is not the right CC for (e.g.) a variable initialization. 3526 AnalyzeImplicitConversions(*this, E, CC); 3527} 3528 3529void Sema::CheckBitFieldInitialization(SourceLocation InitLoc, 3530 FieldDecl *BitField, 3531 Expr *Init) { 3532 (void) AnalyzeBitFieldAssignment(*this, BitField, Init, InitLoc); 3533} 3534 3535/// CheckParmsForFunctionDef - Check that the parameters of the given 3536/// function are appropriate for the definition of a function. This 3537/// takes care of any checks that cannot be performed on the 3538/// declaration itself, e.g., that the types of each of the function 3539/// parameters are complete. 3540bool Sema::CheckParmsForFunctionDef(ParmVarDecl **P, ParmVarDecl **PEnd, 3541 bool CheckParameterNames) { 3542 bool HasInvalidParm = false; 3543 for (; P != PEnd; ++P) { 3544 ParmVarDecl *Param = *P; 3545 3546 // C99 6.7.5.3p4: the parameters in a parameter type list in a 3547 // function declarator that is part of a function definition of 3548 // that function shall not have incomplete type. 3549 // 3550 // This is also C++ [dcl.fct]p6. 3551 if (!Param->isInvalidDecl() && 3552 RequireCompleteType(Param->getLocation(), Param->getType(), 3553 diag::err_typecheck_decl_incomplete_type)) { 3554 Param->setInvalidDecl(); 3555 HasInvalidParm = true; 3556 } 3557 3558 // C99 6.9.1p5: If the declarator includes a parameter type list, the 3559 // declaration of each parameter shall include an identifier. 3560 if (CheckParameterNames && 3561 Param->getIdentifier() == 0 && 3562 !Param->isImplicit() && 3563 !getLangOptions().CPlusPlus) 3564 Diag(Param->getLocation(), diag::err_parameter_name_omitted); 3565 3566 // C99 6.7.5.3p12: 3567 // If the function declarator is not part of a definition of that 3568 // function, parameters may have incomplete type and may use the [*] 3569 // notation in their sequences of declarator specifiers to specify 3570 // variable length array types. 3571 QualType PType = Param->getOriginalType(); 3572 if (const ArrayType *AT = Context.getAsArrayType(PType)) { 3573 if (AT->getSizeModifier() == ArrayType::Star) { 3574 // FIXME: This diagnosic should point the the '[*]' if source-location 3575 // information is added for it. 3576 Diag(Param->getLocation(), diag::err_array_star_in_function_definition); 3577 } 3578 } 3579 } 3580 3581 return HasInvalidParm; 3582} 3583 3584/// CheckCastAlign - Implements -Wcast-align, which warns when a 3585/// pointer cast increases the alignment requirements. 3586void Sema::CheckCastAlign(Expr *Op, QualType T, SourceRange TRange) { 3587 // This is actually a lot of work to potentially be doing on every 3588 // cast; don't do it if we're ignoring -Wcast_align (as is the default). 3589 if (getDiagnostics().getDiagnosticLevel(diag::warn_cast_align, 3590 TRange.getBegin()) 3591 == Diagnostic::Ignored) 3592 return; 3593 3594 // Ignore dependent types. 