SemaChecking.cpp revision 26e10bea3d2e9d2979194890e51b98ecea165a96
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 assert(!shift && "cannot shift polynomial types!"); 202 return (8 << (int)quad) - 1; 203 case 6: // poly16 204 assert(!shift && "cannot shift polynomial types!"); 205 return (4 << (int)quad) - 1; 206 case 7: // float16 207 assert(!shift && "cannot shift float types!"); 208 return (4 << (int)quad) - 1; 209 } 210 return 0; 211} 212 213bool Sema::CheckARMBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) { 214 llvm::APSInt Result; 215 216 unsigned mask = 0; 217 unsigned TV = 0; 218 switch (BuiltinID) { 219#define GET_NEON_OVERLOAD_CHECK 220#include "clang/Basic/arm_neon.inc" 221#undef GET_NEON_OVERLOAD_CHECK 222 } 223 224 // For NEON intrinsics which are overloaded on vector element type, validate 225 // the immediate which specifies which variant to emit. 226 if (mask) { 227 unsigned ArgNo = TheCall->getNumArgs()-1; 228 if (SemaBuiltinConstantArg(TheCall, ArgNo, Result)) 229 return true; 230 231 TV = Result.getLimitedValue(32); 232 if ((TV > 31) || (mask & (1 << TV)) == 0) 233 return Diag(TheCall->getLocStart(), diag::err_invalid_neon_type_code) 234 << TheCall->getArg(ArgNo)->getSourceRange(); 235 } 236 237 // For NEON intrinsics which take an immediate value as part of the 238 // instruction, range check them here. 239 unsigned i = 0, l = 0, u = 0; 240 switch (BuiltinID) { 241 default: return false; 242 case ARM::BI__builtin_arm_ssat: i = 1; l = 1; u = 31; break; 243 case ARM::BI__builtin_arm_usat: i = 1; u = 31; break; 244 case ARM::BI__builtin_arm_vcvtr_f: 245 case ARM::BI__builtin_arm_vcvtr_d: i = 1; u = 1; break; 246#define GET_NEON_IMMEDIATE_CHECK 247#include "clang/Basic/arm_neon.inc" 248#undef GET_NEON_IMMEDIATE_CHECK 249 }; 250 251 // Check that the immediate argument is actually a constant. 252 if (SemaBuiltinConstantArg(TheCall, i, Result)) 253 return true; 254 255 // Range check against the upper/lower values for this isntruction. 256 unsigned Val = Result.getZExtValue(); 257 if (Val < l || Val > (u + l)) 258 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 259 << l << u+l << TheCall->getArg(i)->getSourceRange(); 260 261 // FIXME: VFP Intrinsics should error if VFP not present. 262 return false; 263} 264 265/// CheckFunctionCall - Check a direct function call for various correctness 266/// and safety properties not strictly enforced by the C type system. 267bool Sema::CheckFunctionCall(FunctionDecl *FDecl, CallExpr *TheCall) { 268 // Get the IdentifierInfo* for the called function. 269 IdentifierInfo *FnInfo = FDecl->getIdentifier(); 270 271 // None of the checks below are needed for functions that don't have 272 // simple names (e.g., C++ conversion functions). 273 if (!FnInfo) 274 return false; 275 276 // FIXME: This mechanism should be abstracted to be less fragile and 277 // more efficient. For example, just map function ids to custom 278 // handlers. 279 280 // Printf and scanf checking. 281 for (specific_attr_iterator<FormatAttr> 282 i = FDecl->specific_attr_begin<FormatAttr>(), 283 e = FDecl->specific_attr_end<FormatAttr>(); i != e ; ++i) { 284 285 const FormatAttr *Format = *i; 286 const bool b = Format->getType() == "scanf"; 287 if (b || CheckablePrintfAttr(Format, TheCall)) { 288 bool HasVAListArg = Format->getFirstArg() == 0; 289 CheckPrintfScanfArguments(TheCall, HasVAListArg, 290 Format->getFormatIdx() - 1, 291 HasVAListArg ? 0 : Format->getFirstArg() - 1, 292 !b); 293 } 294 } 295 296 for (specific_attr_iterator<NonNullAttr> 297 i = FDecl->specific_attr_begin<NonNullAttr>(), 298 e = FDecl->specific_attr_end<NonNullAttr>(); i != e; ++i) { 299 CheckNonNullArguments(*i, TheCall); 300 } 301 302 return false; 303} 304 305bool Sema::CheckBlockCall(NamedDecl *NDecl, CallExpr *TheCall) { 306 // Printf checking. 