SemaChecking.cpp revision 284cc8d8a90ae6558e0a4b60b7dc1ddcfd220758
1//===--- SemaChecking.cpp - Extra Semantic Checking -----------------------===// 2// 3// The LLVM Compiler Infrastructure 4// 5// This file is distributed under the University of Illinois Open Source 6// License. See LICENSE.TXT for details. 7// 8//===----------------------------------------------------------------------===// 9// 10// This file implements extra semantic analysis beyond what is enforced 11// by the C type system. 12// 13//===----------------------------------------------------------------------===// 14 15#include "clang/Sema/Sema.h" 16#include "clang/Sema/SemaInternal.h" 17#include "clang/Sema/ScopeInfo.h" 18#include "clang/Analysis/Analyses/FormatString.h" 19#include "clang/AST/ASTContext.h" 20#include "clang/AST/CharUnits.h" 21#include "clang/AST/DeclCXX.h" 22#include "clang/AST/DeclObjC.h" 23#include "clang/AST/ExprCXX.h" 24#include "clang/AST/ExprObjC.h" 25#include "clang/AST/DeclObjC.h" 26#include "clang/AST/StmtCXX.h" 27#include "clang/AST/StmtObjC.h" 28#include "clang/Lex/Preprocessor.h" 29#include "llvm/ADT/BitVector.h" 30#include "llvm/ADT/STLExtras.h" 31#include "llvm/Support/raw_ostream.h" 32#include "clang/Basic/TargetBuiltins.h" 33#include "clang/Basic/TargetInfo.h" 34#include "clang/Basic/ConvertUTF.h" 35#include <limits> 36using namespace clang; 37using namespace sema; 38 39SourceLocation Sema::getLocationOfStringLiteralByte(const StringLiteral *SL, 40 unsigned ByteNo) const { 41 return SL->getLocationOfByte(ByteNo, PP.getSourceManager(), 42 PP.getLangOptions(), PP.getTargetInfo()); 43} 44 45 46/// CheckablePrintfAttr - does a function call have a "printf" attribute 47/// and arguments that merit checking? 48bool Sema::CheckablePrintfAttr(const FormatAttr *Format, CallExpr *TheCall) { 49 if (Format->getType() == "printf") return true; 50 if (Format->getType() == "printf0") { 51 // printf0 allows null "format" string; if so don't check format/args 52 unsigned format_idx = Format->getFormatIdx() - 1; 53 // Does the index refer to the implicit object argument? 54 if (isa<CXXMemberCallExpr>(TheCall)) { 55 if (format_idx == 0) 56 return false; 57 --format_idx; 58 } 59 if (format_idx < TheCall->getNumArgs()) { 60 Expr *Format = TheCall->getArg(format_idx)->IgnoreParenCasts(); 61 if (!Format->isNullPointerConstant(Context, 62 Expr::NPC_ValueDependentIsNull)) 63 return true; 64 } 65 } 66 return false; 67} 68 69ExprResult 70Sema::CheckBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) { 71 ExprResult TheCallResult(Owned(TheCall)); 72 73 // Find out if any arguments are required to be integer constant expressions. 74 unsigned ICEArguments = 0; 75 ASTContext::GetBuiltinTypeError Error; 76 Context.GetBuiltinType(BuiltinID, Error, &ICEArguments); 77 if (Error != ASTContext::GE_None) 78 ICEArguments = 0; // Don't diagnose previously diagnosed errors. 79 80 // If any arguments are required to be ICE's, check and diagnose. 81 for (unsigned ArgNo = 0; ICEArguments != 0; ++ArgNo) { 82 // Skip arguments not required to be ICE's. 83 if ((ICEArguments & (1 << ArgNo)) == 0) continue; 84 85 llvm::APSInt Result; 86 if (SemaBuiltinConstantArg(TheCall, ArgNo, Result)) 87 return true; 88 ICEArguments &= ~(1 << ArgNo); 89 } 90 91 switch (BuiltinID) { 92 case Builtin::BI__builtin___CFStringMakeConstantString: 93 assert(TheCall->getNumArgs() == 1 && 94 "Wrong # arguments to builtin CFStringMakeConstantString"); 95 if (CheckObjCString(TheCall->getArg(0))) 96 return ExprError(); 97 break; 98 case Builtin::BI__builtin_stdarg_start: 99 case Builtin::BI__builtin_va_start: 100 if (SemaBuiltinVAStart(TheCall)) 101 return ExprError(); 102 break; 103 case Builtin::BI__builtin_isgreater: 104 case Builtin::BI__builtin_isgreaterequal: 105 case Builtin::BI__builtin_isless: 106 case Builtin::BI__builtin_islessequal: 107 case Builtin::BI__builtin_islessgreater: 108 case Builtin::BI__builtin_isunordered: 109 if (SemaBuiltinUnorderedCompare(TheCall)) 110 return ExprError(); 111 break; 112 case Builtin::BI__builtin_fpclassify: 113 if (SemaBuiltinFPClassification(TheCall, 6)) 114 return ExprError(); 115 break; 116 case Builtin::BI__builtin_isfinite: 117 case Builtin::BI__builtin_isinf: 118 case Builtin::BI__builtin_isinf_sign: 119 case Builtin::BI__builtin_isnan: 120 case Builtin::BI__builtin_isnormal: 121 if (SemaBuiltinFPClassification(TheCall, 1)) 122 return ExprError(); 123 break; 124 case Builtin::BI__builtin_shufflevector: 125 return SemaBuiltinShuffleVector(TheCall); 126 // TheCall will be freed by the smart pointer here, but that's fine, since 127 // SemaBuiltinShuffleVector guts it, but then doesn't release it. 128 case Builtin::BI__builtin_prefetch: 129 if (SemaBuiltinPrefetch(TheCall)) 130 return ExprError(); 131 break; 132 case Builtin::BI__builtin_object_size: 133 if (SemaBuiltinObjectSize(TheCall)) 134 return ExprError(); 135 break; 136 case Builtin::BI__builtin_longjmp: 137 if (SemaBuiltinLongjmp(TheCall)) 138 return ExprError(); 139 break; 140 case Builtin::BI__builtin_constant_p: 141 if (TheCall->getNumArgs() == 0) 142 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 143 << 0 /*function call*/ << 1 << 0 << TheCall->getSourceRange(); 144 if (TheCall->getNumArgs() > 1) 145 return Diag(TheCall->getArg(1)->getLocStart(), 146 diag::err_typecheck_call_too_many_args) 147 << 0 /*function call*/ << 1 << TheCall->getNumArgs() 148 << TheCall->getArg(1)->getSourceRange(); 149 break; 150 case Builtin::BI__sync_fetch_and_add: 151 case Builtin::BI__sync_fetch_and_sub: 152 case Builtin::BI__sync_fetch_and_or: 153 case Builtin::BI__sync_fetch_and_and: 154 case Builtin::BI__sync_fetch_and_xor: 155 case Builtin::BI__sync_add_and_fetch: 156 case Builtin::BI__sync_sub_and_fetch: 157 case Builtin::BI__sync_and_and_fetch: 158 case Builtin::BI__sync_or_and_fetch: 159 case Builtin::BI__sync_xor_and_fetch: 160 case Builtin::BI__sync_val_compare_and_swap: 161 case Builtin::BI__sync_bool_compare_and_swap: 162 case Builtin::BI__sync_lock_test_and_set: 163 case Builtin::BI__sync_lock_release: 164 return SemaBuiltinAtomicOverloaded(move(TheCallResult)); 165 } 166 167 // Since the target specific builtins for each arch overlap, only check those 168 // of the arch we are compiling for. 169 if (BuiltinID >= Builtin::FirstTSBuiltin) { 170 switch (Context.Target.getTriple().getArch()) { 171 case llvm::Triple::arm: 172 case llvm::Triple::thumb: 173 if (CheckARMBuiltinFunctionCall(BuiltinID, TheCall)) 174 return ExprError(); 175 break; 176 default: 177 break; 178 } 179 } 180 181 return move(TheCallResult); 182} 183 184// Get the valid immediate range for the specified NEON type code. 185static unsigned RFT(unsigned t, bool shift = false) { 186 bool quad = t & 0x10; 187 188 switch (t & 0x7) { 189 case 0: // i8 190 return shift ? 7 : (8 << (int)quad) - 1; 191 case 1: // i16 192 return shift ? 15 : (4 << (int)quad) - 1; 193 case 2: // i32 194 return shift ? 31 : (2 << (int)quad) - 1; 195 case 3: // i64 196 return shift ? 63 : (1 << (int)quad) - 1; 197 case 4: // f32 198 assert(!shift && "cannot shift float types!"); 199 return (2 << (int)quad) - 1; 200 case 5: // poly8 201 return shift ? 7 : (8 << (int)quad) - 1; 202 case 6: // poly16 203 return shift ? 15 : (4 << (int)quad) - 1; 204 case 7: // float16 205 assert(!shift && "cannot shift float types!"); 206 return (4 << (int)quad) - 1; 207 } 208 return 0; 209} 210 211bool Sema::CheckARMBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) { 212 llvm::APSInt Result; 213 214 unsigned mask = 0; 215 unsigned TV = 0; 216 switch (BuiltinID) { 217#define GET_NEON_OVERLOAD_CHECK 218#include "clang/Basic/arm_neon.inc" 219#undef GET_NEON_OVERLOAD_CHECK 220 } 221 222 // For NEON intrinsics which are overloaded on vector element type, validate 223 // the immediate which specifies which variant to emit. 224 if (mask) { 225 unsigned ArgNo = TheCall->getNumArgs()-1; 226 if (SemaBuiltinConstantArg(TheCall, ArgNo, Result)) 227 return true; 228 229 TV = Result.getLimitedValue(32); 230 if ((TV > 31) || (mask & (1 << TV)) == 0) 231 return Diag(TheCall->getLocStart(), diag::err_invalid_neon_type_code) 232 << TheCall->getArg(ArgNo)->getSourceRange(); 233 } 234 235 // For NEON intrinsics which take an immediate value as part of the 236 // instruction, range check them here. 237 unsigned i = 0, l = 0, u = 0; 238 switch (BuiltinID) { 239 default: return false; 240 case ARM::BI__builtin_arm_ssat: i = 1; l = 1; u = 31; break; 241 case ARM::BI__builtin_arm_usat: i = 1; u = 31; break; 242 case ARM::BI__builtin_arm_vcvtr_f: 243 case ARM::BI__builtin_arm_vcvtr_d: i = 1; u = 1; break; 244#define GET_NEON_IMMEDIATE_CHECK 245#include "clang/Basic/arm_neon.inc" 246#undef GET_NEON_IMMEDIATE_CHECK 247 }; 248 249 // Check that the immediate argument is actually a constant. 250 if (SemaBuiltinConstantArg(TheCall, i, Result)) 251 return true; 252 253 // Range check against the upper/lower values for this isntruction. 254 unsigned Val = Result.getZExtValue(); 255 if (Val < l || Val > (u + l)) 256 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 257 << l << u+l << TheCall->getArg(i)->getSourceRange(); 258 259 // FIXME: VFP Intrinsics should error if VFP not present. 260 return false; 261} 262 263/// CheckFunctionCall - Check a direct function call for various correctness 264/// and safety properties not strictly enforced by the C type system. 265bool Sema::CheckFunctionCall(FunctionDecl *FDecl, CallExpr *TheCall) { 266 // Get the IdentifierInfo* for the called function. 267 IdentifierInfo *FnInfo = FDecl->getIdentifier(); 268 269 // None of the checks below are needed for functions that don't have 270 // simple names (e.g., C++ conversion functions). 271 if (!FnInfo) 272 return false; 273 274 // FIXME: This mechanism should be abstracted to be less fragile and 275 // more efficient. For example, just map function ids to custom 276 // handlers. 277 278 // Printf and scanf checking. 279 for (specific_attr_iterator<FormatAttr> 280 i = FDecl->specific_attr_begin<FormatAttr>(), 281 e = FDecl->specific_attr_end<FormatAttr>(); i != e ; ++i) { 282 283 const FormatAttr *Format = *i; 284 const bool b = Format->getType() == "scanf"; 285 if (b || CheckablePrintfAttr(Format, TheCall)) { 286 bool HasVAListArg = Format->getFirstArg() == 0; 287 CheckPrintfScanfArguments(TheCall, HasVAListArg, 288 Format->getFormatIdx() - 1, 289 HasVAListArg ? 0 : Format->getFirstArg() - 1, 290 !b); 291 } 292 } 293 294 for (specific_attr_iterator<NonNullAttr> 295 i = FDecl->specific_attr_begin<NonNullAttr>(), 296 e = FDecl->specific_attr_end<NonNullAttr>(); i != e; ++i) { 297 CheckNonNullArguments(*i, TheCall); 298 } 299 300 return false; 301} 302 303bool Sema::CheckBlockCall(NamedDecl *NDecl, CallExpr *TheCall) { 304 // Printf checking. 305 const FormatAttr *Format = NDecl->getAttr<FormatAttr>(); 306 if (!Format) 307 return false; 308 309 const VarDecl *V = dyn_cast<VarDecl>(NDecl); 310 if (!V) 311 return false; 312 313 QualType Ty = V->getType(); 314 if (!Ty->isBlockPointerType()) 315 return false; 316 317 const bool b = Format->getType() == "scanf"; 318 if (!b && !CheckablePrintfAttr(Format, TheCall)) 319 return false; 320 321 bool HasVAListArg = Format->getFirstArg() == 0; 322 CheckPrintfScanfArguments(TheCall, HasVAListArg, Format->getFormatIdx() - 1, 323 HasVAListArg ? 0 : Format->getFirstArg() - 1, !b); 324 325 return false; 326} 327 328/// SemaBuiltinAtomicOverloaded - We have a call to a function like 329/// __sync_fetch_and_add, which is an overloaded function based on the pointer 330/// type of its first argument. The main ActOnCallExpr routines have already 331/// promoted the types of arguments because all of these calls are prototyped as 332/// void(...). 333/// 334/// This function goes through and does final semantic checking for these 335/// builtins, 336ExprResult 337Sema::SemaBuiltinAtomicOverloaded(ExprResult TheCallResult) { 338 CallExpr *TheCall = (CallExpr *)TheCallResult.get(); 339 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 340 FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl()); 341 342 // Ensure that we have at least one argument to do type inference from. 343 if (TheCall->getNumArgs() < 1) { 344 Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args_at_least) 345 << 0 << 1 << TheCall->getNumArgs() 346 << TheCall->getCallee()->getSourceRange(); 347 return ExprError(); 348 } 349 350 // Inspect the first argument of the atomic builtin. This should always be 351 // a pointer type, whose element is an integral scalar or pointer type. 352 // Because it is a pointer type, we don't have to worry about any implicit 353 // casts here. 354 // FIXME: We don't allow floating point scalars as input. 355 Expr *FirstArg = TheCall->getArg(0); 356 if (!FirstArg->getType()->isPointerType()) { 357 Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer) 358 << FirstArg->getType() << FirstArg->getSourceRange(); 359 return ExprError(); 360 } 361 362 QualType ValType = 363 FirstArg->getType()->getAs<PointerType>()->getPointeeType(); 364 if (!ValType->isIntegerType() && !ValType->isAnyPointerType() && 365 !ValType->isBlockPointerType()) { 366 Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer_intptr) 367 << FirstArg->getType() << FirstArg->getSourceRange(); 368 return ExprError(); 369 } 370 371 // The majority of builtins return a value, but a few have special return 372 // types, so allow them to override appropriately below. 373 QualType ResultType = ValType; 374 375 // We need to figure out which concrete builtin this maps onto. For example, 376 // __sync_fetch_and_add with a 2 byte object turns into 377 // __sync_fetch_and_add_2. 