3595 if (T->isDependentType() || Op->getType()->isDependentType()) 3596 return; 3597 3598 // Require that the destination be a pointer type. 3599 const PointerType *DestPtr = T->getAs<PointerType>(); 3600 if (!DestPtr) return; 3601 3602 // If the destination has alignment 1, we're done. 3603 QualType DestPointee = DestPtr->getPointeeType(); 3604 if (DestPointee->isIncompleteType()) return; 3605 CharUnits DestAlign = Context.getTypeAlignInChars(DestPointee); 3606 if (DestAlign.isOne()) return; 3607 3608 // Require that the source be a pointer type. 3609 const PointerType *SrcPtr = Op->getType()->getAs<PointerType>(); 3610 if (!SrcPtr) return; 3611 QualType SrcPointee = SrcPtr->getPointeeType(); 3612 3613 // Whitelist casts from cv void*. We already implicitly 3614 // whitelisted casts to cv void*, since they have alignment 1. 3615 // Also whitelist casts involving incomplete types, which implicitly 3616 // includes 'void'. 3617 if (SrcPointee->isIncompleteType()) return; 3618 3619 CharUnits SrcAlign = Context.getTypeAlignInChars(SrcPointee); 3620 if (SrcAlign >= DestAlign) return; 3621 3622 Diag(TRange.getBegin(), diag::warn_cast_align) 3623 << Op->getType() << T 3624 << static_cast<unsigned>(SrcAlign.getQuantity()) 3625 << static_cast<unsigned>(DestAlign.getQuantity()) 3626 << TRange << Op->getSourceRange(); 3627} 3628 3629static const Type* getElementType(const Expr *BaseExpr) { 3630 const Type* EltType = BaseExpr->getType().getTypePtr(); 3631 if (EltType->isAnyPointerType()) 3632 return EltType->getPointeeType().getTypePtr(); 3633 else if (EltType->isArrayType()) 3634 return EltType->getBaseElementTypeUnsafe(); 3635 return EltType; 3636} 3637 3638/// \brief Check whether this array fits the idiom of a size-one tail padded 3639/// array member of a struct. 3640/// 3641/// We avoid emitting out-of-bounds access warnings for such arrays as they are 3642/// commonly used to emulate flexible arrays in C89 code. 3643static bool IsTailPaddedMemberArray(Sema &S, llvm::APInt Size, 3644 const NamedDecl *ND) { 3645 if (Size != 1 || !ND) return false; 3646 3647 const FieldDecl *FD = dyn_cast<FieldDecl>(ND); 3648 if (!FD) return false; 3649 3650 // Don't consider sizes resulting from macro expansions or template argument 3651 // substitution to form C89 tail-padded arrays. 3652 ConstantArrayTypeLoc TL = 3653 cast<ConstantArrayTypeLoc>(FD->getTypeSourceInfo()->getTypeLoc()); 3654 const Expr *SizeExpr = dyn_cast<IntegerLiteral>(TL.getSizeExpr()); 3655 if (!SizeExpr || SizeExpr->getExprLoc().isMacroID()) 3656 return false; 3657 3658 const RecordDecl *RD = dyn_cast<RecordDecl>(FD->getDeclContext()); 3659 if (!RD || !RD->isStruct()) 3660 return false; 3661 3662 // See if this is the last field decl in the record. 3663 const Decl *D = FD; 3664 while ((D = D->getNextDeclInContext())) 3665 if (isa<FieldDecl>(D)) 3666 return false; 3667 return true; 3668} 3669 3670void Sema::CheckArrayAccess(const Expr *BaseExpr, const Expr *IndexExpr, 3671 bool isSubscript, bool AllowOnePastEnd) { 3672 const Type* EffectiveType = getElementType(BaseExpr); 3673 BaseExpr = BaseExpr->IgnoreParenCasts(); 3674 IndexExpr = IndexExpr->IgnoreParenCasts(); 3675 3676 const ConstantArrayType *ArrayTy = 3677 Context.getAsConstantArrayType(BaseExpr->getType()); 3678 if (!ArrayTy) 3679 return; 3680 3681 if (IndexExpr->isValueDependent()) 3682 return; 3683 llvm::APSInt index; 3684 if (!IndexExpr->isIntegerConstantExpr(index, Context)) 3685 return; 3686 3687 const NamedDecl *ND = NULL; 3688 if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(BaseExpr)) 3689 ND = dyn_cast<NamedDecl>(DRE->getDecl()); 3690 if (const MemberExpr *ME = dyn_cast<MemberExpr>(BaseExpr)) 3691 ND = dyn_cast<NamedDecl>(ME->getMemberDecl()); 3692 3693 if (index.isUnsigned() || !index.isNegative()) { 3694 llvm::APInt size = ArrayTy->getSize(); 3695 if (!size.isStrictlyPositive()) 3696 return; 3697 3698 const Type* BaseType = getElementType(BaseExpr); 3699 if (!isSubscript && BaseType != EffectiveType) { 3700 // Make sure we're comparing apples to apples when comparing index to size 3701 uint64_t ptrarith_typesize = Context.getTypeSize(EffectiveType); 3702 uint64_t array_typesize = Context.getTypeSize(BaseType); 3703 // Handle ptrarith_typesize being zero, such as when casting to void* 3704 if (!ptrarith_typesize) ptrarith_typesize = 1; 3705 if (ptrarith_typesize != array_typesize) { 3706 // There's a cast to a different size type involved 3707 uint64_t ratio = array_typesize / ptrarith_typesize; 3708 // TODO: Be smarter about handling cases where array_typesize is not a 3709 // multiple of ptrarith_typesize 3710 if (ptrarith_typesize * ratio == array_typesize) 3711 size *= llvm::APInt(size.getBitWidth(), ratio); 3712 } 3713 } 3714 3715 if (size.getBitWidth() > index.getBitWidth()) 3716 index = index.sext(size.getBitWidth()); 3717 else if (size.getBitWidth() < index.getBitWidth()) 3718 size = size.sext(index.getBitWidth()); 3719 3720 // For array subscripting the index must be less than size, but for pointer 3721 // arithmetic also allow the index (offset) to be equal to size since 3722 // computing the next address after the end of the array is legal and 3723 // commonly done e.g. in C++ iterators and range-based for loops. 3724 if (AllowOnePastEnd ? index.sle(size) : index.slt(size)) 3725 return; 3726 3727 // Also don't warn for arrays of size 1 which are members of some 3728 // structure. These are often used to approximate flexible arrays in C89 3729 // code. 3730 if (IsTailPaddedMemberArray(*this, size, ND)) 3731 return; 3732 3733 unsigned DiagID = diag::warn_ptr_arith_exceeds_bounds; 3734 if (isSubscript) 3735 DiagID = diag::warn_array_index_exceeds_bounds; 3736 3737 DiagRuntimeBehavior(BaseExpr->getLocStart(), BaseExpr, 3738 PDiag(DiagID) << index.toString(10, true) 3739 << size.toString(10, true) 3740 << (unsigned)size.getLimitedValue(~0U) 3741 << IndexExpr->getSourceRange()); 3742 } else { 3743 unsigned DiagID = diag::warn_array_index_precedes_bounds; 3744 if (!isSubscript) { 3745 DiagID = diag::warn_ptr_arith_precedes_bounds; 3746 if (index.isNegative()) index = -index; 3747 } 3748 3749 DiagRuntimeBehavior(BaseExpr->getLocStart(), BaseExpr, 3750 PDiag(DiagID) << index.toString(10, true) 3751 << IndexExpr->getSourceRange()); 3752 } 3753 3754 if (ND) 3755 DiagRuntimeBehavior(ND->getLocStart(), BaseExpr, 3756 PDiag(diag::note_array_index_out_of_bounds) 3757 << ND->getDeclName()); 3758} 3759 3760void Sema::CheckArrayAccess(const Expr *expr) { 3761 int AllowOnePastEnd = 0; 3762 while (expr) { 3763 expr = expr->IgnoreParenImpCasts(); 3764 switch (expr->getStmtClass()) { 3765 case Stmt::ArraySubscriptExprClass: { 