307 const FormatAttr *Format = NDecl->getAttr<FormatAttr>(); 308 if (!Format) 309 return false; 310 311 const VarDecl *V = dyn_cast<VarDecl>(NDecl); 312 if (!V) 313 return false; 314 315 QualType Ty = V->getType(); 316 if (!Ty->isBlockPointerType()) 317 return false; 318 319 const bool b = Format->getType() == "scanf"; 320 if (!b && !CheckablePrintfAttr(Format, TheCall)) 321 return false; 322 323 bool HasVAListArg = Format->getFirstArg() == 0; 324 CheckPrintfScanfArguments(TheCall, HasVAListArg, Format->getFormatIdx() - 1, 325 HasVAListArg ? 0 : Format->getFirstArg() - 1, !b); 326 327 return false; 328} 329 330/// SemaBuiltinAtomicOverloaded - We have a call to a function like 331/// __sync_fetch_and_add, which is an overloaded function based on the pointer 332/// type of its first argument. The main ActOnCallExpr routines have already 333/// promoted the types of arguments because all of these calls are prototyped as 334/// void(...). 335/// 336/// This function goes through and does final semantic checking for these 337/// builtins, 338ExprResult 339Sema::SemaBuiltinAtomicOverloaded(ExprResult TheCallResult) { 340 CallExpr *TheCall = (CallExpr *)TheCallResult.get(); 341 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 342 FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl()); 343 344 // Ensure that we have at least one argument to do type inference from. 345 if (TheCall->getNumArgs() < 1) { 346 Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args_at_least) 347 << 0 << 1 << TheCall->getNumArgs() 348 << TheCall->getCallee()->getSourceRange(); 349 return ExprError(); 350 } 351 352 // Inspect the first argument of the atomic builtin. This should always be 353 // a pointer type, whose element is an integral scalar or pointer type. 354 // Because it is a pointer type, we don't have to worry about any implicit 355 // casts here. 356 // FIXME: We don't allow floating point scalars as input. 357 Expr *FirstArg = TheCall->getArg(0); 358 if (!FirstArg->getType()->isPointerType()) { 359 Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer) 360 << FirstArg->getType() << FirstArg->getSourceRange(); 361 return ExprError(); 362 } 363 364 QualType ValType = 365 FirstArg->getType()->getAs<PointerType>()->getPointeeType(); 366 if (!ValType->isIntegerType() && !ValType->isAnyPointerType() && 367 !ValType->isBlockPointerType()) { 368 Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer_intptr) 369 << FirstArg->getType() << FirstArg->getSourceRange(); 370 return ExprError(); 371 } 372 373 // The majority of builtins return a value, but a few have special return 374 // types, so allow them to override appropriately below. 375 QualType ResultType = ValType; 376 377 // We need to figure out which concrete builtin this maps onto. For example, 378 // __sync_fetch_and_add with a 2 byte object turns into 379 // __sync_fetch_and_add_2. 380#define BUILTIN_ROW(x) \ 381 { Builtin::BI##x##_1, Builtin::BI##x##_2, Builtin::BI##x##_4, \ 382 Builtin::BI##x##_8, Builtin::BI##x##_16 } 383 384 static const unsigned BuiltinIndices[][5] = { 385 BUILTIN_ROW(__sync_fetch_and_add), 386 BUILTIN_ROW(__sync_fetch_and_sub), 387 BUILTIN_ROW(__sync_fetch_and_or), 388 BUILTIN_ROW(__sync_fetch_and_and), 389 BUILTIN_ROW(__sync_fetch_and_xor), 390 391 BUILTIN_ROW(__sync_add_and_fetch), 392 BUILTIN_ROW(__sync_sub_and_fetch), 393 BUILTIN_ROW(__sync_and_and_fetch), 394 BUILTIN_ROW(__sync_or_and_fetch), 395 BUILTIN_ROW(__sync_xor_and_fetch), 396 397 BUILTIN_ROW(__sync_val_compare_and_swap), 398 BUILTIN_ROW(__sync_bool_compare_and_swap), 399 BUILTIN_ROW(__sync_lock_test_and_set), 400 BUILTIN_ROW(__sync_lock_release) 401 }; 402#undef BUILTIN_ROW 403 404 // Determine the index of the size. 405 unsigned SizeIndex; 406 switch (Context.getTypeSizeInChars(ValType).getQuantity()) { 407 case 1: SizeIndex = 0; break; 408 case 2: SizeIndex = 1; break; 409 case 4: SizeIndex = 2; break; 410 case 8: SizeIndex = 3; break; 411 case 16: SizeIndex = 4; break; 412 default: 413 Diag(DRE->getLocStart(), diag::err_atomic_builtin_pointer_size) 414 << FirstArg->getType() << FirstArg->getSourceRange(); 415 return ExprError(); 