378#define BUILTIN_ROW(x) \ 379 { Builtin::BI##x##_1, Builtin::BI##x##_2, Builtin::BI##x##_4, \ 380 Builtin::BI##x##_8, Builtin::BI##x##_16 } 381 382 static const unsigned BuiltinIndices[][5] = { 383 BUILTIN_ROW(__sync_fetch_and_add), 384 BUILTIN_ROW(__sync_fetch_and_sub), 385 BUILTIN_ROW(__sync_fetch_and_or), 386 BUILTIN_ROW(__sync_fetch_and_and), 387 BUILTIN_ROW(__sync_fetch_and_xor), 388 389 BUILTIN_ROW(__sync_add_and_fetch), 390 BUILTIN_ROW(__sync_sub_and_fetch), 391 BUILTIN_ROW(__sync_and_and_fetch), 392 BUILTIN_ROW(__sync_or_and_fetch), 393 BUILTIN_ROW(__sync_xor_and_fetch), 394 395 BUILTIN_ROW(__sync_val_compare_and_swap), 396 BUILTIN_ROW(__sync_bool_compare_and_swap), 397 BUILTIN_ROW(__sync_lock_test_and_set), 398 BUILTIN_ROW(__sync_lock_release) 399 }; 400#undef BUILTIN_ROW 401 402 // Determine the index of the size. 403 unsigned SizeIndex; 404 switch (Context.getTypeSizeInChars(ValType).getQuantity()) { 405 case 1: SizeIndex = 0; break; 406 case 2: SizeIndex = 1; break; 407 case 4: SizeIndex = 2; break; 408 case 8: SizeIndex = 3; break; 409 case 16: SizeIndex = 4; break; 410 default: 411 Diag(DRE->getLocStart(), diag::err_atomic_builtin_pointer_size) 412 << FirstArg->getType() << FirstArg->getSourceRange(); 413 return ExprError(); 414 } 415 416 // Each of these builtins has one pointer argument, followed by some number of 417 // values (0, 1 or 2) followed by a potentially empty varags list of stuff 418 // that we ignore. Find out which row of BuiltinIndices to read from as well 419 // as the number of fixed args. 420 unsigned BuiltinID = FDecl->getBuiltinID(); 421 unsigned BuiltinIndex, NumFixed = 1; 422 switch (BuiltinID) { 423 default: assert(0 && "Unknown overloaded atomic builtin!"); 424 case Builtin::BI__sync_fetch_and_add: BuiltinIndex = 0; break; 425 case Builtin::BI__sync_fetch_and_sub: BuiltinIndex = 1; break; 426 case Builtin::BI__sync_fetch_and_or: BuiltinIndex = 2; break; 427 case Builtin::BI__sync_fetch_and_and: BuiltinIndex = 3; break; 428 case Builtin::BI__sync_fetch_and_xor: BuiltinIndex = 4; break; 429 430 case Builtin::BI__sync_add_and_fetch: BuiltinIndex = 5; break; 431 case Builtin::BI__sync_sub_and_fetch: BuiltinIndex = 6; break; 432 case Builtin::BI__sync_and_and_fetch: BuiltinIndex = 7; break; 433 case Builtin::BI__sync_or_and_fetch: BuiltinIndex = 8; break; 434 case Builtin::BI__sync_xor_and_fetch: BuiltinIndex = 9; break; 435 436 case Builtin::BI__sync_val_compare_and_swap: 437 BuiltinIndex = 10; 438 NumFixed = 2; 439 break; 440 case Builtin::BI__sync_bool_compare_and_swap: 441 BuiltinIndex = 11; 442 NumFixed = 2; 443 ResultType = Context.BoolTy; 444 break; 445 case Builtin::BI__sync_lock_test_and_set: BuiltinIndex = 12; break; 446 case Builtin::BI__sync_lock_release: 447 BuiltinIndex = 13; 448 NumFixed = 0; 449 ResultType = Context.VoidTy; 450 break; 451 } 452 453 // Now that we know how many fixed arguments we expect, first check that we 454 // have at least that many. 455 if (TheCall->getNumArgs() < 1+NumFixed) { 456 Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args_at_least) 457 << 0 << 1+NumFixed << TheCall->getNumArgs() 458 << TheCall->getCallee()->getSourceRange(); 459 return ExprError(); 460 } 461 462 // Get the decl for the concrete builtin from this, we can tell what the 463 // concrete integer type we should convert to is. 464 unsigned NewBuiltinID = BuiltinIndices[BuiltinIndex][SizeIndex]; 465 const char *NewBuiltinName = Context.BuiltinInfo.GetName(NewBuiltinID); 466 IdentifierInfo *NewBuiltinII = PP.getIdentifierInfo(NewBuiltinName); 467 FunctionDecl *NewBuiltinDecl = 468 cast<FunctionDecl>(LazilyCreateBuiltin(NewBuiltinII, NewBuiltinID, 469 TUScope, false, DRE->getLocStart())); 470 471 // The first argument --- the pointer --- has a fixed type; we 472 // deduce the types of the rest of the arguments accordingly. Walk 473 // the remaining arguments, converting them to the deduced value type. 474 for (unsigned i = 0; i != NumFixed; ++i) { 475 Expr *Arg = TheCall->getArg(i+1); 476 477 // If the argument is an implicit cast, then there was a promotion due to 478 // "...", just remove it now. 479 if (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(Arg)) { 480 Arg = ICE->getSubExpr(); 481 ICE->setSubExpr(0); 482 TheCall->setArg(i+1, Arg); 483 } 484 485 // GCC does an implicit conversion to the pointer or integer ValType. This 486 // can fail in some cases (1i -> int**), check for this error case now. 487 CastKind Kind = CK_Invalid; 488 ExprValueKind VK = VK_RValue; 489 CXXCastPath BasePath; 490 if (CheckCastTypes(Arg->getSourceRange(), ValType, Arg, Kind, VK, BasePath)) 491 return ExprError(); 492 493 // Okay, we have something that *can* be converted to the right type. Check 494 // to see if there is a potentially weird extension going on here. This can 495 // happen when you do an atomic operation on something like an char* and 496 // pass in 42. The 42 gets converted to char. This is even more strange 497 // for things like 45.123 -> char, etc. 498 // FIXME: Do this check. 499 ImpCastExprToType(Arg, ValType, Kind, VK, &BasePath); 500 TheCall->setArg(i+1, Arg); 501 } 502 503 // Switch the DeclRefExpr to refer to the new decl. 504 DRE->setDecl(NewBuiltinDecl); 505 DRE->setType(NewBuiltinDecl->getType()); 506 507 // Set the callee in the CallExpr. 508 // FIXME: This leaks the original parens and implicit casts. 509 Expr *PromotedCall = DRE; 510 UsualUnaryConversions(PromotedCall); 511 TheCall->setCallee(PromotedCall); 512 513 // Change the result type of the call to match the original value type. This 514 // is arbitrary, but the codegen for these builtins ins design to handle it 515 // gracefully. 516 TheCall->setType(ResultType); 517 518 return move(TheCallResult); 519} 520 521 522/// CheckObjCString - Checks that the argument to the builtin 523/// CFString constructor is correct 524/// Note: It might also make sense to do the UTF-16 conversion here (would 525/// simplify the backend). 526bool Sema::CheckObjCString(Expr *Arg) { 527 Arg = Arg->IgnoreParenCasts(); 528 StringLiteral *Literal = dyn_cast<StringLiteral>(Arg); 529 530 if (!Literal || Literal->isWide()) { 531 Diag(Arg->getLocStart(), diag::err_cfstring_literal_not_string_constant) 532 << Arg->getSourceRange(); 533 return true; 534 } 535 536 size_t NulPos = Literal->getString().find('\0'); 537 if (NulPos != llvm::StringRef::npos) { 538 Diag(getLocationOfStringLiteralByte(Literal, NulPos), 539 diag::warn_cfstring_literal_contains_nul_character) 540 << Arg->getSourceRange(); 541 } 542 if (Literal->containsNonAsciiOrNull()) { 543 llvm::StringRef String = Literal->getString(); 544 unsigned NumBytes = String.size(); 545 llvm::SmallVector<UTF16, 128> ToBuf(NumBytes); 546 const UTF8 *FromPtr = (UTF8 *)String.data(); 547 UTF16 *ToPtr = &ToBuf[0]; 548 549 ConversionResult Result = ConvertUTF8toUTF16(&FromPtr, FromPtr + NumBytes, 550 &ToPtr, ToPtr + NumBytes, 551 strictConversion); 552 // Check for conversion failure. 553 if (Result != conversionOK) 554 Diag(Arg->getLocStart(), 555 diag::warn_cfstring_truncated) << Arg->getSourceRange(); 556 } 557 return false; 558} 559 560/// SemaBuiltinVAStart - Check the arguments to __builtin_va_start for validity. 561/// Emit an error and return true on failure, return false on success. 562bool Sema::SemaBuiltinVAStart(CallExpr *TheCall) { 563 Expr *Fn = TheCall->getCallee(); 564 if (TheCall->getNumArgs() > 2) { 565 Diag(TheCall->getArg(2)->getLocStart(), 566 diag::err_typecheck_call_too_many_args) 567 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 568 << Fn->getSourceRange() 569 << SourceRange(TheCall->getArg(2)->getLocStart(), 570 (*(TheCall->arg_end()-1))->getLocEnd()); 571 return true; 572 } 573 574 if (TheCall->getNumArgs() < 2) { 575 return Diag(TheCall->getLocEnd(), 576 diag::err_typecheck_call_too_few_args_at_least) 577 << 0 /*function call*/ << 2 << TheCall->getNumArgs(); 578 } 579 580 // Determine whether the current function is variadic or not. 581 BlockScopeInfo *CurBlock = getCurBlock(); 582 bool isVariadic; 583 if (CurBlock) 584 isVariadic = CurBlock->TheDecl->isVariadic(); 585 else if (FunctionDecl *FD = getCurFunctionDecl()) 586 isVariadic = FD->isVariadic(); 587 else 588 isVariadic = getCurMethodDecl()->isVariadic(); 589 590 if (!isVariadic) { 591 Diag(Fn->getLocStart(), diag::err_va_start_used_in_non_variadic_function); 592 return true; 593 } 594 595 // Verify that the second argument to the builtin is the last argument of the 596 // current function or method. 597 bool SecondArgIsLastNamedArgument = false; 598 const Expr *Arg = TheCall->getArg(1)->IgnoreParenCasts(); 599 600 if (const DeclRefExpr *DR = dyn_cast<DeclRefExpr>(Arg)) { 601 if (const ParmVarDecl *PV = dyn_cast<ParmVarDecl>(DR->getDecl())) { 602 // FIXME: This isn't correct for methods (results in bogus warning). 603 // Get the last formal in the current function. 604 const ParmVarDecl *LastArg; 605 if (CurBlock) 606 LastArg = *(CurBlock->TheDecl->param_end()-1); 607 else if (FunctionDecl *FD = getCurFunctionDecl()) 608 LastArg = *(FD->param_end()-1); 609 else 610 LastArg = *(getCurMethodDecl()->param_end()-1); 611 SecondArgIsLastNamedArgument = PV == LastArg; 612 } 613 } 614 615 if (!SecondArgIsLastNamedArgument) 616 Diag(TheCall->getArg(1)->getLocStart(), 617 diag::warn_second_parameter_of_va_start_not_last_named_argument); 618 return false; 619} 620 621/// SemaBuiltinUnorderedCompare - Handle functions like __builtin_isgreater and 622/// friends. This is declared to take (...), so we have to check everything. 623bool Sema::SemaBuiltinUnorderedCompare(CallExpr *TheCall) { 624 if (TheCall->getNumArgs() < 2) 625 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 626 << 0 << 2 << TheCall->getNumArgs()/*function call*/; 627 if (TheCall->getNumArgs() > 2) 628 return Diag(TheCall->getArg(2)->getLocStart(), 629 diag::err_typecheck_call_too_many_args) 630 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 631 << SourceRange(TheCall->getArg(2)->getLocStart(), 632 (*(TheCall->arg_end()-1))->getLocEnd()); 633 634 Expr *OrigArg0 = TheCall->getArg(0); 635 Expr *OrigArg1 = TheCall->getArg(1); 636 637 // Do standard promotions between the two arguments, returning their common 638 // type. 639 QualType Res = UsualArithmeticConversions(OrigArg0, OrigArg1, false); 640 641 // Make sure any conversions are pushed back into the call; this is 642 // type safe since unordered compare builtins are declared as "_Bool 643 // foo(...)". 644 TheCall->setArg(0, OrigArg0); 645 TheCall->setArg(1, OrigArg1); 646 647 if (OrigArg0->isTypeDependent() || OrigArg1->isTypeDependent()) 648 return false; 649 650 // If the common type isn't a real floating type, then the arguments were 651 // invalid for this operation. 652 if (!Res->isRealFloatingType()) 653 return Diag(OrigArg0->getLocStart(), 654 diag::err_typecheck_call_invalid_ordered_compare) 655 << OrigArg0->getType() << OrigArg1->getType() 656 << SourceRange(OrigArg0->getLocStart(), OrigArg1->getLocEnd()); 657 658 return false; 659} 660 661/// SemaBuiltinSemaBuiltinFPClassification - Handle functions like 662/// __builtin_isnan and friends. This is declared to take (...), so we have 663/// to check everything. We expect the last argument to be a floating point 664/// value. 665bool Sema::SemaBuiltinFPClassification(CallExpr *TheCall, unsigned NumArgs) { 666 if (TheCall->getNumArgs() < NumArgs) 667 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 668 << 0 << NumArgs << TheCall->getNumArgs()/*function call*/; 669 if (TheCall->getNumArgs() > NumArgs) 670 return Diag(TheCall->getArg(NumArgs)->getLocStart(), 671 diag::err_typecheck_call_too_many_args) 672 << 0 /*function call*/ << NumArgs << TheCall->getNumArgs() 673 << SourceRange(TheCall->getArg(NumArgs)->getLocStart(), 674 (*(TheCall->arg_end()-1))->getLocEnd()); 675 676 Expr *OrigArg = TheCall->getArg(NumArgs-1); 677 678 if (OrigArg->isTypeDependent()) 679 return false; 680 681 // This operation requires a non-_Complex floating-point number. 682 if (!OrigArg->getType()->isRealFloatingType()) 683 return Diag(OrigArg->getLocStart(), 684 diag::err_typecheck_call_invalid_unary_fp) 685 << OrigArg->getType() << OrigArg->getSourceRange(); 686 687 // If this is an implicit conversion from float -> double, remove it. 688 if (ImplicitCastExpr *Cast = dyn_cast<ImplicitCastExpr>(OrigArg)) { 689 Expr *CastArg = Cast->getSubExpr(); 690 if (CastArg->getType()->isSpecificBuiltinType(BuiltinType::Float)) { 691 assert(Cast->getType()->isSpecificBuiltinType(BuiltinType::Double) && 692 "promotion from float to double is the only expected cast here"); 693 Cast->setSubExpr(0); 694 TheCall->setArg(NumArgs-1, CastArg); 695 OrigArg = CastArg; 696 } 697 } 698 699 return false; 700} 701 702/// SemaBuiltinShuffleVector - Handle __builtin_shufflevector. 703// This is declared to take (...), so we have to check everything. 704ExprResult Sema::SemaBuiltinShuffleVector(CallExpr *TheCall) { 705 if (TheCall->getNumArgs() < 2) 706 return ExprError(Diag(TheCall->getLocEnd(), 707 diag::err_typecheck_call_too_few_args_at_least) 708 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 709 << TheCall->getSourceRange()); 710 711 // Determine which of the following types of shufflevector we're checking: 712 // 1) unary, vector mask: (lhs, mask) 713 // 2) binary, vector mask: (lhs, rhs, mask) 714 // 3) binary, scalar mask: (lhs, rhs, index, ..., index) 715 QualType resType = TheCall->getArg(0)->getType(); 716 unsigned numElements = 0; 717 718 if (!TheCall->getArg(0)->isTypeDependent() && 719 !TheCall->getArg(1)->isTypeDependent()) { 720 QualType LHSType = TheCall->getArg(0)->getType(); 721 QualType RHSType = TheCall->getArg(1)->getType(); 722 723 if (!LHSType->isVectorType() || !RHSType->isVectorType()) { 724 Diag(TheCall->getLocStart(), diag::err_shufflevector_non_vector) 725 << SourceRange(TheCall->getArg(0)->getLocStart(), 726 TheCall->getArg(1)->getLocEnd()); 727 return ExprError(); 728 } 729 730 numElements = LHSType->getAs<VectorType>()->getNumElements(); 731 unsigned numResElements = TheCall->getNumArgs() - 2; 732 733 // Check to see if we have a call with 2 vector arguments, the unary shuffle 734 // with mask. If so, verify that RHS is an integer vector type with the 735 // same number of elts as lhs. 736 if (TheCall->getNumArgs() == 2) { 737 if (!RHSType->hasIntegerRepresentation() || 738 RHSType->getAs<VectorType>()->getNumElements() != numElements) 739 Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector) 740 << SourceRange(TheCall->getArg(1)->getLocStart(), 741 TheCall->getArg(1)->getLocEnd()); 742 numResElements = numElements; 743 } 744 else if (!Context.hasSameUnqualifiedType(LHSType, RHSType)) { 745 Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector) 746 << SourceRange(TheCall->getArg(0)->getLocStart(), 747 TheCall->getArg(1)->getLocEnd()); 748 return ExprError(); 749 } else if (numElements != numResElements) { 750 QualType eltType = LHSType->getAs<VectorType>()->getElementType(); 751 resType = Context.getVectorType(eltType, numResElements, 752 VectorType::GenericVector); 753 } 754 } 755 756 for (unsigned i = 2; i < TheCall->getNumArgs(); i++) { 757 if (TheCall->getArg(i)->isTypeDependent() || 758 TheCall->getArg(i)->isValueDependent()) 759 continue; 760 761 llvm::APSInt Result(32); 762 if (!TheCall->getArg(i)->isIntegerConstantExpr(Result, Context)) 763 return ExprError(Diag(TheCall->getLocStart(), 764 diag::err_shufflevector_nonconstant_argument) 765 << TheCall->getArg(i)->getSourceRange()); 766 767 if (Result.getActiveBits() > 64 || Result.getZExtValue() >= numElements*2) 768 return ExprError(Diag(TheCall->getLocStart(), 769 diag::err_shufflevector_argument_too_large) 770 << TheCall->getArg(i)->getSourceRange()); 771 } 772 773 llvm::SmallVector<Expr*, 32> exprs; 774 775 for (unsigned i = 0, e = TheCall->getNumArgs(); i != e; i++) { 776 exprs.push_back(TheCall->getArg(i)); 777 TheCall->setArg(i, 0); 778 } 779 780 return Owned(new (Context) ShuffleVectorExpr(Context, exprs.begin(), 781 exprs.size(), resType, 782 TheCall->getCallee()->getLocStart(), 783 TheCall->getRParenLoc())); 784} 785 786/// SemaBuiltinPrefetch - Handle __builtin_prefetch. 