3766 const ArraySubscriptExpr *ASE = cast<ArraySubscriptExpr>(expr); 3767 CheckArrayAccess(ASE->getBase(), ASE->getIdx(), true, 3768 AllowOnePastEnd > 0); 3769 return; 3770 } 3771 case Stmt::UnaryOperatorClass: { 3772 // Only unwrap the * and & unary operators 3773 const UnaryOperator *UO = cast<UnaryOperator>(expr); 3774 expr = UO->getSubExpr(); 3775 switch (UO->getOpcode()) { 3776 case UO_AddrOf: 3777 AllowOnePastEnd++; 3778 break; 3779 case UO_Deref: 3780 AllowOnePastEnd--; 3781 break; 3782 default: 3783 return; 3784 } 3785 break; 3786 } 3787 case Stmt::ConditionalOperatorClass: { 3788 const ConditionalOperator *cond = cast<ConditionalOperator>(expr); 3789 if (const Expr *lhs = cond->getLHS()) 3790 CheckArrayAccess(lhs); 3791 if (const Expr *rhs = cond->getRHS()) 3792 CheckArrayAccess(rhs); 3793 return; 3794 } 3795 default: 3796 return; 3797 } 3798 } 3799} 3800 3801//===--- CHECK: Objective-C retain cycles ----------------------------------// 3802 3803namespace { 3804 struct RetainCycleOwner { 3805 RetainCycleOwner() : Variable(0), Indirect(false) {} 3806 VarDecl *Variable; 3807 SourceRange Range; 3808 SourceLocation Loc; 3809 bool Indirect; 3810 3811 void setLocsFrom(Expr *e) { 3812 Loc = e->getExprLoc(); 3813 Range = e->getSourceRange(); 3814 } 3815 }; 3816} 3817 3818/// Consider whether capturing the given variable can possibly lead to 3819/// a retain cycle. 3820static bool considerVariable(VarDecl *var, Expr *ref, RetainCycleOwner &owner) { 3821 // In ARC, it's captured strongly iff the variable has __strong 3822 // lifetime. In MRR, it's captured strongly if the variable is 3823 // __block and has an appropriate type. 3824 if (var->getType().getObjCLifetime() != Qualifiers::OCL_Strong) 3825 return false; 3826 3827 owner.Variable = var; 3828 owner.setLocsFrom(ref); 3829 return true; 3830} 3831 3832static bool findRetainCycleOwner(Expr *e, RetainCycleOwner &owner) { 3833 while (true) { 3834 e = e->IgnoreParens(); 3835 if (CastExpr *cast = dyn_cast<CastExpr>(e)) { 3836 switch (cast->getCastKind()) { 3837 case CK_BitCast: 3838 case CK_LValueBitCast: 3839 case CK_LValueToRValue: 3840 case CK_ObjCReclaimReturnedObject: 3841 e = cast->getSubExpr(); 3842 continue; 3843 3844 case CK_GetObjCProperty: { 3845 // Bail out if this isn't a strong explicit property. 3846 const ObjCPropertyRefExpr *pre = cast->getSubExpr()->getObjCProperty(); 3847 if (pre->isImplicitProperty()) return false; 3848 ObjCPropertyDecl *property = pre->getExplicitProperty(); 3849 if (!(property->getPropertyAttributes() & 3850 (ObjCPropertyDecl::OBJC_PR_retain | 3851 ObjCPropertyDecl::OBJC_PR_copy | 3852 ObjCPropertyDecl::OBJC_PR_strong)) && 3853 !(property->getPropertyIvarDecl() && 3854 property->getPropertyIvarDecl()->getType() 3855 .getObjCLifetime() == Qualifiers::OCL_Strong)) 3856 return false; 3857 3858 owner.Indirect = true; 3859 e = const_cast<Expr*>(pre->getBase()); 3860 continue; 3861 } 3862 3863 default: 3864 return false; 3865 } 3866 } 3867 3868 if (ObjCIvarRefExpr *ref = dyn_cast<ObjCIvarRefExpr>(e)) { 3869 ObjCIvarDecl *ivar = ref->getDecl(); 3870 if (ivar->getType().getObjCLifetime() != Qualifiers::OCL_Strong) 3871 return false; 3872 