416 } 417 418 // Each of these builtins has one pointer argument, followed by some number of 419 // values (0, 1 or 2) followed by a potentially empty varags list of stuff 420 // that we ignore. Find out which row of BuiltinIndices to read from as well 421 // as the number of fixed args. 422 unsigned BuiltinID = FDecl->getBuiltinID(); 423 unsigned BuiltinIndex, NumFixed = 1; 424 switch (BuiltinID) { 425 default: assert(0 && "Unknown overloaded atomic builtin!"); 426 case Builtin::BI__sync_fetch_and_add: BuiltinIndex = 0; break; 427 case Builtin::BI__sync_fetch_and_sub: BuiltinIndex = 1; break; 428 case Builtin::BI__sync_fetch_and_or: BuiltinIndex = 2; break; 429 case Builtin::BI__sync_fetch_and_and: BuiltinIndex = 3; break; 430 case Builtin::BI__sync_fetch_and_xor: BuiltinIndex = 4; break; 431 432 case Builtin::BI__sync_add_and_fetch: BuiltinIndex = 5; break; 433 case Builtin::BI__sync_sub_and_fetch: BuiltinIndex = 6; break; 434 case Builtin::BI__sync_and_and_fetch: BuiltinIndex = 7; break; 435 case Builtin::BI__sync_or_and_fetch: BuiltinIndex = 8; break; 436 case Builtin::BI__sync_xor_and_fetch: BuiltinIndex = 9; break; 437 438 case Builtin::BI__sync_val_compare_and_swap: 439 BuiltinIndex = 10; 440 NumFixed = 2; 441 break; 442 case Builtin::BI__sync_bool_compare_and_swap: 443 BuiltinIndex = 11; 444 NumFixed = 2; 445 ResultType = Context.BoolTy; 446 break; 447 case Builtin::BI__sync_lock_test_and_set: BuiltinIndex = 12; break; 448 case Builtin::BI__sync_lock_release: 449 BuiltinIndex = 13; 450 NumFixed = 0; 451 ResultType = Context.VoidTy; 452 break; 453 } 454 455 // Now that we know how many fixed arguments we expect, first check that we 456 // have at least that many. 457 if (TheCall->getNumArgs() < 1+NumFixed) { 458 Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args_at_least) 459 << 0 << 1+NumFixed << TheCall->getNumArgs() 460 << TheCall->getCallee()->getSourceRange(); 461 return ExprError(); 462 } 463 464 // Get the decl for the concrete builtin from this, we can tell what the 465 // concrete integer type we should convert to is. 466 unsigned NewBuiltinID = BuiltinIndices[BuiltinIndex][SizeIndex]; 467 const char *NewBuiltinName = Context.BuiltinInfo.GetName(NewBuiltinID); 468 IdentifierInfo *NewBuiltinII = PP.getIdentifierInfo(NewBuiltinName); 469 FunctionDecl *NewBuiltinDecl = 470 cast<FunctionDecl>(LazilyCreateBuiltin(NewBuiltinII, NewBuiltinID, 471 TUScope, false, DRE->getLocStart())); 472 473 // The first argument --- the pointer --- has a fixed type; we 474 // deduce the types of the rest of the arguments accordingly. Walk 475 // the remaining arguments, converting them to the deduced value type. 476 for (unsigned i = 0; i != NumFixed; ++i) { 477 Expr *Arg = TheCall->getArg(i+1); 478 479 // If the argument is an implicit cast, then there was a promotion due to 480 // "...", just remove it now. 481 if (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(Arg)) { 482 Arg = ICE->getSubExpr(); 483 ICE->setSubExpr(0); 484 TheCall->setArg(i+1, Arg); 485 } 486 487 // GCC does an implicit conversion to the pointer or integer ValType. This 488 // can fail in some cases (1i -> int**), check for this error case now. 489 CastKind Kind = CK_Invalid; 490 ExprValueKind VK = VK_RValue; 491 CXXCastPath BasePath; 492 if (CheckCastTypes(Arg->getSourceRange(), ValType, Arg, Kind, VK, BasePath)) 493 return ExprError(); 494 495 // Okay, we have something that *can* be converted to the right type. Check 496 // to see if there is a potentially weird extension going on here. This can 497 // happen when you do an atomic operation on something like an char* and 498 // pass in 42. The 42 gets converted to char. This is even more strange 499 // for things like 45.123 -> char, etc. 500 // FIXME: Do this check. 501 ImpCastExprToType(Arg, ValType, Kind, VK, &BasePath); 502 TheCall->setArg(i+1, Arg); 503 } 504 505 // Switch the DeclRefExpr to refer to the new decl. 506 DRE->setDecl(NewBuiltinDecl); 507 DRE->setType(NewBuiltinDecl->getType()); 508 509 // Set the callee in the CallExpr. 