787// This is declared to take (const void*, ...) and can take two 788// optional constant int args. 789bool Sema::SemaBuiltinPrefetch(CallExpr *TheCall) { 790 unsigned NumArgs = TheCall->getNumArgs(); 791 792 if (NumArgs > 3) 793 return Diag(TheCall->getLocEnd(), 794 diag::err_typecheck_call_too_many_args_at_most) 795 << 0 /*function call*/ << 3 << NumArgs 796 << TheCall->getSourceRange(); 797 798 // Argument 0 is checked for us and the remaining arguments must be 799 // constant integers. 800 for (unsigned i = 1; i != NumArgs; ++i) { 801 Expr *Arg = TheCall->getArg(i); 802 803 llvm::APSInt Result; 804 if (SemaBuiltinConstantArg(TheCall, i, Result)) 805 return true; 806 807 // FIXME: gcc issues a warning and rewrites these to 0. These 808 // seems especially odd for the third argument since the default 809 // is 3. 810 if (i == 1) { 811 if (Result.getLimitedValue() > 1) 812 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 813 << "0" << "1" << Arg->getSourceRange(); 814 } else { 815 if (Result.getLimitedValue() > 3) 816 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 817 << "0" << "3" << Arg->getSourceRange(); 818 } 819 } 820 821 return false; 822} 823 824/// SemaBuiltinConstantArg - Handle a check if argument ArgNum of CallExpr 825/// TheCall is a constant expression. 826bool Sema::SemaBuiltinConstantArg(CallExpr *TheCall, int ArgNum, 827 llvm::APSInt &Result) { 828 Expr *Arg = TheCall->getArg(ArgNum); 829 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 830 FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl()); 831 832 if (Arg->isTypeDependent() || Arg->isValueDependent()) return false; 833 834 if (!Arg->isIntegerConstantExpr(Result, Context)) 835 return Diag(TheCall->getLocStart(), diag::err_constant_integer_arg_type) 836 << FDecl->getDeclName() << Arg->getSourceRange(); 837 838 return false; 839} 840 841/// SemaBuiltinObjectSize - Handle __builtin_object_size(void *ptr, 842/// int type). This simply type checks that type is one of the defined 843/// constants (0-3). 844// For compatability check 0-3, llvm only handles 0 and 2. 845bool Sema::SemaBuiltinObjectSize(CallExpr *TheCall) { 846 llvm::APSInt Result; 847 848 // Check constant-ness first. 849 if (SemaBuiltinConstantArg(TheCall, 1, Result)) 850 return true; 851 852 Expr *Arg = TheCall->getArg(1); 853 if (Result.getSExtValue() < 0 || Result.getSExtValue() > 3) { 854 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 855 << "0" << "3" << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 856 } 857 858 return false; 859} 860 861/// SemaBuiltinLongjmp - Handle __builtin_longjmp(void *env[5], int val). 862/// This checks that val is a constant 1. 863bool Sema::SemaBuiltinLongjmp(CallExpr *TheCall) { 864 Expr *Arg = TheCall->getArg(1); 865 llvm::APSInt Result; 866 867 // TODO: This is less than ideal. Overload this to take a value. 868 if (SemaBuiltinConstantArg(TheCall, 1, Result)) 869 return true; 870 871 if (Result != 1) 872 return Diag(TheCall->getLocStart(), diag::err_builtin_longjmp_invalid_val) 873 << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 874 875 return false; 876} 877 878// Handle i > 1 ? "x" : "y", recursivelly 879bool Sema::SemaCheckStringLiteral(const Expr *E, const CallExpr *TheCall, 880 bool HasVAListArg, 881 unsigned format_idx, unsigned firstDataArg, 882 bool isPrintf) { 883 tryAgain: 884 if (E->isTypeDependent() || E->isValueDependent()) 885 return false; 886 887 switch (E->getStmtClass()) { 888 case Stmt::BinaryConditionalOperatorClass: 889 case Stmt::ConditionalOperatorClass: { 890 const AbstractConditionalOperator *C = cast<AbstractConditionalOperator>(E); 891 return SemaCheckStringLiteral(C->getTrueExpr(), TheCall, HasVAListArg, 892 format_idx, firstDataArg, isPrintf) 893 && SemaCheckStringLiteral(C->getFalseExpr(), TheCall, HasVAListArg, 894 format_idx, firstDataArg, isPrintf); 895 } 896 897 case Stmt::IntegerLiteralClass: 898 // Technically -Wformat-nonliteral does not warn about this case. 899 // The behavior of printf and friends in this case is implementation 900 // dependent. Ideally if the format string cannot be null then 901 // it should have a 'nonnull' attribute in the function prototype. 902 return true; 903 904 case Stmt::ImplicitCastExprClass: { 905 E = cast<ImplicitCastExpr>(E)->getSubExpr(); 906 goto tryAgain; 907 } 908 909 case Stmt::ParenExprClass: { 910 E = cast<ParenExpr>(E)->getSubExpr(); 911 goto tryAgain; 912 } 913 914 case Stmt::OpaqueValueExprClass: 915 if (const Expr *src = cast<OpaqueValueExpr>(E)->getSourceExpr()) { 916 E = src; 917 goto tryAgain; 918 } 919 return false; 920 921 case Stmt::DeclRefExprClass: { 922 const DeclRefExpr *DR = cast<DeclRefExpr>(E); 923 924 // As an exception, do not flag errors for variables binding to 925 // const string literals. 926 if (const VarDecl *VD = dyn_cast<VarDecl>(DR->getDecl())) { 927 bool isConstant = false; 928 QualType T = DR->getType(); 929 930 if (const ArrayType *AT = Context.getAsArrayType(T)) { 931 isConstant = AT->getElementType().isConstant(Context); 932 } else if (const PointerType *PT = T->getAs<PointerType>()) { 933 isConstant = T.isConstant(Context) && 934 PT->getPointeeType().isConstant(Context); 935 } 936 937 if (isConstant) { 938 if (const Expr *Init = VD->getAnyInitializer()) 939 return SemaCheckStringLiteral(Init, TheCall, 940 HasVAListArg, format_idx, firstDataArg, 941 isPrintf); 942 } 943 944 // For vprintf* functions (i.e., HasVAListArg==true), we add a 945 // special check to see if the format string is a function parameter 946 // of the function calling the printf function. If the function 947 // has an attribute indicating it is a printf-like function, then we 948 // should suppress warnings concerning non-literals being used in a call 949 // to a vprintf function. For example: 950 // 951 // void 952 // logmessage(char const *fmt __attribute__ (format (printf, 1, 2)), ...){ 953 // va_list ap; 954 // va_start(ap, fmt); 955 // vprintf(fmt, ap); // Do NOT emit a warning about "fmt". 956 // ... 957 // 958 // 959 // FIXME: We don't have full attribute support yet, so just check to see 960 // if the argument is a DeclRefExpr that references a parameter. We'll 961 // add proper support for checking the attribute later. 962 if (HasVAListArg) 963 if (isa<ParmVarDecl>(VD)) 964 return true; 965 } 966 967 return false; 968 } 969 970 case Stmt::CallExprClass: { 971 const CallExpr *CE = cast<CallExpr>(E); 972 if (const ImplicitCastExpr *ICE 973 = dyn_cast<ImplicitCastExpr>(CE->getCallee())) { 974 if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(ICE->getSubExpr())) { 975 if (const FunctionDecl *FD = dyn_cast<FunctionDecl>(DRE->getDecl())) { 976 if (const FormatArgAttr *FA = FD->getAttr<FormatArgAttr>()) { 977 unsigned ArgIndex = FA->getFormatIdx(); 978 const Expr *Arg = CE->getArg(ArgIndex - 1); 979 980 return SemaCheckStringLiteral(Arg, TheCall, HasVAListArg, 981 format_idx, firstDataArg, isPrintf); 982 } 983 } 984 } 985 } 986 987 return false; 988 } 989 case Stmt::ObjCStringLiteralClass: 990 case Stmt::StringLiteralClass: { 991 const StringLiteral *StrE = NULL; 992 993 if (const ObjCStringLiteral *ObjCFExpr = dyn_cast<ObjCStringLiteral>(E)) 994 StrE = ObjCFExpr->getString(); 995 else 996 StrE = cast<StringLiteral>(E); 997 998 if (StrE) { 999 CheckFormatString(StrE, E, TheCall, HasVAListArg, format_idx, 1000 firstDataArg, isPrintf); 1001 return true; 1002 } 1003 1004 return false; 1005 } 1006 1007 default: 1008 return false; 1009 } 1010} 1011 1012void 1013Sema::CheckNonNullArguments(const NonNullAttr *NonNull, 1014 const CallExpr *TheCall) { 1015 for (NonNullAttr::args_iterator i = NonNull->args_begin(), 1016 e = NonNull->args_end(); 1017 i != e; ++i) { 1018 const Expr *ArgExpr = TheCall->getArg(*i); 1019 if (ArgExpr->isNullPointerConstant(Context, 1020 Expr::NPC_ValueDependentIsNotNull)) 1021 Diag(TheCall->getCallee()->getLocStart(), diag::warn_null_arg) 1022 << ArgExpr->getSourceRange(); 1023 } 1024} 1025 1026/// CheckPrintfScanfArguments - Check calls to printf and scanf (and similar 1027/// functions) for correct use of format strings. 1028void 1029Sema::CheckPrintfScanfArguments(const CallExpr *TheCall, bool HasVAListArg, 1030 unsigned format_idx, unsigned firstDataArg, 1031 bool isPrintf) { 1032 1033 const Expr *Fn = TheCall->getCallee(); 1034 1035 // The way the format attribute works in GCC, the implicit this argument 1036 // of member functions is counted. However, it doesn't appear in our own 1037 // lists, so decrement format_idx in that case. 1038 if (isa<CXXMemberCallExpr>(TheCall)) { 1039 const CXXMethodDecl *method_decl = 1040 dyn_cast<CXXMethodDecl>(TheCall->getCalleeDecl()); 1041 if (method_decl && method_decl->isInstance()) { 1042 // Catch a format attribute mistakenly referring to the object argument. 1043 if (format_idx == 0) 1044 return; 1045 --format_idx; 1046 if(firstDataArg != 0) 1047 --firstDataArg; 1048 } 1049 } 1050 1051 // CHECK: printf/scanf-like function is called with no format string. 1052 if (format_idx >= TheCall->getNumArgs()) { 1053 Diag(TheCall->getRParenLoc(), diag::warn_missing_format_string) 1054 << Fn->getSourceRange(); 1055 return; 1056 } 1057 1058 const Expr *OrigFormatExpr = TheCall->getArg(format_idx)->IgnoreParenCasts(); 1059 1060 // CHECK: format string is not a string literal. 1061 // 1062 // Dynamically generated format strings are difficult to 1063 // automatically vet at compile time. Requiring that format strings 1064 // are string literals: (1) permits the checking of format strings by 1065 // the compiler and thereby (2) can practically remove the source of 1066 // many format string exploits. 1067 1068 // Format string can be either ObjC string (e.g. @"%d") or 1069 // C string (e.g. "%d") 1070 // ObjC string uses the same format specifiers as C string, so we can use 1071 // the same format string checking logic for both ObjC and C strings. 1072 if (SemaCheckStringLiteral(OrigFormatExpr, TheCall, HasVAListArg, format_idx, 1073 firstDataArg, isPrintf)) 1074 return; // Literal format string found, check done! 1075 1076 // If there are no arguments specified, warn with -Wformat-security, otherwise 1077 // warn only with -Wformat-nonliteral. 1078 if (TheCall->getNumArgs() == format_idx+1) 1079 Diag(TheCall->getArg(format_idx)->getLocStart(), 1080 diag::warn_format_nonliteral_noargs) 1081 << OrigFormatExpr->getSourceRange(); 1082 else 1083 Diag(TheCall->getArg(format_idx)->getLocStart(), 1084 diag::warn_format_nonliteral) 1085 << OrigFormatExpr->getSourceRange(); 1086} 1087 1088namespace { 1089class CheckFormatHandler : public analyze_format_string::FormatStringHandler { 1090protected: 1091 Sema &S; 1092 const StringLiteral *FExpr; 1093 const Expr *OrigFormatExpr; 1094 const unsigned FirstDataArg; 1095 const unsigned NumDataArgs; 1096 const bool IsObjCLiteral; 1097 const char *Beg; // Start of format string. 1098 const bool HasVAListArg; 1099 const CallExpr *TheCall; 1100 unsigned FormatIdx; 1101 llvm::BitVector CoveredArgs; 1102 bool usesPositionalArgs; 1103 bool atFirstArg; 1104public: 1105 CheckFormatHandler(Sema &s, const StringLiteral *fexpr, 1106 const Expr *origFormatExpr, unsigned firstDataArg, 1107 unsigned numDataArgs, bool isObjCLiteral, 1108 const char *beg, bool hasVAListArg, 1109 const CallExpr *theCall, unsigned formatIdx) 1110 : S(s), FExpr(fexpr), OrigFormatExpr(origFormatExpr), 1111 FirstDataArg(firstDataArg), 1112 NumDataArgs(numDataArgs), 1113 IsObjCLiteral(isObjCLiteral), Beg(beg), 1114 HasVAListArg(hasVAListArg), 1115 TheCall(theCall), FormatIdx(formatIdx), 1116 usesPositionalArgs(false), atFirstArg(true) { 1117 CoveredArgs.resize(numDataArgs); 1118 CoveredArgs.reset(); 1119 } 1120 1121 void DoneProcessing(); 1122 1123 void HandleIncompleteSpecifier(const char *startSpecifier, 1124 unsigned specifierLen); 1125 1126 virtual void HandleInvalidPosition(const char *startSpecifier, 1127 unsigned specifierLen, 1128 analyze_format_string::PositionContext p); 1129 1130 virtual void HandleZeroPosition(const char *startPos, unsigned posLen); 1131 1132 void HandleNullChar(const char *nullCharacter); 1133 1134protected: 1135 bool HandleInvalidConversionSpecifier(unsigned argIndex, SourceLocation Loc, 1136 const char *startSpec, 1137 unsigned specifierLen, 1138 const char *csStart, unsigned csLen); 1139 1140 SourceRange getFormatStringRange(); 1141 CharSourceRange getSpecifierRange(const char *startSpecifier, 1142 unsigned specifierLen); 1143 SourceLocation getLocationOfByte(const char *x); 1144 1145 const Expr *getDataArg(unsigned i) const; 1146 1147 bool CheckNumArgs(const analyze_format_string::FormatSpecifier &FS, 1148 const analyze_format_string::ConversionSpecifier &CS, 1149 const char *startSpecifier, unsigned specifierLen, 1150 unsigned argIndex); 1151}; 1152} 1153 1154SourceRange CheckFormatHandler::getFormatStringRange() { 1155 return OrigFormatExpr->getSourceRange(); 1156} 1157 1158CharSourceRange CheckFormatHandler:: 1159getSpecifierRange(const char *startSpecifier, unsigned specifierLen) { 1160 SourceLocation Start = getLocationOfByte(startSpecifier); 1161 SourceLocation End = getLocationOfByte(startSpecifier + specifierLen - 1); 1162 1163 // Advance the end SourceLocation by one due to half-open ranges. 1164 End = End.getFileLocWithOffset(1); 1165 1166 return CharSourceRange::getCharRange(Start, End); 1167} 1168 1169SourceLocation CheckFormatHandler::getLocationOfByte(const char *x) { 1170 return S.getLocationOfStringLiteralByte(FExpr, x - Beg); 1171} 1172 1173void CheckFormatHandler::HandleIncompleteSpecifier(const char *startSpecifier, 1174 unsigned specifierLen){ 1175 SourceLocation Loc = getLocationOfByte(startSpecifier); 1176 S.Diag(Loc, diag::warn_printf_incomplete_specifier) 1177 << getSpecifierRange(startSpecifier, specifierLen); 1178} 1179 1180void 1181CheckFormatHandler::HandleInvalidPosition(const char *startPos, unsigned posLen, 1182 analyze_format_string::PositionContext p) { 1183 SourceLocation Loc = getLocationOfByte(startPos); 1184 S.Diag(Loc, diag::warn_format_invalid_positional_specifier) 1185 << (unsigned) p << getSpecifierRange(startPos, posLen); 1186} 1187 1188void CheckFormatHandler::HandleZeroPosition(const char *startPos, 1189 unsigned posLen) { 1190 SourceLocation Loc = getLocationOfByte(startPos); 1191 S.Diag(Loc, diag::warn_format_zero_positional_specifier) 1192 << getSpecifierRange(startPos, posLen); 1193} 1194 1195void CheckFormatHandler::HandleNullChar(const char *nullCharacter) { 1196 // The presence of a null character is likely an error. 