3873 // Try to find a retain cycle in the base. 3874 if (!findRetainCycleOwner(ref->getBase(), owner)) 3875 return false; 3876 3877 if (ref->isFreeIvar()) owner.setLocsFrom(ref); 3878 owner.Indirect = true; 3879 return true; 3880 } 3881 3882 if (DeclRefExpr *ref = dyn_cast<DeclRefExpr>(e)) { 3883 VarDecl *var = dyn_cast<VarDecl>(ref->getDecl()); 3884 if (!var) return false; 3885 return considerVariable(var, ref, owner); 3886 } 3887 3888 if (BlockDeclRefExpr *ref = dyn_cast<BlockDeclRefExpr>(e)) { 3889 owner.Variable = ref->getDecl(); 3890 owner.setLocsFrom(ref); 3891 return true; 3892 } 3893 3894 if (MemberExpr *member = dyn_cast<MemberExpr>(e)) { 3895 if (member->isArrow()) return false; 3896 3897 // Don't count this as an indirect ownership. 3898 e = member->getBase(); 3899 continue; 3900 } 3901 3902 // Array ivars? 3903 3904 return false; 3905 } 3906} 3907 3908namespace { 3909 struct FindCaptureVisitor : EvaluatedExprVisitor<FindCaptureVisitor> { 3910 FindCaptureVisitor(ASTContext &Context, VarDecl *variable) 3911 : EvaluatedExprVisitor<FindCaptureVisitor>(Context), 3912 Variable(variable), Capturer(0) {} 3913 3914 VarDecl *Variable; 3915 Expr *Capturer; 3916 3917 void VisitDeclRefExpr(DeclRefExpr *ref) { 3918 if (ref->getDecl() == Variable && !Capturer) 3919 Capturer = ref; 3920 } 3921 3922 void VisitBlockDeclRefExpr(BlockDeclRefExpr *ref) { 3923 if (ref->getDecl() == Variable && !Capturer) 3924 Capturer = ref; 3925 } 3926 3927 void VisitObjCIvarRefExpr(ObjCIvarRefExpr *ref) { 3928 if (Capturer) return; 3929 Visit(ref->getBase()); 3930 if (Capturer && ref->isFreeIvar()) 3931 Capturer = ref; 3932 } 3933 3934 void VisitBlockExpr(BlockExpr *block) { 3935 // Look inside nested blocks 3936 if (block->getBlockDecl()->capturesVariable(Variable)) 3937 Visit(block->getBlockDecl()->getBody()); 3938 } 3939 }; 3940} 3941 3942/// Check whether the given argument is a block which captures a 3943/// variable. 3944static Expr *findCapturingExpr(Sema &S, Expr *e, RetainCycleOwner &owner) { 3945 assert(owner.Variable && owner.Loc.isValid()); 3946 3947 e = e->IgnoreParenCasts(); 3948 BlockExpr *block = dyn_cast<BlockExpr>(e); 3949 if (!block || !block->getBlockDecl()->capturesVariable(owner.Variable)) 3950 return 0; 3951 3952 FindCaptureVisitor visitor(S.Context, owner.Variable); 3953 visitor.Visit(block->getBlockDecl()->getBody()); 3954 return visitor.Capturer; 3955} 3956 3957static void diagnoseRetainCycle(Sema &S, Expr *capturer, 3958 RetainCycleOwner &owner) { 3959 assert(capturer); 3960 assert(owner.Variable && owner.Loc.isValid()); 3961 3962 S.Diag(capturer->getExprLoc(), diag::warn_arc_retain_cycle) 3963 << owner.Variable << capturer->getSourceRange(); 3964 S.Diag(owner.Loc, diag::note_arc_retain_cycle_owner) 3965 << owner.Indirect << owner.Range; 3966} 3967 3968/// Check for a keyword selector that starts with the word 'add' or 3969/// 'set'. 3970static bool isSetterLikeSelector(Selector sel) { 3971 if (sel.isUnarySelector()) return false; 3972 3973 StringRef str = sel.getNameForSlot(0); 3974 while (!str.empty() && str.front() == '_') str = str.substr(1); 3975 if (str.startswith("set") || str.startswith("add")) 3976 str = str.substr(3); 3977 else 3978 return false; 3979 3980 if (str.empty()) return true; 3981 return !islower(str.front()); 3982} 3983 3984/// Check a message send to see if it's likely to cause a retain cycle. 