510 // FIXME: This leaks the original parens and implicit casts. 511 Expr *PromotedCall = DRE; 512 UsualUnaryConversions(PromotedCall); 513 TheCall->setCallee(PromotedCall); 514 515 // Change the result type of the call to match the original value type. This 516 // is arbitrary, but the codegen for these builtins ins design to handle it 517 // gracefully. 518 TheCall->setType(ResultType); 519 520 return move(TheCallResult); 521} 522 523 524/// CheckObjCString - Checks that the argument to the builtin 525/// CFString constructor is correct 526/// Note: It might also make sense to do the UTF-16 conversion here (would 527/// simplify the backend). 528bool Sema::CheckObjCString(Expr *Arg) { 529 Arg = Arg->IgnoreParenCasts(); 530 StringLiteral *Literal = dyn_cast<StringLiteral>(Arg); 531 532 if (!Literal || Literal->isWide()) { 533 Diag(Arg->getLocStart(), diag::err_cfstring_literal_not_string_constant) 534 << Arg->getSourceRange(); 535 return true; 536 } 537 538 size_t NulPos = Literal->getString().find('\0'); 539 if (NulPos != llvm::StringRef::npos) { 540 Diag(getLocationOfStringLiteralByte(Literal, NulPos), 541 diag::warn_cfstring_literal_contains_nul_character) 542 << Arg->getSourceRange(); 543 } 544 if (Literal->containsNonAsciiOrNull()) { 545 llvm::StringRef String = Literal->getString(); 546 unsigned NumBytes = String.size(); 547 llvm::SmallVector<UTF16, 128> ToBuf(NumBytes); 548 const UTF8 *FromPtr = (UTF8 *)String.data(); 549 UTF16 *ToPtr = &ToBuf[0]; 550 551 ConversionResult Result = ConvertUTF8toUTF16(&FromPtr, FromPtr + NumBytes, 552 &ToPtr, ToPtr + NumBytes, 553 strictConversion); 554 // Check for conversion failure. 555 if (Result != conversionOK) 556 Diag(Arg->getLocStart(), 557 diag::warn_cfstring_truncated) << Arg->getSourceRange(); 558 } 559 return false; 560} 561 562/// SemaBuiltinVAStart - Check the arguments to __builtin_va_start for validity. 563/// Emit an error and return true on failure, return false on success. 564bool Sema::SemaBuiltinVAStart(CallExpr *TheCall) { 565 Expr *Fn = TheCall->getCallee(); 566 if (TheCall->getNumArgs() > 2) { 567 Diag(TheCall->getArg(2)->getLocStart(), 568 diag::err_typecheck_call_too_many_args) 569 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 570 << Fn->getSourceRange() 571 << SourceRange(TheCall->getArg(2)->getLocStart(), 572 (*(TheCall->arg_end()-1))->getLocEnd()); 573 return true; 574 } 575 576 if (TheCall->getNumArgs() < 2) { 577 return Diag(TheCall->getLocEnd(), 578 diag::err_typecheck_call_too_few_args_at_least) 579 << 0 /*function call*/ << 2 << TheCall->getNumArgs(); 580 } 581 582 // Determine whether the current function is variadic or not. 583 BlockScopeInfo *CurBlock = getCurBlock(); 584 bool isVariadic; 585 if (CurBlock) 586 isVariadic = CurBlock->TheDecl->isVariadic(); 587 else if (FunctionDecl *FD = getCurFunctionDecl()) 588 isVariadic = FD->isVariadic(); 589 else 590 isVariadic = getCurMethodDecl()->isVariadic(); 591 592 if (!isVariadic) { 593 Diag(Fn->getLocStart(), diag::err_va_start_used_in_non_variadic_function); 594 return true; 595 } 596 597 // Verify that the second argument to the builtin is the last argument of the 598 // current function or method. 599 bool SecondArgIsLastNamedArgument = false; 600 const Expr *Arg = TheCall->getArg(1)->IgnoreParenCasts(); 601 602 if (const DeclRefExpr *DR = dyn_cast<DeclRefExpr>(Arg)) { 603 if (const ParmVarDecl *PV = dyn_cast<ParmVarDecl>(DR->getDecl())) { 604 // FIXME: This isn't correct for methods (results in bogus warning). 605 // Get the last formal in the current function. 606 const ParmVarDecl *LastArg; 607 if (CurBlock) 608 LastArg = *(CurBlock->TheDecl->param_end()-1); 609 else if (FunctionDecl *FD = getCurFunctionDecl()) 610 LastArg = *(FD->param_end()-1); 611 else 612 LastArg = *(getCurMethodDecl()->param_end()-1); 613 SecondArgIsLastNamedArgument = PV == LastArg; 614 } 615 } 616 617 if (!SecondArgIsLastNamedArgument) 618 Diag(TheCall->getArg(1)->getLocStart(), 619 diag::warn_second_parameter_of_va_start_not_last_named_argument); 620 return false; 621} 622 623/// SemaBuiltinUnorderedCompare - Handle functions like __builtin_isgreater and 624/// friends. This is declared to take (...), so we have to check everything. 