1197 S.Diag(getLocationOfByte(nullCharacter), 1198 diag::warn_printf_format_string_contains_null_char) 1199 << getFormatStringRange(); 1200} 1201 1202const Expr *CheckFormatHandler::getDataArg(unsigned i) const { 1203 return TheCall->getArg(FirstDataArg + i); 1204} 1205 1206void CheckFormatHandler::DoneProcessing() { 1207 // Does the number of data arguments exceed the number of 1208 // format conversions in the format string? 1209 if (!HasVAListArg) { 1210 // Find any arguments that weren't covered. 1211 CoveredArgs.flip(); 1212 signed notCoveredArg = CoveredArgs.find_first(); 1213 if (notCoveredArg >= 0) { 1214 assert((unsigned)notCoveredArg < NumDataArgs); 1215 S.Diag(getDataArg((unsigned) notCoveredArg)->getLocStart(), 1216 diag::warn_printf_data_arg_not_used) 1217 << getFormatStringRange(); 1218 } 1219 } 1220} 1221 1222bool 1223CheckFormatHandler::HandleInvalidConversionSpecifier(unsigned argIndex, 1224 SourceLocation Loc, 1225 const char *startSpec, 1226 unsigned specifierLen, 1227 const char *csStart, 1228 unsigned csLen) { 1229 1230 bool keepGoing = true; 1231 if (argIndex < NumDataArgs) { 1232 // Consider the argument coverered, even though the specifier doesn't 1233 // make sense. 1234 CoveredArgs.set(argIndex); 1235 } 1236 else { 1237 // If argIndex exceeds the number of data arguments we 1238 // don't issue a warning because that is just a cascade of warnings (and 1239 // they may have intended '%%' anyway). We don't want to continue processing 1240 // the format string after this point, however, as we will like just get 1241 // gibberish when trying to match arguments. 1242 keepGoing = false; 1243 } 1244 1245 S.Diag(Loc, diag::warn_format_invalid_conversion) 1246 << llvm::StringRef(csStart, csLen) 1247 << getSpecifierRange(startSpec, specifierLen); 1248 1249 return keepGoing; 1250} 1251 1252bool 1253CheckFormatHandler::CheckNumArgs( 1254 const analyze_format_string::FormatSpecifier &FS, 1255 const analyze_format_string::ConversionSpecifier &CS, 1256 const char *startSpecifier, unsigned specifierLen, unsigned argIndex) { 1257 1258 if (argIndex >= NumDataArgs) { 1259 if (FS.usesPositionalArg()) { 1260 S.Diag(getLocationOfByte(CS.getStart()), 1261 diag::warn_printf_positional_arg_exceeds_data_args) 1262 << (argIndex+1) << NumDataArgs 1263 << getSpecifierRange(startSpecifier, specifierLen); 1264 } 1265 else { 1266 S.Diag(getLocationOfByte(CS.getStart()), 1267 diag::warn_printf_insufficient_data_args) 1268 << getSpecifierRange(startSpecifier, specifierLen); 1269 } 1270 1271 return false; 1272 } 1273 return true; 1274} 1275 1276//===--- CHECK: Printf format string checking ------------------------------===// 1277 1278namespace { 1279class CheckPrintfHandler : public CheckFormatHandler { 1280public: 1281 CheckPrintfHandler(Sema &s, const StringLiteral *fexpr, 1282 const Expr *origFormatExpr, unsigned firstDataArg, 1283 unsigned numDataArgs, bool isObjCLiteral, 1284 const char *beg, bool hasVAListArg, 1285 const CallExpr *theCall, unsigned formatIdx) 1286 : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg, 1287 numDataArgs, isObjCLiteral, beg, hasVAListArg, 1288 theCall, formatIdx) {} 1289 1290 1291 bool HandleInvalidPrintfConversionSpecifier( 1292 const analyze_printf::PrintfSpecifier &FS, 1293 const char *startSpecifier, 1294 unsigned specifierLen); 1295 1296 bool HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier &FS, 1297 const char *startSpecifier, 1298 unsigned specifierLen); 1299 1300 bool HandleAmount(const analyze_format_string::OptionalAmount &Amt, unsigned k, 1301 const char *startSpecifier, unsigned specifierLen); 1302 void HandleInvalidAmount(const analyze_printf::PrintfSpecifier &FS, 1303 const analyze_printf::OptionalAmount &Amt, 1304 unsigned type, 1305 const char *startSpecifier, unsigned specifierLen); 1306 void HandleFlag(const analyze_printf::PrintfSpecifier &FS, 1307 const analyze_printf::OptionalFlag &flag, 1308 const char *startSpecifier, unsigned specifierLen); 1309 void HandleIgnoredFlag(const analyze_printf::PrintfSpecifier &FS, 1310 const analyze_printf::OptionalFlag &ignoredFlag, 1311 const analyze_printf::OptionalFlag &flag, 1312 const char *startSpecifier, unsigned specifierLen); 1313}; 1314} 1315 1316bool CheckPrintfHandler::HandleInvalidPrintfConversionSpecifier( 1317 const analyze_printf::PrintfSpecifier &FS, 1318 const char *startSpecifier, 1319 unsigned specifierLen) { 1320 const analyze_printf::PrintfConversionSpecifier &CS = 1321 FS.getConversionSpecifier(); 1322 1323 return HandleInvalidConversionSpecifier(FS.getArgIndex(), 1324 getLocationOfByte(CS.getStart()), 1325 startSpecifier, specifierLen, 1326 CS.getStart(), CS.getLength()); 1327} 1328 1329bool CheckPrintfHandler::HandleAmount( 1330 const analyze_format_string::OptionalAmount &Amt, 1331 unsigned k, const char *startSpecifier, 1332 unsigned specifierLen) { 1333 1334 if (Amt.hasDataArgument()) { 1335 if (!HasVAListArg) { 1336 unsigned argIndex = Amt.getArgIndex(); 1337 if (argIndex >= NumDataArgs) { 1338 S.Diag(getLocationOfByte(Amt.getStart()), 1339 diag::warn_printf_asterisk_missing_arg) 1340 << k << getSpecifierRange(startSpecifier, specifierLen); 1341 // Don't do any more checking. We will just emit 1342 // spurious errors. 1343 return false; 1344 } 1345 1346 // Type check the data argument. It should be an 'int'. 1347 // Although not in conformance with C99, we also allow the argument to be 1348 // an 'unsigned int' as that is a reasonably safe case. GCC also 1349 // doesn't emit a warning for that case. 1350 CoveredArgs.set(argIndex); 1351 const Expr *Arg = getDataArg(argIndex); 1352 QualType T = Arg->getType(); 1353 1354 const analyze_printf::ArgTypeResult &ATR = Amt.getArgType(S.Context); 1355 assert(ATR.isValid()); 1356 1357 if (!ATR.matchesType(S.Context, T)) { 1358 S.Diag(getLocationOfByte(Amt.getStart()), 1359 diag::warn_printf_asterisk_wrong_type) 1360 << k 1361 << ATR.getRepresentativeType(S.Context) << T 1362 << getSpecifierRange(startSpecifier, specifierLen) 1363 << Arg->getSourceRange(); 1364 // Don't do any more checking. We will just emit 1365 // spurious errors. 1366 return false; 1367 } 1368 } 1369 } 1370 return true; 1371} 1372 1373void CheckPrintfHandler::HandleInvalidAmount( 1374 const analyze_printf::PrintfSpecifier &FS, 1375 const analyze_printf::OptionalAmount &Amt, 1376 unsigned type, 1377 const char *startSpecifier, 1378 unsigned specifierLen) { 1379 const analyze_printf::PrintfConversionSpecifier &CS = 1380 FS.getConversionSpecifier(); 1381 switch (Amt.getHowSpecified()) { 1382 case analyze_printf::OptionalAmount::Constant: 1383 S.Diag(getLocationOfByte(Amt.getStart()), 1384 diag::warn_printf_nonsensical_optional_amount) 1385 << type 1386 << CS.toString() 1387 << getSpecifierRange(startSpecifier, specifierLen) 1388 << FixItHint::CreateRemoval(getSpecifierRange(Amt.getStart(), 1389 Amt.getConstantLength())); 1390 break; 1391 1392 default: 1393 S.Diag(getLocationOfByte(Amt.getStart()), 1394 diag::warn_printf_nonsensical_optional_amount) 1395 << type 1396 << CS.toString() 1397 << getSpecifierRange(startSpecifier, specifierLen); 1398 break; 1399 } 1400} 1401 1402void CheckPrintfHandler::HandleFlag(const analyze_printf::PrintfSpecifier &FS, 1403 const analyze_printf::OptionalFlag &flag, 1404 const char *startSpecifier, 1405 unsigned specifierLen) { 1406 // Warn about pointless flag with a fixit removal. 1407 const analyze_printf::PrintfConversionSpecifier &CS = 1408 FS.getConversionSpecifier(); 1409 S.Diag(getLocationOfByte(flag.getPosition()), 1410 diag::warn_printf_nonsensical_flag) 1411 << flag.toString() << CS.toString() 1412 << getSpecifierRange(startSpecifier, specifierLen) 1413 << FixItHint::CreateRemoval(getSpecifierRange(flag.getPosition(), 1)); 1414} 1415 1416void CheckPrintfHandler::HandleIgnoredFlag( 1417 const analyze_printf::PrintfSpecifier &FS, 1418 const analyze_printf::OptionalFlag &ignoredFlag, 1419 const analyze_printf::OptionalFlag &flag, 1420 const char *startSpecifier, 1421 unsigned specifierLen) { 1422 // Warn about ignored flag with a fixit removal. 1423 S.Diag(getLocationOfByte(ignoredFlag.getPosition()), 1424 diag::warn_printf_ignored_flag) 1425 << ignoredFlag.toString() << flag.toString() 1426 << getSpecifierRange(startSpecifier, specifierLen) 1427 << FixItHint::CreateRemoval(getSpecifierRange( 1428 ignoredFlag.getPosition(), 1)); 1429} 1430 1431bool 1432CheckPrintfHandler::HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier 1433 &FS, 1434 const char *startSpecifier, 1435 unsigned specifierLen) { 1436 1437 using namespace analyze_format_string; 1438 using namespace analyze_printf; 1439 const PrintfConversionSpecifier &CS = FS.getConversionSpecifier(); 1440 1441 if (FS.consumesDataArgument()) { 1442 if (atFirstArg) { 1443 atFirstArg = false; 1444 usesPositionalArgs = FS.usesPositionalArg(); 1445 } 1446 else if (usesPositionalArgs != FS.usesPositionalArg()) { 1447 // Cannot mix-and-match positional and non-positional arguments. 1448 S.Diag(getLocationOfByte(CS.getStart()), 1449 diag::warn_format_mix_positional_nonpositional_args) 1450 << getSpecifierRange(startSpecifier, specifierLen); 1451 return false; 1452 } 1453 } 1454 1455 // First check if the field width, precision, and conversion specifier 1456 // have matching data arguments. 1457 if (!HandleAmount(FS.getFieldWidth(), /* field width */ 0, 1458 startSpecifier, specifierLen)) { 1459 return false; 1460 } 1461 1462 if (!HandleAmount(FS.getPrecision(), /* precision */ 1, 1463 startSpecifier, specifierLen)) { 1464 return false; 1465 } 1466 1467 if (!CS.consumesDataArgument()) { 1468 // FIXME: Technically specifying a precision or field width here 1469 // makes no sense. Worth issuing a warning at some point. 1470 return true; 1471 } 1472 1473 // Consume the argument. 1474 unsigned argIndex = FS.getArgIndex(); 1475 if (argIndex < NumDataArgs) { 1476 // The check to see if the argIndex is valid will come later. 1477 // We set the bit here because we may exit early from this 1478 // function if we encounter some other error. 1479 CoveredArgs.set(argIndex); 1480 } 1481 1482 // Check for using an Objective-C specific conversion specifier 1483 // in a non-ObjC literal. 1484 if (!IsObjCLiteral && CS.isObjCArg()) { 1485 return HandleInvalidPrintfConversionSpecifier(FS, startSpecifier, 1486 specifierLen); 1487 } 1488 1489 // Check for invalid use of field width 1490 if (!FS.hasValidFieldWidth()) { 1491 HandleInvalidAmount(FS, FS.getFieldWidth(), /* field width */ 0, 1492 startSpecifier, specifierLen); 1493 } 1494 1495 // Check for invalid use of precision 1496 if (!FS.hasValidPrecision()) { 1497 HandleInvalidAmount(FS, FS.getPrecision(), /* precision */ 1, 1498 startSpecifier, specifierLen); 1499 } 1500 1501 // Check each flag does not conflict with any other component. 1502 if (!FS.hasValidThousandsGroupingPrefix()) 1503 HandleFlag(FS, FS.hasThousandsGrouping(), startSpecifier, specifierLen); 1504 if (!FS.hasValidLeadingZeros()) 1505 HandleFlag(FS, FS.hasLeadingZeros(), startSpecifier, specifierLen); 1506 if (!FS.hasValidPlusPrefix()) 1507 HandleFlag(FS, FS.hasPlusPrefix(), startSpecifier, specifierLen); 1508 if (!FS.hasValidSpacePrefix()) 1509 HandleFlag(FS, FS.hasSpacePrefix(), startSpecifier, specifierLen); 1510 if (!FS.hasValidAlternativeForm()) 1511 HandleFlag(FS, FS.hasAlternativeForm(), startSpecifier, specifierLen); 1512 if (!FS.hasValidLeftJustified()) 1513 HandleFlag(FS, FS.isLeftJustified(), startSpecifier, specifierLen); 1514 1515 // Check that flags are not ignored by another flag 1516 if (FS.hasSpacePrefix() && FS.hasPlusPrefix()) // ' ' ignored by '+' 1517 HandleIgnoredFlag(FS, FS.hasSpacePrefix(), FS.hasPlusPrefix(), 1518 startSpecifier, specifierLen); 1519 if (FS.hasLeadingZeros() && FS.isLeftJustified()) // '0' ignored by '-' 1520 HandleIgnoredFlag(FS, FS.hasLeadingZeros(), FS.isLeftJustified(), 1521 startSpecifier, specifierLen); 1522 1523 // Check the length modifier is valid with the given conversion specifier. 1524 const LengthModifier &LM = FS.getLengthModifier(); 1525 if (!FS.hasValidLengthModifier()) 1526 S.Diag(getLocationOfByte(LM.getStart()), 1527 diag::warn_format_nonsensical_length) 1528 << LM.toString() << CS.toString() 1529 << getSpecifierRange(startSpecifier, specifierLen) 1530 << FixItHint::CreateRemoval(getSpecifierRange(LM.getStart(), 1531 LM.getLength())); 1532 1533 // Are we using '%n'? 1534 if (CS.getKind() == ConversionSpecifier::nArg) { 1535 // Issue a warning about this being a possible security issue. 1536 S.Diag(getLocationOfByte(CS.getStart()), diag::warn_printf_write_back) 1537 << getSpecifierRange(startSpecifier, specifierLen); 1538 // Continue checking the other format specifiers. 1539 return true; 1540 } 1541 1542 // The remaining checks depend on the data arguments. 1543 if (HasVAListArg) 1544 return true; 1545 1546 if (!CheckNumArgs(FS, CS, startSpecifier, specifierLen, argIndex)) 1547 return false; 1548 1549 // Now type check the data expression that matches the 1550 // format specifier. 1551 const Expr *Ex = getDataArg(argIndex); 1552 const analyze_printf::ArgTypeResult &ATR = FS.getArgType(S.Context); 1553 if (ATR.isValid() && !ATR.matchesType(S.Context, Ex->getType())) { 1554 // Check if we didn't match because of an implicit cast from a 'char' 1555 // or 'short' to an 'int'. This is done because printf is a varargs 1556 // function. 1557 if (const ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(Ex)) 1558 if (ICE->getType() == S.Context.IntTy) { 1559 // All further checking is done on the subexpression. 1560 Ex = ICE->getSubExpr(); 1561 if (ATR.matchesType(S.Context, Ex->getType())) 1562 return true; 1563 } 1564 1565 // We may be able to offer a FixItHint if it is a supported type. 1566 PrintfSpecifier fixedFS = FS; 1567 bool success = fixedFS.fixType(Ex->getType()); 1568 1569 if (success) { 1570 // Get the fix string from the fixed format specifier 1571 llvm::SmallString<128> buf; 1572 llvm::raw_svector_ostream os(buf); 1573 fixedFS.toString(os); 1574 1575 // FIXME: getRepresentativeType() perhaps should return a string 1576 // instead of a QualType to better handle when the representative 1577 // type is 'wint_t' (which is defined in the system headers). 