3985void Sema::checkRetainCycles(ObjCMessageExpr *msg) { 3986 // Only check instance methods whose selector looks like a setter. 3987 if (!msg->isInstanceMessage() || !isSetterLikeSelector(msg->getSelector())) 3988 return; 3989 3990 // Try to find a variable that the receiver is strongly owned by. 3991 RetainCycleOwner owner; 3992 if (msg->getReceiverKind() == ObjCMessageExpr::Instance) { 3993 if (!findRetainCycleOwner(msg->getInstanceReceiver(), owner)) 3994 return; 3995 } else { 3996 assert(msg->getReceiverKind() == ObjCMessageExpr::SuperInstance); 3997 owner.Variable = getCurMethodDecl()->getSelfDecl(); 3998 owner.Loc = msg->getSuperLoc(); 3999 owner.Range = msg->getSuperLoc(); 4000 } 4001 4002 // Check whether the receiver is captured by any of the arguments. 4003 for (unsigned i = 0, e = msg->getNumArgs(); i != e; ++i) 4004 if (Expr *capturer = findCapturingExpr(*this, msg->getArg(i), owner)) 4005 return diagnoseRetainCycle(*this, capturer, owner); 4006} 4007 4008/// Check a property assign to see if it's likely to cause a retain cycle. 4009void Sema::checkRetainCycles(Expr *receiver, Expr *argument) { 4010 RetainCycleOwner owner; 4011 if (!findRetainCycleOwner(receiver, owner)) 4012 return; 4013 4014 if (Expr *capturer = findCapturingExpr(*this, argument, owner)) 4015 diagnoseRetainCycle(*this, capturer, owner); 4016} 4017 4018bool Sema::checkUnsafeAssigns(SourceLocation Loc, 4019 QualType LHS, Expr *RHS) { 4020 Qualifiers::ObjCLifetime LT = LHS.getObjCLifetime(); 4021 if (LT != Qualifiers::OCL_Weak && LT != Qualifiers::OCL_ExplicitNone) 4022 return false; 4023 // strip off any implicit cast added to get to the one arc-specific 4024 while (ImplicitCastExpr *cast = dyn_cast<ImplicitCastExpr>(RHS)) { 4025 if (cast->getCastKind() == CK_ObjCConsumeObject) { 4026 Diag(Loc, diag::warn_arc_retained_assign) 4027 << (LT == Qualifiers::OCL_ExplicitNone) 4028 << RHS->getSourceRange(); 4029 return true; 4030 } 4031 RHS = cast->getSubExpr(); 4032 } 4033 return false; 4034} 4035 4036void Sema::checkUnsafeExprAssigns(SourceLocation Loc, 4037 Expr *LHS, Expr *RHS) { 4038 QualType LHSType = LHS->getType(); 4039 if (checkUnsafeAssigns(Loc, LHSType, RHS)) 4040 return; 4041 Qualifiers::ObjCLifetime LT = LHSType.getObjCLifetime(); 4042 // FIXME. Check for other life times. 4043 if (LT != Qualifiers::OCL_None) 4044 return; 4045 4046 if (ObjCPropertyRefExpr *PRE = dyn_cast<ObjCPropertyRefExpr>(LHS)) { 4047 if (PRE->isImplicitProperty()) 4048 return; 4049 const ObjCPropertyDecl *PD = PRE->getExplicitProperty(); 4050 if (!PD) 4051 return; 4052 4053 unsigned Attributes = PD->getPropertyAttributes(); 4054 if (Attributes & ObjCPropertyDecl::OBJC_PR_assign) 4055 while (ImplicitCastExpr *cast = dyn_cast<ImplicitCastExpr>(RHS)) { 4056 if (cast->getCastKind() == CK_ObjCConsumeObject) { 4057 Diag(Loc, diag::warn_arc_retained_property_assign) 4058 << RHS->getSourceRange(); 4059 return; 4060 } 4061 RHS = cast->getSubExpr(); 4062 } 4063 } 4064} 4065