625bool Sema::SemaBuiltinUnorderedCompare(CallExpr *TheCall) { 626 if (TheCall->getNumArgs() < 2) 627 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 628 << 0 << 2 << TheCall->getNumArgs()/*function call*/; 629 if (TheCall->getNumArgs() > 2) 630 return Diag(TheCall->getArg(2)->getLocStart(), 631 diag::err_typecheck_call_too_many_args) 632 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 633 << SourceRange(TheCall->getArg(2)->getLocStart(), 634 (*(TheCall->arg_end()-1))->getLocEnd()); 635 636 Expr *OrigArg0 = TheCall->getArg(0); 637 Expr *OrigArg1 = TheCall->getArg(1); 638 639 // Do standard promotions between the two arguments, returning their common 640 // type. 641 QualType Res = UsualArithmeticConversions(OrigArg0, OrigArg1, false); 642 643 // Make sure any conversions are pushed back into the call; this is 644 // type safe since unordered compare builtins are declared as "_Bool 645 // foo(...)". 646 TheCall->setArg(0, OrigArg0); 647 TheCall->setArg(1, OrigArg1); 648 649 if (OrigArg0->isTypeDependent() || OrigArg1->isTypeDependent()) 650 return false; 651 652 // If the common type isn't a real floating type, then the arguments were 653 // invalid for this operation. 654 if (!Res->isRealFloatingType()) 655 return Diag(OrigArg0->getLocStart(), 656 diag::err_typecheck_call_invalid_ordered_compare) 657 << OrigArg0->getType() << OrigArg1->getType() 658 << SourceRange(OrigArg0->getLocStart(), OrigArg1->getLocEnd()); 659 660 return false; 661} 662 663/// SemaBuiltinSemaBuiltinFPClassification - Handle functions like 664/// __builtin_isnan and friends. This is declared to take (...), so we have 665/// to check everything. We expect the last argument to be a floating point 666/// value. 667bool Sema::SemaBuiltinFPClassification(CallExpr *TheCall, unsigned NumArgs) { 668 if (TheCall->getNumArgs() < NumArgs) 669 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 670 << 0 << NumArgs << TheCall->getNumArgs()/*function call*/; 671 if (TheCall->getNumArgs() > NumArgs) 672 return Diag(TheCall->getArg(NumArgs)->getLocStart(), 673 diag::err_typecheck_call_too_many_args) 674 << 0 /*function call*/ << NumArgs << TheCall->getNumArgs() 675 << SourceRange(TheCall->getArg(NumArgs)->getLocStart(), 676 (*(TheCall->arg_end()-1))->getLocEnd()); 677 678 Expr *OrigArg = TheCall->getArg(NumArgs-1); 679 680 if (OrigArg->isTypeDependent()) 681 return false; 682 683 // This operation requires a non-_Complex floating-point number. 684 if (!OrigArg->getType()->isRealFloatingType()) 685 return Diag(OrigArg->getLocStart(), 686 diag::err_typecheck_call_invalid_unary_fp) 687 << OrigArg->getType() << OrigArg->getSourceRange(); 688 689 // If this is an implicit conversion from float -> double, remove it. 690 if (ImplicitCastExpr *Cast = dyn_cast<ImplicitCastExpr>(OrigArg)) { 691 Expr *CastArg = Cast->getSubExpr(); 692 if (CastArg->getType()->isSpecificBuiltinType(BuiltinType::Float)) { 693 assert(Cast->getType()->isSpecificBuiltinType(BuiltinType::Double) && 694 "promotion from float to double is the only expected cast here"); 695 Cast->setSubExpr(0); 696 TheCall->setArg(NumArgs-1, CastArg); 697 OrigArg = CastArg; 698 } 699 } 700 701 return false; 702} 703 704/// SemaBuiltinShuffleVector - Handle __builtin_shufflevector. 705// This is declared to take (...), so we have to check everything. 