1578 S.Diag(getLocationOfByte(CS.getStart()), 1579 diag::warn_printf_conversion_argument_type_mismatch) 1580 << ATR.getRepresentativeType(S.Context) << Ex->getType() 1581 << getSpecifierRange(startSpecifier, specifierLen) 1582 << Ex->getSourceRange() 1583 << FixItHint::CreateReplacement( 1584 getSpecifierRange(startSpecifier, specifierLen), 1585 os.str()); 1586 } 1587 else { 1588 S.Diag(getLocationOfByte(CS.getStart()), 1589 diag::warn_printf_conversion_argument_type_mismatch) 1590 << ATR.getRepresentativeType(S.Context) << Ex->getType() 1591 << getSpecifierRange(startSpecifier, specifierLen) 1592 << Ex->getSourceRange(); 1593 } 1594 } 1595 1596 return true; 1597} 1598 1599//===--- CHECK: Scanf format string checking ------------------------------===// 1600 1601namespace { 1602class CheckScanfHandler : public CheckFormatHandler { 1603public: 1604 CheckScanfHandler(Sema &s, const StringLiteral *fexpr, 1605 const Expr *origFormatExpr, unsigned firstDataArg, 1606 unsigned numDataArgs, bool isObjCLiteral, 1607 const char *beg, bool hasVAListArg, 1608 const CallExpr *theCall, unsigned formatIdx) 1609 : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg, 1610 numDataArgs, isObjCLiteral, beg, hasVAListArg, 1611 theCall, formatIdx) {} 1612 1613 bool HandleScanfSpecifier(const analyze_scanf::ScanfSpecifier &FS, 1614 const char *startSpecifier, 1615 unsigned specifierLen); 1616 1617 bool HandleInvalidScanfConversionSpecifier( 1618 const analyze_scanf::ScanfSpecifier &FS, 1619 const char *startSpecifier, 1620 unsigned specifierLen); 1621 1622 void HandleIncompleteScanList(const char *start, const char *end); 1623}; 1624} 1625 1626void CheckScanfHandler::HandleIncompleteScanList(const char *start, 1627 const char *end) { 1628 S.Diag(getLocationOfByte(end), diag::warn_scanf_scanlist_incomplete) 1629 << getSpecifierRange(start, end - start); 1630} 1631 1632bool CheckScanfHandler::HandleInvalidScanfConversionSpecifier( 1633 const analyze_scanf::ScanfSpecifier &FS, 1634 const char *startSpecifier, 1635 unsigned specifierLen) { 1636 1637 const analyze_scanf::ScanfConversionSpecifier &CS = 1638 FS.getConversionSpecifier(); 1639 1640 return HandleInvalidConversionSpecifier(FS.getArgIndex(), 1641 getLocationOfByte(CS.getStart()), 1642 startSpecifier, specifierLen, 1643 CS.getStart(), CS.getLength()); 1644} 1645 1646bool CheckScanfHandler::HandleScanfSpecifier( 1647 const analyze_scanf::ScanfSpecifier &FS, 1648 const char *startSpecifier, 1649 unsigned specifierLen) { 1650 1651 using namespace analyze_scanf; 1652 using namespace analyze_format_string; 1653 1654 const ScanfConversionSpecifier &CS = FS.getConversionSpecifier(); 1655 1656 // Handle case where '%' and '*' don't consume an argument. These shouldn't 1657 // be used to decide if we are using positional arguments consistently. 1658 if (FS.consumesDataArgument()) { 1659 if (atFirstArg) { 1660 atFirstArg = false; 1661 usesPositionalArgs = FS.usesPositionalArg(); 1662 } 1663 else if (usesPositionalArgs != FS.usesPositionalArg()) { 1664 // Cannot mix-and-match positional and non-positional arguments. 1665 S.Diag(getLocationOfByte(CS.getStart()), 1666 diag::warn_format_mix_positional_nonpositional_args) 1667 << getSpecifierRange(startSpecifier, specifierLen); 1668 return false; 1669 } 1670 } 1671 1672 // Check if the field with is non-zero. 1673 const OptionalAmount &Amt = FS.getFieldWidth(); 1674 if (Amt.getHowSpecified() == OptionalAmount::Constant) { 1675 if (Amt.getConstantAmount() == 0) { 1676 const CharSourceRange &R = getSpecifierRange(Amt.getStart(), 1677 Amt.getConstantLength()); 1678 S.Diag(getLocationOfByte(Amt.getStart()), 1679 diag::warn_scanf_nonzero_width) 1680 << R << FixItHint::CreateRemoval(R); 1681 } 1682 } 1683 1684 if (!FS.consumesDataArgument()) { 1685 // FIXME: Technically specifying a precision or field width here 1686 // makes no sense. Worth issuing a warning at some point. 1687 return true; 1688 } 1689 1690 // Consume the argument. 1691 unsigned argIndex = FS.getArgIndex(); 1692 if (argIndex < NumDataArgs) { 1693 // The check to see if the argIndex is valid will come later. 1694 // We set the bit here because we may exit early from this 1695 // function if we encounter some other error. 1696 CoveredArgs.set(argIndex); 1697 } 1698 1699 // Check the length modifier is valid with the given conversion specifier. 1700 const LengthModifier &LM = FS.getLengthModifier(); 1701 if (!FS.hasValidLengthModifier()) { 1702 S.Diag(getLocationOfByte(LM.getStart()), 1703 diag::warn_format_nonsensical_length) 1704 << LM.toString() << CS.toString() 1705 << getSpecifierRange(startSpecifier, specifierLen) 1706 << FixItHint::CreateRemoval(getSpecifierRange(LM.getStart(), 1707 LM.getLength())); 1708 } 1709 1710 // The remaining checks depend on the data arguments. 1711 if (HasVAListArg) 1712 return true; 1713 1714 if (!CheckNumArgs(FS, CS, startSpecifier, specifierLen, argIndex)) 1715 return false; 1716 1717 // FIXME: Check that the argument type matches the format specifier. 1718 1719 return true; 1720} 1721 1722void Sema::CheckFormatString(const StringLiteral *FExpr, 1723 const Expr *OrigFormatExpr, 1724 const CallExpr *TheCall, bool HasVAListArg, 1725 unsigned format_idx, unsigned firstDataArg, 1726 bool isPrintf) { 1727 1728 // CHECK: is the format string a wide literal? 1729 if (FExpr->isWide()) { 1730 Diag(FExpr->getLocStart(), 1731 diag::warn_format_string_is_wide_literal) 1732 << OrigFormatExpr->getSourceRange(); 1733 return; 1734 } 1735 1736 // Str - The format string. NOTE: this is NOT null-terminated! 1737 llvm::StringRef StrRef = FExpr->getString(); 1738 const char *Str = StrRef.data(); 1739 unsigned StrLen = StrRef.size(); 1740 1741 // CHECK: empty format string? 1742 if (StrLen == 0) { 1743 Diag(FExpr->getLocStart(), diag::warn_empty_format_string) 1744 << OrigFormatExpr->getSourceRange(); 1745 return; 1746 } 1747 1748 if (isPrintf) { 1749 CheckPrintfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg, 1750 TheCall->getNumArgs() - firstDataArg, 1751 isa<ObjCStringLiteral>(OrigFormatExpr), Str, 1752 HasVAListArg, TheCall, format_idx); 1753 1754 if (!analyze_format_string::ParsePrintfString(H, Str, Str + StrLen)) 1755 H.DoneProcessing(); 1756 } 1757 else { 1758 CheckScanfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg, 1759 TheCall->getNumArgs() - firstDataArg, 1760 isa<ObjCStringLiteral>(OrigFormatExpr), Str, 1761 HasVAListArg, TheCall, format_idx); 1762 1763 if (!analyze_format_string::ParseScanfString(H, Str, Str + StrLen)) 1764 H.DoneProcessing(); 1765 } 1766} 1767 1768//===--- CHECK: Return Address of Stack Variable --------------------------===// 1769 1770static Expr *EvalVal(Expr *E, llvm::SmallVectorImpl<DeclRefExpr *> &refVars); 1771static Expr *EvalAddr(Expr* E, llvm::SmallVectorImpl<DeclRefExpr *> &refVars); 1772 1773/// CheckReturnStackAddr - Check if a return statement returns the address 1774/// of a stack variable. 1775void 1776Sema::CheckReturnStackAddr(Expr *RetValExp, QualType lhsType, 1777 SourceLocation ReturnLoc) { 1778 1779 Expr *stackE = 0; 1780 llvm::SmallVector<DeclRefExpr *, 8> refVars; 1781 1782 // Perform checking for returned stack addresses, local blocks, 1783 // label addresses or references to temporaries. 1784 if (lhsType->isPointerType() || lhsType->isBlockPointerType()) { 1785 stackE = EvalAddr(RetValExp, refVars); 1786 } else if (lhsType->isReferenceType()) { 1787 stackE = EvalVal(RetValExp, refVars); 1788 } 1789 1790 if (stackE == 0) 1791 return; // Nothing suspicious was found. 1792 1793 SourceLocation diagLoc; 1794 SourceRange diagRange; 1795 if (refVars.empty()) { 1796 diagLoc = stackE->getLocStart(); 1797 diagRange = stackE->getSourceRange(); 1798 } else { 1799 // We followed through a reference variable. 'stackE' contains the 1800 // problematic expression but we will warn at the return statement pointing 1801 // at the reference variable. We will later display the "trail" of 1802 // reference variables using notes. 1803 diagLoc = refVars[0]->getLocStart(); 1804 diagRange = refVars[0]->getSourceRange(); 1805 } 1806 1807 if (DeclRefExpr *DR = dyn_cast<DeclRefExpr>(stackE)) { //address of local var. 1808 Diag(diagLoc, lhsType->isReferenceType() ? diag::warn_ret_stack_ref 1809 : diag::warn_ret_stack_addr) 1810 << DR->getDecl()->getDeclName() << diagRange; 1811 } else if (isa<BlockExpr>(stackE)) { // local block. 1812 Diag(diagLoc, diag::err_ret_local_block) << diagRange; 1813 } else if (isa<AddrLabelExpr>(stackE)) { // address of label. 1814 Diag(diagLoc, diag::warn_ret_addr_label) << diagRange; 1815 } else { // local temporary. 1816 Diag(diagLoc, lhsType->isReferenceType() ? diag::warn_ret_local_temp_ref 1817 : diag::warn_ret_local_temp_addr) 1818 << diagRange; 1819 } 1820 1821 // Display the "trail" of reference variables that we followed until we 1822 // found the problematic expression using notes. 1823 for (unsigned i = 0, e = refVars.size(); i != e; ++i) { 1824 VarDecl *VD = cast<VarDecl>(refVars[i]->getDecl()); 1825 // If this var binds to another reference var, show the range of the next 1826 // var, otherwise the var binds to the problematic expression, in which case 1827 // show the range of the expression. 1828 SourceRange range = (i < e-1) ? refVars[i+1]->getSourceRange() 1829 : stackE->getSourceRange(); 1830 Diag(VD->getLocation(), diag::note_ref_var_local_bind) 1831 << VD->getDeclName() << range; 1832 } 1833} 1834 1835/// EvalAddr - EvalAddr and EvalVal are mutually recursive functions that 1836/// check if the expression in a return statement evaluates to an address 1837/// to a location on the stack, a local block, an address of a label, or a 1838/// reference to local temporary. The recursion is used to traverse the 1839/// AST of the return expression, with recursion backtracking when we 1840/// encounter a subexpression that (1) clearly does not lead to one of the 1841/// above problematic expressions (2) is something we cannot determine leads to 1842/// a problematic expression based on such local checking. 1843/// 1844/// Both EvalAddr and EvalVal follow through reference variables to evaluate 1845/// the expression that they point to. Such variables are added to the 1846/// 'refVars' vector so that we know what the reference variable "trail" was. 1847/// 1848/// EvalAddr processes expressions that are pointers that are used as 1849/// references (and not L-values). EvalVal handles all other values. 1850/// At the base case of the recursion is a check for the above problematic 1851/// expressions. 1852/// 1853/// This implementation handles: 1854/// 1855/// * pointer-to-pointer casts 1856/// * implicit conversions from array references to pointers 1857/// * taking the address of fields 1858/// * arbitrary interplay between "&" and "*" operators 1859/// * pointer arithmetic from an address of a stack variable 1860/// * taking the address of an array element where the array is on the stack 1861static Expr *EvalAddr(Expr *E, llvm::SmallVectorImpl<DeclRefExpr *> &refVars) { 1862 if (E->isTypeDependent()) 1863 return NULL; 1864 1865 // We should only be called for evaluating pointer expressions. 1866 assert((E->getType()->isAnyPointerType() || 1867 E->getType()->isBlockPointerType() || 1868 E->getType()->isObjCQualifiedIdType()) && 1869 "EvalAddr only works on pointers"); 1870 1871 // Our "symbolic interpreter" is just a dispatch off the currently 1872 // viewed AST node. We then recursively traverse the AST by calling 1873 // EvalAddr and EvalVal appropriately. 1874 switch (E->getStmtClass()) { 1875 case Stmt::ParenExprClass: 1876 // Ignore parentheses. 1877 return EvalAddr(cast<ParenExpr>(E)->getSubExpr(), refVars); 1878 1879 case Stmt::DeclRefExprClass: { 1880 DeclRefExpr *DR = cast<DeclRefExpr>(E); 1881 1882 if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl())) 1883 // If this is a reference variable, follow through to the expression that 1884 // it points to. 1885 if (V->hasLocalStorage() && 1886 V->getType()->isReferenceType() && V->hasInit()) { 1887 // Add the reference variable to the "trail". 1888 refVars.push_back(DR); 1889 return EvalAddr(V->getInit(), refVars); 1890 } 1891 1892 return NULL; 1893 } 1894 1895 case Stmt::UnaryOperatorClass: { 1896 // The only unary operator that make sense to handle here 1897 // is AddrOf. All others don't make sense as pointers. 1898 UnaryOperator *U = cast<UnaryOperator>(E); 1899 1900 if (U->getOpcode() == UO_AddrOf) 1901 return EvalVal(U->getSubExpr(), refVars); 1902 else 1903 return NULL; 1904 } 1905 1906 case Stmt::BinaryOperatorClass: { 1907 // Handle pointer arithmetic. All other binary operators are not valid 1908 // in this context. 1909 BinaryOperator *B = cast<BinaryOperator>(E); 1910 BinaryOperatorKind op = B->getOpcode(); 1911 1912 if (op != BO_Add && op != BO_Sub) 1913 return NULL; 1914 1915 Expr *Base = B->getLHS(); 1916 1917 // Determine which argument is the real pointer base. It could be 1918 // the RHS argument instead of the LHS. 1919 if (!Base->getType()->isPointerType()) Base = B->getRHS(); 1920 1921 assert (Base->getType()->isPointerType()); 1922 return EvalAddr(Base, refVars); 1923 } 1924 1925 // For conditional operators we need to see if either the LHS or RHS are 1926 // valid DeclRefExpr*s. If one of them is valid, we return it. 1927 case Stmt::ConditionalOperatorClass: { 1928 ConditionalOperator *C = cast<ConditionalOperator>(E); 1929 1930 // Handle the GNU extension for missing LHS. 1931 if (Expr *lhsExpr = C->getLHS()) { 1932 // In C++, we can have a throw-expression, which has 'void' type. 1933 if (!lhsExpr->getType()->isVoidType()) 1934 if (Expr* LHS = EvalAddr(lhsExpr, refVars)) 1935 return LHS; 1936 } 1937 1938 // In C++, we can have a throw-expression, which has 'void' type. 1939 if (C->getRHS()->getType()->isVoidType()) 1940 return NULL; 1941 1942 return EvalAddr(C->getRHS(), refVars); 1943 } 1944 1945 case Stmt::BlockExprClass: 1946 if (cast<BlockExpr>(E)->getBlockDecl()->hasCaptures()) 1947 return E; // local block. 1948 return NULL; 1949 1950 case Stmt::AddrLabelExprClass: 1951 return E; // address of label. 1952 1953 // For casts, we need to handle conversions from arrays to 1954 // pointer values, and pointer-to-pointer conversions. 1955 case Stmt::ImplicitCastExprClass: 1956 case Stmt::CStyleCastExprClass: 1957 case Stmt::CXXFunctionalCastExprClass: { 1958 Expr* SubExpr = cast<CastExpr>(E)->getSubExpr(); 1959 QualType T = SubExpr->getType(); 1960 1961 if (SubExpr->getType()->isPointerType() || 1962 SubExpr->getType()->isBlockPointerType() || 1963 SubExpr->getType()->isObjCQualifiedIdType()) 1964 return EvalAddr(SubExpr, refVars); 1965 else if (T->isArrayType()) 1966 return EvalVal(SubExpr, refVars); 1967 else 1968 return 0; 1969 } 1970 1971 // C++ casts. For dynamic casts, static casts, and const casts, we 1972 // are always converting from a pointer-to-pointer, so we just blow 1973 // through the cast. In the case the dynamic cast doesn't fail (and 1974 // return NULL), we take the conservative route and report cases 1975 // where we return the address of a stack variable. For Reinterpre 1976 // FIXME: The comment about is wrong; we're not always converting 1977 // from pointer to pointer. I'm guessing that this code should also 1978 // handle references to objects. 