706ExprResult Sema::SemaBuiltinShuffleVector(CallExpr *TheCall) { 707 if (TheCall->getNumArgs() < 2) 708 return ExprError(Diag(TheCall->getLocEnd(), 709 diag::err_typecheck_call_too_few_args_at_least) 710 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 711 << TheCall->getSourceRange()); 712 713 // Determine which of the following types of shufflevector we're checking: 714 // 1) unary, vector mask: (lhs, mask) 715 // 2) binary, vector mask: (lhs, rhs, mask) 716 // 3) binary, scalar mask: (lhs, rhs, index, ..., index) 717 QualType resType = TheCall->getArg(0)->getType(); 718 unsigned numElements = 0; 719 720 if (!TheCall->getArg(0)->isTypeDependent() && 721 !TheCall->getArg(1)->isTypeDependent()) { 722 QualType LHSType = TheCall->getArg(0)->getType(); 723 QualType RHSType = TheCall->getArg(1)->getType(); 724 725 if (!LHSType->isVectorType() || !RHSType->isVectorType()) { 726 Diag(TheCall->getLocStart(), diag::err_shufflevector_non_vector) 727 << SourceRange(TheCall->getArg(0)->getLocStart(), 728 TheCall->getArg(1)->getLocEnd()); 729 return ExprError(); 730 } 731 732 numElements = LHSType->getAs<VectorType>()->getNumElements(); 733 unsigned numResElements = TheCall->getNumArgs() - 2; 734 735 // Check to see if we have a call with 2 vector arguments, the unary shuffle 736 // with mask. If so, verify that RHS is an integer vector type with the 737 // same number of elts as lhs. 738 if (TheCall->getNumArgs() == 2) { 739 if (!RHSType->hasIntegerRepresentation() || 740 RHSType->getAs<VectorType>()->getNumElements() != numElements) 741 Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector) 742 << SourceRange(TheCall->getArg(1)->getLocStart(), 743 TheCall->getArg(1)->getLocEnd()); 744 numResElements = numElements; 745 } 746 else if (!Context.hasSameUnqualifiedType(LHSType, RHSType)) { 747 Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector) 748 << SourceRange(TheCall->getArg(0)->getLocStart(), 749 TheCall->getArg(1)->getLocEnd()); 750 return ExprError(); 751 } else if (numElements != numResElements) { 752 QualType eltType = LHSType->getAs<VectorType>()->getElementType(); 753 resType = Context.getVectorType(eltType, numResElements, 754 VectorType::GenericVector); 755 } 756 } 757 758 for (unsigned i = 2; i < TheCall->getNumArgs(); i++) { 759 if (TheCall->getArg(i)->isTypeDependent() || 760 TheCall->getArg(i)->isValueDependent()) 761 continue; 762 763 llvm::APSInt Result(32); 764 if (!TheCall->getArg(i)->isIntegerConstantExpr(Result, Context)) 765 return ExprError(Diag(TheCall->getLocStart(), 766 diag::err_shufflevector_nonconstant_argument) 767 << TheCall->getArg(i)->getSourceRange()); 768 769 if (Result.getActiveBits() > 64 || Result.getZExtValue() >= numElements*2) 770 return ExprError(Diag(TheCall->getLocStart(), 771 diag::err_shufflevector_argument_too_large) 772 << TheCall->getArg(i)->getSourceRange()); 773 } 774 775 llvm::SmallVector<Expr*, 32> exprs; 776 777 for (unsigned i = 0, e = TheCall->getNumArgs(); i != e; i++) { 778 exprs.push_back(TheCall->getArg(i)); 779 TheCall->setArg(i, 0); 780 } 781 782 return Owned(new (Context) ShuffleVectorExpr(Context, exprs.begin(), 783 exprs.size(), resType, 784 TheCall->getCallee()->getLocStart(), 785 TheCall->getRParenLoc())); 786} 787 788/// SemaBuiltinPrefetch - Handle __builtin_prefetch. 789// This is declared to take (const void*, ...) and can take two 790// optional constant int args. 791bool Sema::SemaBuiltinPrefetch(CallExpr *TheCall) { 792 unsigned NumArgs = TheCall->getNumArgs(); 793 794 if (NumArgs > 3) 795 return Diag(TheCall->getLocEnd(), 796 diag::err_typecheck_call_too_many_args_at_most) 797 << 0 /*function call*/ << 3 << NumArgs 798 << TheCall->getSourceRange(); 799 800 // Argument 0 is checked for us and the remaining arguments must be 801 // constant integers. 802 for (unsigned i = 1; i != NumArgs; ++i) { 803 Expr *Arg = TheCall->getArg(i); 804 805 llvm::APSInt Result; 806 if (SemaBuiltinConstantArg(TheCall, i, Result)) 807 return true; 808 809 // FIXME: gcc issues a warning and rewrites these to 0. These 810 // seems especially odd for the third argument since the default 811 // is 3. 812 if (i == 1) { 813 if (Result.getLimitedValue() > 1) 814 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 815 << "0" << "1" << Arg->getSourceRange(); 816 } else { 817 if (Result.getLimitedValue() > 3) 818 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 819 << "0" << "3" << Arg->getSourceRange(); 820 } 821 } 822 823 return false; 824} 825 826/// SemaBuiltinConstantArg - Handle a check if argument ArgNum of CallExpr 827/// TheCall is a constant expression. 828bool Sema::SemaBuiltinConstantArg(CallExpr *TheCall, int ArgNum, 829 llvm::APSInt &Result) { 830 Expr *Arg = TheCall->getArg(ArgNum); 831 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 832 FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl()); 833 834 if (Arg->isTypeDependent() || Arg->isValueDependent()) return false; 835 836 if (!Arg->isIntegerConstantExpr(Result, Context)) 837 return Diag(TheCall->getLocStart(), diag::err_constant_integer_arg_type) 838 << FDecl->getDeclName() << Arg->getSourceRange(); 839 840 return false; 841} 842 843/// SemaBuiltinObjectSize - Handle __builtin_object_size(void *ptr, 844/// int type). This simply type checks that type is one of the defined 845/// constants (0-3). 846// For compatability check 0-3, llvm only handles 0 and 2. 847bool Sema::SemaBuiltinObjectSize(CallExpr *TheCall) { 848 llvm::APSInt Result; 849 850 // Check constant-ness first. 851 if (SemaBuiltinConstantArg(TheCall, 1, Result)) 852 return true; 853 854 Expr *Arg = TheCall->getArg(1); 855 if (Result.getSExtValue() < 0 || Result.getSExtValue() > 3) { 856 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 857 << "0" << "3" << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 858 } 859 860 return false; 861} 862 863/// SemaBuiltinLongjmp - Handle __builtin_longjmp(void *env[5], int val). 864/// This checks that val is a constant 1. 865bool Sema::SemaBuiltinLongjmp(CallExpr *TheCall) { 866 Expr *Arg = TheCall->getArg(1); 867 llvm::APSInt Result; 868 869 // TODO: This is less than ideal. Overload this to take a value. 870 if (SemaBuiltinConstantArg(TheCall, 1, Result)) 871 return true; 872 873 if (Result != 1) 874 return Diag(TheCall->getLocStart(), diag::err_builtin_longjmp_invalid_val) 875 << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 876 877 return false; 878} 879 880// Handle i > 1 ? "x" : "y", recursivelly 881bool Sema::SemaCheckStringLiteral(const Expr *E, const CallExpr *TheCall, 882 bool HasVAListArg, 883 unsigned format_idx, unsigned firstDataArg, 884 bool isPrintf) { 885 tryAgain: 886 if (E->isTypeDependent() || E->isValueDependent()) 887 return false; 888 889 switch (E->getStmtClass()) { 890 case Stmt::ConditionalOperatorClass: { 891 const ConditionalOperator *C = cast<ConditionalOperator>(E); 892 return SemaCheckStringLiteral(C->getTrueExpr(), TheCall, HasVAListArg, 893 format_idx, firstDataArg, isPrintf) 894 && SemaCheckStringLiteral(C->getRHS(), TheCall, HasVAListArg, 895 format_idx, firstDataArg, isPrintf); 896 } 897 898 case Stmt::IntegerLiteralClass: 899 // Technically -Wformat-nonliteral does not warn about this case. 900 // The behavior of printf and friends in this case is implementation 901 // dependent. Ideally if the format string cannot be null then 902 // it should have a 'nonnull' attribute in the function prototype. 903 return true; 904 905 case Stmt::ImplicitCastExprClass: { 906 E = cast<ImplicitCastExpr>(E)->getSubExpr(); 907 goto tryAgain; 908 } 909 910 case Stmt::ParenExprClass: { 911 E = cast<ParenExpr>(E)->getSubExpr(); 912 goto tryAgain; 913 } 914 915 case Stmt::DeclRefExprClass: { 916 const DeclRefExpr *DR = cast<DeclRefExpr>(E); 917 918 // As an exception, do not flag errors for variables binding to 919 // const string literals. 