1979 case Stmt::CXXStaticCastExprClass: 1980 case Stmt::CXXDynamicCastExprClass: 1981 case Stmt::CXXConstCastExprClass: 1982 case Stmt::CXXReinterpretCastExprClass: { 1983 Expr *S = cast<CXXNamedCastExpr>(E)->getSubExpr(); 1984 if (S->getType()->isPointerType() || S->getType()->isBlockPointerType()) 1985 return EvalAddr(S, refVars); 1986 else 1987 return NULL; 1988 } 1989 1990 // Everything else: we simply don't reason about them. 1991 default: 1992 return NULL; 1993 } 1994} 1995 1996 1997/// EvalVal - This function is complements EvalAddr in the mutual recursion. 1998/// See the comments for EvalAddr for more details. 1999static Expr *EvalVal(Expr *E, llvm::SmallVectorImpl<DeclRefExpr *> &refVars) { 2000do { 2001 // We should only be called for evaluating non-pointer expressions, or 2002 // expressions with a pointer type that are not used as references but instead 2003 // are l-values (e.g., DeclRefExpr with a pointer type). 2004 2005 // Our "symbolic interpreter" is just a dispatch off the currently 2006 // viewed AST node. We then recursively traverse the AST by calling 2007 // EvalAddr and EvalVal appropriately. 2008 switch (E->getStmtClass()) { 2009 case Stmt::ImplicitCastExprClass: { 2010 ImplicitCastExpr *IE = cast<ImplicitCastExpr>(E); 2011 if (IE->getValueKind() == VK_LValue) { 2012 E = IE->getSubExpr(); 2013 continue; 2014 } 2015 return NULL; 2016 } 2017 2018 case Stmt::DeclRefExprClass: { 2019 // When we hit a DeclRefExpr we are looking at code that refers to a 2020 // variable's name. If it's not a reference variable we check if it has 2021 // local storage within the function, and if so, return the expression. 2022 DeclRefExpr *DR = cast<DeclRefExpr>(E); 2023 2024 if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl())) 2025 if (V->hasLocalStorage()) { 2026 if (!V->getType()->isReferenceType()) 2027 return DR; 2028 2029 // Reference variable, follow through to the expression that 2030 // it points to. 2031 if (V->hasInit()) { 2032 // Add the reference variable to the "trail". 2033 refVars.push_back(DR); 2034 return EvalVal(V->getInit(), refVars); 2035 } 2036 } 2037 2038 return NULL; 2039 } 2040 2041 case Stmt::ParenExprClass: { 2042 // Ignore parentheses. 2043 E = cast<ParenExpr>(E)->getSubExpr(); 2044 continue; 2045 } 2046 2047 case Stmt::UnaryOperatorClass: { 2048 // The only unary operator that make sense to handle here 2049 // is Deref. All others don't resolve to a "name." This includes 2050 // handling all sorts of rvalues passed to a unary operator. 2051 UnaryOperator *U = cast<UnaryOperator>(E); 2052 2053 if (U->getOpcode() == UO_Deref) 2054 return EvalAddr(U->getSubExpr(), refVars); 2055 2056 return NULL; 2057 } 2058 2059 case Stmt::ArraySubscriptExprClass: { 2060 // Array subscripts are potential references to data on the stack. We 2061 // retrieve the DeclRefExpr* for the array variable if it indeed 2062 // has local storage. 2063 return EvalAddr(cast<ArraySubscriptExpr>(E)->getBase(), refVars); 2064 } 2065 2066 case Stmt::ConditionalOperatorClass: { 2067 // For conditional operators we need to see if either the LHS or RHS are 2068 // non-NULL Expr's. If one is non-NULL, we return it. 2069 ConditionalOperator *C = cast<ConditionalOperator>(E); 2070 2071 // Handle the GNU extension for missing LHS. 2072 if (Expr *lhsExpr = C->getLHS()) 2073 if (Expr *LHS = EvalVal(lhsExpr, refVars)) 2074 return LHS; 2075 2076 return EvalVal(C->getRHS(), refVars); 2077 } 2078 2079 // Accesses to members are potential references to data on the stack. 2080 case Stmt::MemberExprClass: { 2081 MemberExpr *M = cast<MemberExpr>(E); 2082 2083 // Check for indirect access. We only want direct field accesses. 2084 if (M->isArrow()) 2085 return NULL; 2086 2087 // Check whether the member type is itself a reference, in which case 2088 // we're not going to refer to the member, but to what the member refers to. 2089 if (M->getMemberDecl()->getType()->isReferenceType()) 2090 return NULL; 2091 2092 return EvalVal(M->getBase(), refVars); 2093 } 2094 2095 default: 2096 // Check that we don't return or take the address of a reference to a 2097 // temporary. This is only useful in C++. 2098 if (!E->isTypeDependent() && E->isRValue()) 2099 return E; 2100 2101 // Everything else: we simply don't reason about them. 2102 return NULL; 2103 } 2104} while (true); 2105} 2106 2107//===--- CHECK: Floating-Point comparisons (-Wfloat-equal) ---------------===// 2108 2109/// Check for comparisons of floating point operands using != and ==. 2110/// Issue a warning if these are no self-comparisons, as they are not likely 2111/// to do what the programmer intended. 2112void Sema::CheckFloatComparison(SourceLocation loc, Expr* lex, Expr *rex) { 2113 bool EmitWarning = true; 2114 2115 Expr* LeftExprSansParen = lex->IgnoreParenImpCasts(); 2116 Expr* RightExprSansParen = rex->IgnoreParenImpCasts(); 2117 2118 // Special case: check for x == x (which is OK). 2119 // Do not emit warnings for such cases. 2120 if (DeclRefExpr* DRL = dyn_cast<DeclRefExpr>(LeftExprSansParen)) 2121 if (DeclRefExpr* DRR = dyn_cast<DeclRefExpr>(RightExprSansParen)) 2122 if (DRL->getDecl() == DRR->getDecl()) 2123 EmitWarning = false; 2124 2125 2126 // Special case: check for comparisons against literals that can be exactly 2127 // represented by APFloat. In such cases, do not emit a warning. This 2128 // is a heuristic: often comparison against such literals are used to 2129 // detect if a value in a variable has not changed. This clearly can 2130 // lead to false negatives. 2131 if (EmitWarning) { 2132 if (FloatingLiteral* FLL = dyn_cast<FloatingLiteral>(LeftExprSansParen)) { 2133 if (FLL->isExact()) 2134 EmitWarning = false; 2135 } else 2136 if (FloatingLiteral* FLR = dyn_cast<FloatingLiteral>(RightExprSansParen)){ 2137 if (FLR->isExact()) 2138 EmitWarning = false; 2139 } 2140 } 2141 2142 // Check for comparisons with builtin types. 2143 if (EmitWarning) 2144 if (CallExpr* CL = dyn_cast<CallExpr>(LeftExprSansParen)) 2145 if (CL->isBuiltinCall(Context)) 2146 EmitWarning = false; 2147 2148 if (EmitWarning) 2149 if (CallExpr* CR = dyn_cast<CallExpr>(RightExprSansParen)) 2150 if (CR->isBuiltinCall(Context)) 2151 EmitWarning = false; 2152 2153 // Emit the diagnostic. 2154 if (EmitWarning) 2155 Diag(loc, diag::warn_floatingpoint_eq) 2156 << lex->getSourceRange() << rex->getSourceRange(); 2157} 2158 2159//===--- CHECK: Integer mixed-sign comparisons (-Wsign-compare) --------===// 2160//===--- CHECK: Lossy implicit conversions (-Wconversion) --------------===// 2161 2162namespace { 2163 2164/// Structure recording the 'active' range of an integer-valued 2165/// expression. 2166struct IntRange { 2167 /// The number of bits active in the int. 2168 unsigned Width; 2169 2170 /// True if the int is known not to have negative values. 2171 bool NonNegative; 2172 2173 IntRange(unsigned Width, bool NonNegative) 2174 : Width(Width), NonNegative(NonNegative) 2175 {} 2176 2177 /// Returns the range of the bool type. 2178 static IntRange forBoolType() { 2179 return IntRange(1, true); 2180 } 2181 2182 /// Returns the range of an opaque value of the given integral type. 2183 static IntRange forValueOfType(ASTContext &C, QualType T) { 2184 return forValueOfCanonicalType(C, 2185 T->getCanonicalTypeInternal().getTypePtr()); 2186 } 2187 2188 /// Returns the range of an opaque value of a canonical integral type. 2189 static IntRange forValueOfCanonicalType(ASTContext &C, const Type *T) { 2190 assert(T->isCanonicalUnqualified()); 2191 2192 if (const VectorType *VT = dyn_cast<VectorType>(T)) 2193 T = VT->getElementType().getTypePtr(); 2194 if (const ComplexType *CT = dyn_cast<ComplexType>(T)) 2195 T = CT->getElementType().getTypePtr(); 2196 2197 // For enum types, use the known bit width of the enumerators. 2198 if (const EnumType *ET = dyn_cast<EnumType>(T)) { 2199 EnumDecl *Enum = ET->getDecl(); 2200 if (!Enum->isDefinition()) 2201 return IntRange(C.getIntWidth(QualType(T, 0)), false); 2202 2203 unsigned NumPositive = Enum->getNumPositiveBits(); 2204 unsigned NumNegative = Enum->getNumNegativeBits(); 2205 2206 return IntRange(std::max(NumPositive, NumNegative), NumNegative == 0); 2207 } 2208 2209 const BuiltinType *BT = cast<BuiltinType>(T); 2210 assert(BT->isInteger()); 2211 2212 return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger()); 2213 } 2214 2215 /// Returns the "target" range of a canonical integral type, i.e. 2216 /// the range of values expressible in the type. 2217 /// 2218 /// This matches forValueOfCanonicalType except that enums have the 2219 /// full range of their type, not the range of their enumerators. 2220 static IntRange forTargetOfCanonicalType(ASTContext &C, const Type *T) { 2221 assert(T->isCanonicalUnqualified()); 2222 2223 if (const VectorType *VT = dyn_cast<VectorType>(T)) 2224 T = VT->getElementType().getTypePtr(); 2225 if (const ComplexType *CT = dyn_cast<ComplexType>(T)) 2226 T = CT->getElementType().getTypePtr(); 2227 if (const EnumType *ET = dyn_cast<EnumType>(T)) 2228 T = ET->getDecl()->getIntegerType().getTypePtr(); 2229 2230 const BuiltinType *BT = cast<BuiltinType>(T); 2231 assert(BT->isInteger()); 2232 2233 return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger()); 2234 } 2235 2236 /// Returns the supremum of two ranges: i.e. their conservative merge. 2237 static IntRange join(IntRange L, IntRange R) { 2238 return IntRange(std::max(L.Width, R.Width), 2239 L.NonNegative && R.NonNegative); 2240 } 2241 2242 /// Returns the infinum of two ranges: i.e. their aggressive merge. 2243 static IntRange meet(IntRange L, IntRange R) { 2244 return IntRange(std::min(L.Width, R.Width), 2245 L.NonNegative || R.NonNegative); 2246 } 2247}; 2248 2249IntRange GetValueRange(ASTContext &C, llvm::APSInt &value, unsigned MaxWidth) { 2250 if (value.isSigned() && value.isNegative()) 2251 return IntRange(value.getMinSignedBits(), false); 2252 2253 if (value.getBitWidth() > MaxWidth) 2254 value = value.trunc(MaxWidth); 2255 2256 // isNonNegative() just checks the sign bit without considering 2257 // signedness. 2258 return IntRange(value.getActiveBits(), true); 2259} 2260 2261IntRange GetValueRange(ASTContext &C, APValue &result, QualType Ty, 2262 unsigned MaxWidth) { 2263 if (result.isInt()) 2264 return GetValueRange(C, result.getInt(), MaxWidth); 2265 2266 if (result.isVector()) { 2267 IntRange R = GetValueRange(C, result.getVectorElt(0), Ty, MaxWidth); 2268 for (unsigned i = 1, e = result.getVectorLength(); i != e; ++i) { 2269 IntRange El = GetValueRange(C, result.getVectorElt(i), Ty, MaxWidth); 2270 R = IntRange::join(R, El); 2271 } 2272 return R; 2273 } 2274 2275 if (result.isComplexInt()) { 2276 IntRange R = GetValueRange(C, result.getComplexIntReal(), MaxWidth); 2277 IntRange I = GetValueRange(C, result.getComplexIntImag(), MaxWidth); 2278 return IntRange::join(R, I); 2279 } 2280 2281 // This can happen with lossless casts to intptr_t of "based" lvalues. 2282 // Assume it might use arbitrary bits. 2283 // FIXME: The only reason we need to pass the type in here is to get 2284 // the sign right on this one case. It would be nice if APValue 2285 // preserved this. 2286 assert(result.isLValue()); 2287 return IntRange(MaxWidth, Ty->isUnsignedIntegerType()); 2288} 2289 2290/// Pseudo-evaluate the given integer expression, estimating the 2291/// range of values it might take. 2292/// 2293/// \param MaxWidth - the width to which the value will be truncated 2294IntRange GetExprRange(ASTContext &C, Expr *E, unsigned MaxWidth) { 2295 E = E->IgnoreParens(); 2296 2297 // Try a full evaluation first. 2298 Expr::EvalResult result; 2299 if (E->Evaluate(result, C)) 2300 return GetValueRange(C, result.Val, E->getType(), MaxWidth); 2301 2302 // I think we only want to look through implicit casts here; if the 2303 // user has an explicit widening cast, we should treat the value as 2304 // being of the new, wider type. 2305 if (ImplicitCastExpr *CE = dyn_cast<ImplicitCastExpr>(E)) { 2306 if (CE->getCastKind() == CK_NoOp) 2307 return GetExprRange(C, CE->getSubExpr(), MaxWidth); 2308 2309 IntRange OutputTypeRange = IntRange::forValueOfType(C, CE->getType()); 2310 2311 bool isIntegerCast = (CE->getCastKind() == CK_IntegralCast); 2312 2313 // Assume that non-integer casts can span the full range of the type. 2314 if (!isIntegerCast) 2315 return OutputTypeRange; 2316 2317 IntRange SubRange 2318 = GetExprRange(C, CE->getSubExpr(), 2319 std::min(MaxWidth, OutputTypeRange.Width)); 2320 2321 // Bail out if the subexpr's range is as wide as the cast type. 2322 if (SubRange.Width >= OutputTypeRange.Width) 2323 return OutputTypeRange; 2324 2325 // Otherwise, we take the smaller width, and we're non-negative if 2326 // either the output type or the subexpr is. 2327 return IntRange(SubRange.Width, 2328 SubRange.NonNegative || OutputTypeRange.NonNegative); 2329 } 2330 2331 if (ConditionalOperator *CO = dyn_cast<ConditionalOperator>(E)) { 2332 // If we can fold the condition, just take that operand. 2333 bool CondResult; 2334 if (CO->getCond()->EvaluateAsBooleanCondition(CondResult, C)) 2335 return GetExprRange(C, CondResult ? CO->getTrueExpr() 2336 : CO->getFalseExpr(), 2337 MaxWidth); 2338 2339 // Otherwise, conservatively merge. 2340 IntRange L = GetExprRange(C, CO->getTrueExpr(), MaxWidth); 2341 IntRange R = GetExprRange(C, CO->getFalseExpr(), MaxWidth); 2342 return IntRange::join(L, R); 2343 } 2344 2345 if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) { 2346 switch (BO->getOpcode()) { 2347 2348 // Boolean-valued operations are single-bit and positive. 2349 case BO_LAnd: 2350 case BO_LOr: 2351 case BO_LT: 2352 case BO_GT: 2353 case BO_LE: 2354 case BO_GE: 2355 case BO_EQ: 2356 case BO_NE: 2357 return IntRange::forBoolType(); 2358 2359 // The type of these compound assignments is the type of the LHS, 2360 // so the RHS is not necessarily an integer. 2361 case BO_MulAssign: 2362 case BO_DivAssign: 2363 case BO_RemAssign: 2364 case BO_AddAssign: 2365 case BO_SubAssign: 2366 return IntRange::forValueOfType(C, E->getType()); 2367 2368 // Operations with opaque sources are black-listed. 2369 case BO_PtrMemD: 2370 case BO_PtrMemI: 2371 return IntRange::forValueOfType(C, E->getType()); 2372 2373 // Bitwise-and uses the *infinum* of the two source ranges. 2374 case BO_And: 2375 case BO_AndAssign: 2376 return IntRange::meet(GetExprRange(C, BO->getLHS(), MaxWidth), 2377 GetExprRange(C, BO->getRHS(), MaxWidth)); 2378 2379 // Left shift gets black-listed based on a judgement call. 2380 case BO_Shl: 2381 // ...except that we want to treat '1 << (blah)' as logically 2382 // positive. It's an important idiom. 2383 if (IntegerLiteral *I 2384 = dyn_cast<IntegerLiteral>(BO->getLHS()->IgnoreParenCasts())) { 2385 if (I->getValue() == 1) { 2386 IntRange R = IntRange::forValueOfType(C, E->getType()); 2387 return IntRange(R.Width, /*NonNegative*/ true); 2388 } 2389 } 2390 // fallthrough 2391 2392 case BO_ShlAssign: 2393 return IntRange::forValueOfType(C, E->getType()); 2394 2395 // Right shift by a constant can narrow its left argument. 2396 case BO_Shr: 2397 case BO_ShrAssign: { 2398 IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth); 2399 2400 // If the shift amount is a positive constant, drop the width by 2401 // that much. 2402 llvm::APSInt shift; 2403 if (BO->getRHS()->isIntegerConstantExpr(shift, C) && 2404 shift.isNonNegative()) { 2405 unsigned zext = shift.getZExtValue(); 2406 if (zext >= L.Width) 2407 L.Width = (L.NonNegative ? 0 : 1); 2408 else 2409 L.Width -= zext; 2410 } 2411 2412 return L; 2413 } 2414 2415 // Comma acts as its right operand. 2416 case BO_Comma: 2417 return GetExprRange(C, BO->getRHS(), MaxWidth); 2418 2419 // Black-list pointer subtractions. 2420 case BO_Sub: 2421 if (BO->getLHS()->getType()->isPointerType()) 2422 return IntRange::forValueOfType(C, E->getType()); 2423 // fallthrough 2424 2425 default: 2426 break; 2427 } 2428 2429 // Treat every other operator as if it were closed on the 2430 // narrowest type that encompasses both operands. 2431 IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth); 2432 IntRange R = GetExprRange(C, BO->getRHS(), MaxWidth); 2433 return IntRange::join(L, R); 2434 } 2435 2436 if (UnaryOperator *UO = dyn_cast<UnaryOperator>(E)) { 2437 switch (UO->getOpcode()) { 2438 // Boolean-valued operations are white-listed. 2439 case UO_LNot: 2440 return IntRange::forBoolType(); 2441 2442 // Operations with opaque sources are black-listed. 2443 case UO_Deref: 2444 case UO_AddrOf: // should be impossible 2445 return IntRange::forValueOfType(C, E->getType()); 2446 2447 default: 2448 return GetExprRange(C, UO->getSubExpr(), MaxWidth); 2449 } 2450 } 2451 2452 if (dyn_cast<OffsetOfExpr>(E)) { 2453 IntRange::forValueOfType(C, E->getType()); 2454 } 2455 2456 FieldDecl *BitField = E->getBitField(); 2457 if (BitField) { 2458 llvm::APSInt BitWidthAP = BitField->getBitWidth()->EvaluateAsInt(C); 2459 unsigned BitWidth = BitWidthAP.getZExtValue(); 2460 2461 return IntRange(BitWidth, BitField->getType()->isUnsignedIntegerType()); 2462 } 2463 2464 return IntRange::forValueOfType(C, E->getType()); 2465} 2466 2467IntRange GetExprRange(ASTContext &C, Expr *E) { 2468 return GetExprRange(C, E, C.getIntWidth(E->getType())); 2469} 2470 2471/// Checks whether the given value, which currently has the given 2472/// source semantics, has the same value when coerced through the 2473/// target semantics. 2474bool IsSameFloatAfterCast(const llvm::APFloat &value, 2475 const llvm::fltSemantics &Src, 2476 const llvm::fltSemantics &Tgt) { 2477 llvm::APFloat truncated = value; 2478 2479 bool ignored; 2480 truncated.convert(Src, llvm::APFloat::rmNearestTiesToEven, &ignored); 2481 truncated.convert(Tgt, llvm::APFloat::rmNearestTiesToEven, &ignored); 2482 2483 return truncated.bitwiseIsEqual(value); 2484} 2485 2486/// Checks whether the given value, which currently has the given 2487/// source semantics, has the same value when coerced through the 2488/// target semantics. 2489/// 2490/// The value might be a vector of floats (or a complex number). 2491bool IsSameFloatAfterCast(const APValue &value, 2492 const llvm::fltSemantics &Src, 2493 const llvm::fltSemantics &Tgt) { 2494 if (value.isFloat()) 2495 return IsSameFloatAfterCast(value.getFloat(), Src, Tgt); 2496 2497 if (value.isVector()) { 2498 for (unsigned i = 0, e = value.getVectorLength(); i != e; ++i) 2499 if (!IsSameFloatAfterCast(value.getVectorElt(i), Src, Tgt)) 2500 return false; 2501 return true; 2502 } 2503 2504 assert(value.isComplexFloat()); 2505 return (IsSameFloatAfterCast(value.getComplexFloatReal(), Src, Tgt) && 2506 IsSameFloatAfterCast(value.getComplexFloatImag(), Src, Tgt)); 2507} 2508 2509void AnalyzeImplicitConversions(Sema &S, Expr *E, SourceLocation CC); 2510 2511static bool IsZero(Sema &S, Expr *E) { 2512 // Suppress cases where we are comparing against an enum constant. 2513 if (const DeclRefExpr *DR = 2514 dyn_cast<DeclRefExpr>(E->IgnoreParenImpCasts())) 2515 if (isa<EnumConstantDecl>(DR->getDecl())) 2516 return false; 2517 2518 // Suppress cases where the '0' value is expanded from a macro. 2519 if (E->getLocStart().isMacroID()) 2520 return false; 2521 2522 llvm::APSInt Value; 2523 return E->isIntegerConstantExpr(Value, S.Context) && Value == 0; 2524} 2525 2526static bool HasEnumType(Expr *E) { 2527 // Strip off implicit integral promotions. 2528 while (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(E)) { 2529 if (ICE->getCastKind() != CK_IntegralCast && 2530 ICE->getCastKind() != CK_NoOp) 2531 break; 2532 E = ICE->getSubExpr(); 2533 } 2534 2535 return E->getType()->isEnumeralType(); 2536} 2537 2538void CheckTrivialUnsignedComparison(Sema &S, BinaryOperator *E) { 2539 BinaryOperatorKind op = E->getOpcode(); 2540 if (E->isValueDependent()) 2541 return; 2542 2543 if (op == BO_LT && IsZero(S, E->getRHS())) { 2544 S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison) 2545 << "< 0" << "false" << HasEnumType(E->getLHS()) 2546 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 2547 } else if (op == BO_GE && IsZero(S, E->getRHS())) { 2548 S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison) 2549 << ">= 0" << "true" << HasEnumType(E->getLHS()) 2550 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 2551 } else if (op == BO_GT && IsZero(S, E->getLHS())) { 2552 S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison) 2553 << "0 >" << "false" << HasEnumType(E->getRHS()) 2554 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 2555 } else if (op == BO_LE && IsZero(S, E->getLHS())) { 2556 S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison) 2557 << "0 <=" << "true" << HasEnumType(E->getRHS()) 2558 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 2559 } 2560} 2561 2562/// Analyze the operands of the given comparison. Implements the 2563/// fallback case from AnalyzeComparison. 2564void AnalyzeImpConvsInComparison(Sema &S, BinaryOperator *E) { 2565 AnalyzeImplicitConversions(S, E->getLHS(), E->getOperatorLoc()); 2566 AnalyzeImplicitConversions(S, E->getRHS(), E->getOperatorLoc()); 2567} 2568 2569/// \brief Implements -Wsign-compare. 2570/// 2571/// \param lex the left-hand expression 2572/// \param rex the right-hand expression 2573/// \param OpLoc the location of the joining operator 2574/// \param BinOpc binary opcode or 0 2575void AnalyzeComparison(Sema &S, BinaryOperator *E) { 2576 // The type the comparison is being performed in. 2577 QualType T = E->getLHS()->getType(); 2578 assert(S.Context.hasSameUnqualifiedType(T, E->getRHS()->getType()) 2579 && "comparison with mismatched types"); 2580 2581 // We don't do anything special if this isn't an unsigned integral 2582 // comparison: we're only interested in integral comparisons, and 2583 // signed comparisons only happen in cases we don't care to warn about. 2584 // 2585 // We also don't care about value-dependent expressions or expressions 2586 // whose result is a constant. 2587 if (!T->hasUnsignedIntegerRepresentation() 2588 || E->isValueDependent() || E->isIntegerConstantExpr(S.Context)) 2589 return AnalyzeImpConvsInComparison(S, E); 2590 2591 Expr *lex = E->getLHS()->IgnoreParenImpCasts(); 2592 Expr *rex = E->getRHS()->IgnoreParenImpCasts(); 2593 2594 // Check to see if one of the (unmodified) operands is of different 2595 // signedness. 2596 Expr *signedOperand, *unsignedOperand; 2597 if (lex->getType()->hasSignedIntegerRepresentation()) { 2598 assert(!rex->getType()->hasSignedIntegerRepresentation() && 2599 "unsigned comparison between two signed integer expressions?"); 2600 signedOperand = lex; 2601 unsignedOperand = rex; 2602 } else if (rex->getType()->hasSignedIntegerRepresentation()) { 2603 signedOperand = rex; 2604 unsignedOperand = lex; 2605 } else { 2606 CheckTrivialUnsignedComparison(S, E); 2607 return AnalyzeImpConvsInComparison(S, E); 2608 } 2609 2610 // Otherwise, calculate the effective range of the signed operand. 2611 IntRange signedRange = GetExprRange(S.Context, signedOperand); 2612 2613 // Go ahead and analyze implicit conversions in the operands. Note 2614 // that we skip the implicit conversions on both sides. 2615 AnalyzeImplicitConversions(S, lex, E->getOperatorLoc()); 2616 AnalyzeImplicitConversions(S, rex, E->getOperatorLoc()); 2617 2618 // If the signed range is non-negative, -Wsign-compare won't fire, 2619 // but we should still check for comparisons which are always true 2620 // or false. 2621 if (signedRange.NonNegative) 2622 return CheckTrivialUnsignedComparison(S, E); 2623 2624 // For (in)equality comparisons, if the unsigned operand is a 2625 // constant which cannot collide with a overflowed signed operand, 2626 // then reinterpreting the signed operand as unsigned will not 2627 // change the result of the comparison. 2628 if (E->isEqualityOp()) { 2629 unsigned comparisonWidth = S.Context.getIntWidth(T); 2630 IntRange unsignedRange = GetExprRange(S.Context, unsignedOperand); 2631 2632 // We should never be unable to prove that the unsigned operand is 2633 // non-negative. 2634 assert(unsignedRange.NonNegative && "unsigned range includes negative?"); 2635 2636 if (unsignedRange.Width < comparisonWidth) 2637 return; 2638 } 2639 2640 S.Diag(E->getOperatorLoc(), diag::warn_mixed_sign_comparison) 2641 << lex->getType() << rex->getType() 2642 << lex->getSourceRange() << rex->getSourceRange(); 2643} 2644 2645/// Analyzes an attempt to assign the given value to a bitfield. 2646/// 2647/// Returns true if there was something fishy about the attempt. 2648bool AnalyzeBitFieldAssignment(Sema &S, FieldDecl *Bitfield, Expr *Init, 2649 SourceLocation InitLoc) { 2650 assert(Bitfield->isBitField()); 2651 if (Bitfield->isInvalidDecl()) 2652 return false; 2653 2654 // White-list bool bitfields. 2655 if (Bitfield->getType()->isBooleanType()) 2656 return false; 2657 2658 // Ignore value- or type-dependent expressions. 2659 if (Bitfield->getBitWidth()->isValueDependent() || 2660 Bitfield->getBitWidth()->isTypeDependent() || 2661 Init->isValueDependent() || 2662 Init->isTypeDependent()) 2663 return false; 2664 2665 Expr *OriginalInit = Init->IgnoreParenImpCasts(); 2666 2667 llvm::APSInt Width(32); 2668 Expr::EvalResult InitValue; 2669 if (!Bitfield->getBitWidth()->isIntegerConstantExpr(Width, S.Context) || 2670 !OriginalInit->Evaluate(InitValue, S.Context) || 2671 !InitValue.Val.isInt()) 2672 return false; 2673 2674 const llvm::APSInt &Value = InitValue.Val.getInt(); 2675 unsigned OriginalWidth = Value.getBitWidth(); 2676 unsigned FieldWidth = Width.getZExtValue(); 2677 2678 if (OriginalWidth <= FieldWidth) 2679 return false; 2680 2681 llvm::APSInt TruncatedValue = Value.trunc(FieldWidth); 2682 2683 // It's fairly common to write values into signed bitfields 2684 // that, if sign-extended, would end up becoming a different 2685 // value. We don't want to warn about that. 2686 if (Value.isSigned() && Value.isNegative()) 2687 TruncatedValue = TruncatedValue.sext(OriginalWidth); 2688 else 2689 TruncatedValue = TruncatedValue.zext(OriginalWidth); 2690 2691 if (Value == TruncatedValue) 2692 return false; 2693 2694 std::string PrettyValue = Value.toString(10); 2695 std::string PrettyTrunc = TruncatedValue.toString(10); 2696 2697 S.Diag(InitLoc, diag::warn_impcast_bitfield_precision_constant) 2698 << PrettyValue << PrettyTrunc << OriginalInit->getType() 2699 << Init->getSourceRange(); 2700 2701 return true; 2702} 2703 2704/// Analyze the given simple or compound assignment for warning-worthy 2705/// operations. 2706void AnalyzeAssignment(Sema &S, BinaryOperator *E) { 2707 // Just recurse on the LHS. 2708 AnalyzeImplicitConversions(S, E->getLHS(), E->getOperatorLoc()); 2709 2710 // We want to recurse on the RHS as normal unless we're assigning to 2711 // a bitfield. 2712 if (FieldDecl *Bitfield = E->getLHS()->getBitField()) { 2713 if (AnalyzeBitFieldAssignment(S, Bitfield, E->getRHS(), 2714 E->getOperatorLoc())) { 2715 // Recurse, ignoring any implicit conversions on the RHS. 2716 return AnalyzeImplicitConversions(S, E->getRHS()->IgnoreParenImpCasts(), 2717 E->getOperatorLoc()); 2718 } 2719 } 2720 2721 AnalyzeImplicitConversions(S, E->getRHS(), E->getOperatorLoc()); 2722} 2723 2724/// Diagnose an implicit cast; purely a helper for CheckImplicitConversion. 2725void DiagnoseImpCast(Sema &S, Expr *E, QualType T, SourceLocation CContext, 2726 unsigned diag) { 2727 S.Diag(E->getExprLoc(), diag) 2728 << E->getType() << T << E->getSourceRange() << SourceRange(CContext); 2729} 2730 2731std::string PrettyPrintInRange(const llvm::APSInt &Value, IntRange Range) { 2732 if (!Range.Width) return "0"; 2733 2734 llvm::APSInt ValueInRange = Value; 2735 ValueInRange.setIsSigned(!Range.NonNegative); 2736 ValueInRange = ValueInRange.trunc(Range.Width); 2737 return ValueInRange.toString(10); 2738} 2739 2740void CheckImplicitConversion(Sema &S, Expr *E, QualType T, 2741 SourceLocation CC, bool *ICContext = 0) { 2742 if (E->isTypeDependent() || E->isValueDependent()) return; 2743 2744 const Type *Source = S.Context.getCanonicalType(E->getType()).getTypePtr(); 2745 const Type *Target = S.Context.getCanonicalType(T).getTypePtr(); 2746 if (Source == Target) return; 2747 if (Target->isDependentType()) return; 2748 2749 // If the conversion context location is invalid or instantiated 2750 // from a system macro, don't complain. 2751 if (CC.isInvalid() || 2752 (CC.isMacroID() && S.Context.getSourceManager().isInSystemHeader( 2753 S.Context.getSourceManager().getSpellingLoc(CC)))) 2754 return; 2755 2756 // Never diagnose implicit casts to bool. 2757 if (Target->isSpecificBuiltinType(BuiltinType::Bool)) 2758 return; 2759 2760 // Strip vector types. 2761 if (isa<VectorType>(Source)) { 2762 if (!isa<VectorType>(Target)) 2763 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_vector_scalar); 2764 2765 Source = cast<VectorType>(Source)->getElementType().getTypePtr(); 2766 Target = cast<VectorType>(Target)->getElementType().getTypePtr(); 2767 } 2768 2769 // Strip complex types. 2770 if (isa<ComplexType>(Source)) { 2771 if (!isa<ComplexType>(Target)) 2772 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_complex_scalar); 2773 2774 Source = cast<ComplexType>(Source)->getElementType().getTypePtr(); 2775 Target = cast<ComplexType>(Target)->getElementType().getTypePtr(); 2776 } 2777 2778 const BuiltinType *SourceBT = dyn_cast<BuiltinType>(Source); 2779 const BuiltinType *TargetBT = dyn_cast<BuiltinType>(Target); 2780 2781 // If the source is floating point... 2782 if (SourceBT && SourceBT->isFloatingPoint()) { 2783 // ...and the target is floating point... 2784 if (TargetBT && TargetBT->isFloatingPoint()) { 2785 // ...then warn if we're dropping FP rank. 2786 2787 // Builtin FP kinds are ordered by increasing FP rank. 2788 if (SourceBT->getKind() > TargetBT->getKind()) { 2789 // Don't warn about float constants that are precisely 2790 // representable in the target type. 2791 Expr::EvalResult result; 2792 if (E->Evaluate(result, S.Context)) { 2793 // Value might be a float, a float vector, or a float complex. 2794 if (IsSameFloatAfterCast(result.Val, 2795 S.Context.getFloatTypeSemantics(QualType(TargetBT, 0)), 2796 S.Context.getFloatTypeSemantics(QualType(SourceBT, 0)))) 2797 return; 2798 } 2799 2800 DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_float_precision); 2801 } 2802 return; 2803 } 2804 2805 // If the target is integral, always warn. 2806 if ((TargetBT && TargetBT->isInteger())) { 2807 Expr *InnerE = E->IgnoreParenImpCasts(); 2808 if (FloatingLiteral *LiteralExpr = dyn_cast<FloatingLiteral>(InnerE)) { 2809 DiagnoseImpCast(S, LiteralExpr, T, CC, 2810 diag::warn_impcast_literal_float_to_integer); 2811 } else { 2812 DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_float_integer); 2813 } 2814 } 2815 2816 return; 2817 } 2818 2819 if (!Source->isIntegerType() || !Target->isIntegerType()) 2820 return; 2821 2822 IntRange SourceRange = GetExprRange(S.Context, E); 2823 IntRange TargetRange = IntRange::forTargetOfCanonicalType(S.Context, Target); 2824 2825 if (SourceRange.Width > TargetRange.Width) { 2826 // If the source is a constant, use a default-on diagnostic. 2827 // TODO: this should happen for bitfield stores, too. 2828 llvm::APSInt Value(32); 2829 if (E->isIntegerConstantExpr(Value, S.Context)) { 2830 std::string PrettySourceValue = Value.toString(10); 2831 std::string PrettyTargetValue = PrettyPrintInRange(Value, TargetRange); 2832 2833 S.Diag(E->getExprLoc(), diag::warn_impcast_integer_precision_constant) 2834 << PrettySourceValue << PrettyTargetValue 2835 << E->getType() << T << E->getSourceRange() << clang::SourceRange(CC); 2836 return; 2837 } 2838 2839 // People want to build with -Wshorten-64-to-32 and not -Wconversion 2840 // and by god we'll let them. 2841 if (SourceRange.Width == 64 && TargetRange.Width == 32) 2842 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_integer_64_32); 2843 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_integer_precision); 2844 } 2845 2846 if ((TargetRange.NonNegative && !SourceRange.NonNegative) || 2847 (!TargetRange.NonNegative && SourceRange.NonNegative && 2848 SourceRange.Width == TargetRange.Width)) { 2849 unsigned DiagID = diag::warn_impcast_integer_sign; 2850 2851 // Traditionally, gcc has warned about this under -Wsign-compare. 2852 // We also want to warn about it in -Wconversion. 2853 // So if -Wconversion is off, use a completely identical diagnostic 2854 // in the sign-compare group. 2855 // The conditional-checking code will 2856 if (ICContext) { 2857 DiagID = diag::warn_impcast_integer_sign_conditional; 2858 *ICContext = true; 2859 } 2860 2861 return DiagnoseImpCast(S, E, T, CC, DiagID); 2862 } 2863 2864 // Diagnose conversions between different enumeration types. 2865 if (const EnumType *SourceEnum = Source->getAs<EnumType>()) 2866 if (const EnumType *TargetEnum = Target->getAs<EnumType>()) 2867 if ((SourceEnum->getDecl()->getIdentifier() || 2868 SourceEnum->getDecl()->getTypedefForAnonDecl()) && 2869 (TargetEnum->getDecl()->getIdentifier() || 2870 TargetEnum->getDecl()->getTypedefForAnonDecl()) && 2871 SourceEnum != TargetEnum) 2872 return DiagnoseImpCast(S, E, T, CC, 2873 diag::warn_impcast_different_enum_types); 2874 2875 return; 2876} 2877 2878void CheckConditionalOperator(Sema &S, ConditionalOperator *E, QualType T); 2879 2880void CheckConditionalOperand(Sema &S, Expr *E, QualType T, 2881 SourceLocation CC, bool &ICContext) { 2882 E = E->IgnoreParenImpCasts(); 2883 2884 if (isa<ConditionalOperator>(E)) 2885 return CheckConditionalOperator(S, cast<ConditionalOperator>(E), T); 2886 2887 AnalyzeImplicitConversions(S, E, CC); 2888 if (E->getType() != T) 2889 return CheckImplicitConversion(S, E, T, CC, &ICContext); 2890 return; 2891} 2892 2893void CheckConditionalOperator(Sema &S, ConditionalOperator *E, QualType T) { 2894 SourceLocation CC = E->getQuestionLoc(); 2895 2896 AnalyzeImplicitConversions(S, E->getCond(), CC); 2897 2898 bool Suspicious = false; 2899 CheckConditionalOperand(S, E->getTrueExpr(), T, CC, Suspicious); 2900 CheckConditionalOperand(S, E->getFalseExpr(), T, CC, Suspicious); 2901 2902 // If -Wconversion would have warned about either of the candidates 2903 // for a signedness conversion to the context type... 2904 if (!Suspicious) return; 2905 2906 // ...but it's currently ignored... 2907 if (S.Diags.getDiagnosticLevel(diag::warn_impcast_integer_sign_conditional, 2908 CC)) 2909 return; 2910 2911 // ...and -Wsign-compare isn't... 2912 if (!S.Diags.getDiagnosticLevel(diag::warn_mixed_sign_conditional, CC)) 2913 return; 2914 2915 // ...then check whether it would have warned about either of the 2916 // candidates for a signedness conversion to the condition type. 2917 if (E->getType() != T) { 2918 Suspicious = false; 2919 CheckImplicitConversion(S, E->getTrueExpr()->IgnoreParenImpCasts(), 2920 E->getType(), CC, &Suspicious); 2921 if (!Suspicious) 2922 CheckImplicitConversion(S, E->getFalseExpr()->IgnoreParenImpCasts(), 2923 E->getType(), CC, &Suspicious); 2924 if (!Suspicious) 2925 return; 2926 } 2927 2928 // If so, emit a diagnostic under -Wsign-compare. 2929 Expr *lex = E->getTrueExpr()->IgnoreParenImpCasts(); 2930 Expr *rex = E->getFalseExpr()->IgnoreParenImpCasts(); 2931 S.Diag(E->getQuestionLoc(), diag::warn_mixed_sign_conditional) 2932 << lex->getType() << rex->getType() 2933 << lex->getSourceRange() << rex->getSourceRange(); 2934} 2935 2936/// AnalyzeImplicitConversions - Find and report any interesting 2937/// implicit conversions in the given expression. There are a couple 2938/// of competing diagnostics here, -Wconversion and -Wsign-compare. 2939void AnalyzeImplicitConversions(Sema &S, Expr *OrigE, SourceLocation CC) { 2940 QualType T = OrigE->getType(); 2941 Expr *E = OrigE->IgnoreParenImpCasts(); 2942 2943 // For conditional operators, we analyze the arguments as if they 2944 // were being fed directly into the output. 2945 if (isa<ConditionalOperator>(E)) { 2946 ConditionalOperator *CO = cast<ConditionalOperator>(E); 2947 CheckConditionalOperator(S, CO, T); 2948 return; 2949 } 2950 2951 // Go ahead and check any implicit conversions we might have skipped. 2952 // The non-canonical typecheck is just an optimization; 2953 // CheckImplicitConversion will filter out dead implicit conversions. 2954 if (E->getType() != T) 2955 CheckImplicitConversion(S, E, T, CC); 2956 2957 // Now continue drilling into this expression. 2958 2959 // Skip past explicit casts. 2960 if (isa<ExplicitCastExpr>(E)) { 2961 E = cast<ExplicitCastExpr>(E)->getSubExpr()->IgnoreParenImpCasts(); 2962 return AnalyzeImplicitConversions(S, E, CC); 2963 } 2964 2965 if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) { 2966 // Do a somewhat different check with comparison operators. 2967 if (BO->isComparisonOp()) 2968 return AnalyzeComparison(S, BO); 2969 2970 // And with assignments and compound assignments. 2971 if (BO->isAssignmentOp()) 2972 return AnalyzeAssignment(S, BO); 2973 } 2974 2975 // These break the otherwise-useful invariant below. Fortunately, 2976 // we don't really need to recurse into them, because any internal 2977 // expressions should have been analyzed already when they were 2978 // built into statements. 2979 if (isa<StmtExpr>(E)) return; 2980 2981 // Don't descend into unevaluated contexts. 2982 if (isa<SizeOfAlignOfExpr>(E)) return; 2983 2984 // Now just recurse over the expression's children. 2985 CC = E->getExprLoc(); 2986 for (Stmt::child_range I = E->children(); I; ++I) 2987 AnalyzeImplicitConversions(S, cast<Expr>(*I), CC); 2988} 2989 2990} // end anonymous namespace 2991 2992/// Diagnoses "dangerous" implicit conversions within the given 2993/// expression (which is a full expression). Implements -Wconversion 2994/// and -Wsign-compare. 2995/// 2996/// \param CC the "context" location of the implicit conversion, i.e. 2997/// the most location of the syntactic entity requiring the implicit 2998/// conversion 2999void Sema::CheckImplicitConversions(Expr *E, SourceLocation CC) { 3000 // Don't diagnose in unevaluated contexts. 3001 if (ExprEvalContexts.back().Context == Sema::Unevaluated) 3002 return; 3003 3004 // Don't diagnose for value- or type-dependent expressions. 3005 if (E->isTypeDependent() || E->isValueDependent()) 3006 return; 3007 3008 // This is not the right CC for (e.g.) a variable initialization. 3009 AnalyzeImplicitConversions(*this, E, CC); 3010} 3011 3012void Sema::CheckBitFieldInitialization(SourceLocation InitLoc, 3013 FieldDecl *BitField, 3014 Expr *Init) { 3015 (void) AnalyzeBitFieldAssignment(*this, BitField, Init, InitLoc); 3016} 3017 3018/// CheckParmsForFunctionDef - Check that the parameters of the given 3019/// function are appropriate for the definition of a function. This 3020/// takes care of any checks that cannot be performed on the 3021/// declaration itself, e.g., that the types of each of the function 3022/// parameters are complete. 3023bool Sema::CheckParmsForFunctionDef(ParmVarDecl **P, ParmVarDecl **PEnd, 3024 bool CheckParameterNames) { 3025 bool HasInvalidParm = false; 3026 for (; P != PEnd; ++P) { 3027 ParmVarDecl *Param = *P; 3028 3029 // C99 6.7.5.3p4: the parameters in a parameter type list in a 3030 // function declarator that is part of a function definition of 3031 // that function shall not have incomplete type. 3032 // 3033 // This is also C++ [dcl.fct]p6. 3034 if (!Param->isInvalidDecl() && 3035 RequireCompleteType(Param->getLocation(), Param->getType(), 3036 diag::err_typecheck_decl_incomplete_type)) { 3037 Param->setInvalidDecl(); 3038 HasInvalidParm = true; 3039 } 3040 3041 // C99 6.9.1p5: If the declarator includes a parameter type list, the 3042 // declaration of each parameter shall include an identifier. 3043 if (CheckParameterNames && 3044 Param->getIdentifier() == 0 && 3045 !Param->isImplicit() && 3046 !getLangOptions().CPlusPlus) 3047 Diag(Param->getLocation(), diag::err_parameter_name_omitted); 3048 3049 // C99 6.7.5.3p12: 3050 // If the function declarator is not part of a definition of that 3051 // function, parameters may have incomplete type and may use the [*] 3052 // notation in their sequences of declarator specifiers to specify 3053 // variable length array types. 3054 QualType PType = Param->getOriginalType(); 3055 if (const ArrayType *AT = Context.getAsArrayType(PType)) { 3056 if (AT->getSizeModifier() == ArrayType::Star) { 3057 // FIXME: This diagnosic should point the the '[*]' if source-location 3058 // information is added for it. 3059 Diag(Param->getLocation(), diag::err_array_star_in_function_definition); 3060 } 3061 } 3062 } 3063 3064 return HasInvalidParm; 3065} 3066 3067/// CheckCastAlign - Implements -Wcast-align, which warns when a 3068/// pointer cast increases the alignment requirements. 3069void Sema::CheckCastAlign(Expr *Op, QualType T, SourceRange TRange) { 3070 // This is actually a lot of work to potentially be doing on every 3071 // cast; don't do it if we're ignoring -Wcast_align (as is the default). 3072 if (getDiagnostics().getDiagnosticLevel(diag::warn_cast_align, 3073 TRange.getBegin()) 3074 == Diagnostic::Ignored) 3075 return; 3076 3077 // Ignore dependent types. 3078 if (T->isDependentType() || Op->getType()->isDependentType()) 3079 return; 3080 3081 // Require that the destination be a pointer type. 3082 const PointerType *DestPtr = T->getAs<PointerType>(); 3083 if (!DestPtr) return; 3084 3085 // If the destination has alignment 1, we're done. 3086 QualType DestPointee = DestPtr->getPointeeType(); 3087 if (DestPointee->isIncompleteType()) return; 3088 CharUnits DestAlign = Context.getTypeAlignInChars(DestPointee); 3089 if (DestAlign.isOne()) return; 3090 3091 // Require that the source be a pointer type. 3092 const PointerType *SrcPtr = Op->getType()->getAs<PointerType>(); 3093 if (!SrcPtr) return; 3094 QualType SrcPointee = SrcPtr->getPointeeType(); 3095 3096 // Whitelist casts from cv void*. We already implicitly 3097 // whitelisted casts to cv void*, since they have alignment 1. 3098 // Also whitelist casts involving incomplete types, which implicitly 3099 // includes 'void'. 3100 if (SrcPointee->isIncompleteType()) return; 3101 3102 CharUnits SrcAlign = Context.getTypeAlignInChars(SrcPointee); 3103 if (SrcAlign >= DestAlign) return; 3104 3105 Diag(TRange.getBegin(), diag::warn_cast_align) 3106 << Op->getType() << T 3107 << static_cast<unsigned>(SrcAlign.getQuantity()) 3108 << static_cast<unsigned>(DestAlign.getQuantity()) 3109 << TRange << Op->getSourceRange(); 3110} 3111 3112void Sema::CheckArrayAccess(const clang::ArraySubscriptExpr *E) { 3113 const Expr *BaseExpr = E->getBase()->IgnoreParenImpCasts(); 3114 const ConstantArrayType *ArrayTy = 3115 Context.getAsConstantArrayType(BaseExpr->getType()); 3116 if (!ArrayTy) 3117 return; 3118 3119 const Expr *IndexExpr = E->getIdx(); 3120 if (IndexExpr->isValueDependent()) 3121 return; 3122 llvm::APSInt index; 3123 if (!IndexExpr->isIntegerConstantExpr(index, Context)) 3124 return; 3125 3126 if (!index.isNegative()) { 3127 llvm::APInt size = ArrayTy->getSize(); 3128 if (!size.isStrictlyPositive()) 3129 return; 3130 if (size.getBitWidth() > index.getBitWidth()) 3131 index = index.sext(size.getBitWidth()); 3132 else if (size.getBitWidth() < index.getBitWidth()) 3133 size = size.sext(index.getBitWidth()); 3134 3135 if (index.slt(size)) 3136 return; 3137 3138 Diag(E->getBase()->getLocStart(), diag::warn_array_index_exceeds_bounds) 3139 << index.toString(10, true) << size.toString(10, true) 3140 << IndexExpr->getSourceRange(); 3141 } else { 3142 Diag(E->getBase()->getLocStart(), diag::warn_array_index_precedes_bounds) 3143 << index.toString(10, true) << IndexExpr->getSourceRange(); 3144 } 3145 3146 const NamedDecl *ND = NULL; 3147 if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(BaseExpr)) 3148 ND = dyn_cast<NamedDecl>(DRE->getDecl()); 3149 if (const MemberExpr *ME = dyn_cast<MemberExpr>(BaseExpr)) 3150 ND = dyn_cast<NamedDecl>(ME->getMemberDecl()); 3151 if (ND) 3152 Diag(ND->getLocStart(), diag::note_array_index_out_of_bounds) 3153 << ND->getDeclName(); 3154} 3155 3156