920 if (const VarDecl *VD = dyn_cast<VarDecl>(DR->getDecl())) { 921 bool isConstant = false; 922 QualType T = DR->getType(); 923 924 if (const ArrayType *AT = Context.getAsArrayType(T)) { 925 isConstant = AT->getElementType().isConstant(Context); 926 } else if (const PointerType *PT = T->getAs<PointerType>()) { 927 isConstant = T.isConstant(Context) && 928 PT->getPointeeType().isConstant(Context); 929 } 930 931 if (isConstant) { 932 if (const Expr *Init = VD->getAnyInitializer()) 933 return SemaCheckStringLiteral(Init, TheCall, 934 HasVAListArg, format_idx, firstDataArg, 935 isPrintf); 936 } 937 938 // For vprintf* functions (i.e., HasVAListArg==true), we add a 939 // special check to see if the format string is a function parameter 940 // of the function calling the printf function. If the function 941 // has an attribute indicating it is a printf-like function, then we 942 // should suppress warnings concerning non-literals being used in a call 943 // to a vprintf function. For example: 944 // 945 // void 946 // logmessage(char const *fmt __attribute__ (format (printf, 1, 2)), ...){ 947 // va_list ap; 948 // va_start(ap, fmt); 949 // vprintf(fmt, ap); // Do NOT emit a warning about "fmt". 950 // ... 951 // 952 // 953 // FIXME: We don't have full attribute support yet, so just check to see 954 // if the argument is a DeclRefExpr that references a parameter. We'll 955 // add proper support for checking the attribute later. 956 if (HasVAListArg) 957 if (isa<ParmVarDecl>(VD)) 958 return true; 959 } 960 961 return false; 962 } 963 964 case Stmt::CallExprClass: { 965 const CallExpr *CE = cast<CallExpr>(E); 966 if (const ImplicitCastExpr *ICE 967 = dyn_cast<ImplicitCastExpr>(CE->getCallee())) { 968 if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(ICE->getSubExpr())) { 969 if (const FunctionDecl *FD = dyn_cast<FunctionDecl>(DRE->getDecl())) { 970 if (const FormatArgAttr *FA = FD->getAttr<FormatArgAttr>()) { 971 unsigned ArgIndex = FA->getFormatIdx(); 972 const Expr *Arg = CE->getArg(ArgIndex - 1); 973 974 return SemaCheckStringLiteral(Arg, TheCall, HasVAListArg, 975 format_idx, firstDataArg, isPrintf); 976 } 977 } 978 } 979 } 980 981 return false; 982 } 983 case Stmt::ObjCStringLiteralClass: 984 case Stmt::StringLiteralClass: { 985 const StringLiteral *StrE = NULL; 986 987 if (const ObjCStringLiteral *ObjCFExpr = dyn_cast<ObjCStringLiteral>(E)) 988 StrE = ObjCFExpr->getString(); 989 else 990 StrE = cast<StringLiteral>(E); 991 992 if (StrE) { 993 CheckFormatString(StrE, E, TheCall, HasVAListArg, format_idx, 994 firstDataArg, isPrintf); 995 return true; 996 } 997 998 return false; 999 } 1000 1001 default: 1002 return false; 1003 } 1004} 1005 1006void 1007Sema::CheckNonNullArguments(const NonNullAttr *NonNull, 1008 const CallExpr *TheCall) { 1009 for (NonNullAttr::args_iterator i = NonNull->args_begin(), 1010 e = NonNull->args_end(); 1011 i != e; ++i) { 1012 const Expr *ArgExpr = TheCall->getArg(*i); 1013 if (ArgExpr->isNullPointerConstant(Context, 1014 Expr::NPC_ValueDependentIsNotNull)) 1015 Diag(TheCall->getCallee()->getLocStart(), diag::warn_null_arg) 1016 << ArgExpr->getSourceRange(); 1017 } 1018} 1019 1020/// CheckPrintfScanfArguments - Check calls to printf and scanf (and similar 1021/// functions) for correct use of format strings. 1022void 1023Sema::CheckPrintfScanfArguments(const CallExpr *TheCall, bool HasVAListArg, 1024 unsigned format_idx, unsigned firstDataArg, 1025 bool isPrintf) { 1026 1027 const Expr *Fn = TheCall->getCallee(); 1028 1029 // The way the format attribute works in GCC, the implicit this argument 1030 // of member functions is counted. However, it doesn't appear in our own 1031 // lists, so decrement format_idx in that case. 1032 if (isa<CXXMemberCallExpr>(TheCall)) { 1033 const CXXMethodDecl *method_decl = 1034 dyn_cast<CXXMethodDecl>(TheCall->getCalleeDecl()); 1035 if (method_decl && method_decl->isInstance()) { 1036 // Catch a format attribute mistakenly referring to the object argument. 1037 if (format_idx == 0) 1038 return; 1039 --format_idx; 1040 if(firstDataArg != 0) 1041 --firstDataArg; 1042 } 1043 } 1044 1045 // CHECK: printf/scanf-like function is called with no format string. 1046 if (format_idx >= TheCall->getNumArgs()) { 1047 Diag(TheCall->getRParenLoc(), diag::warn_missing_format_string) 1048 << Fn->getSourceRange(); 1049 return; 1050 } 1051 1052 const Expr *OrigFormatExpr = TheCall->getArg(format_idx)->IgnoreParenCasts(); 1053 1054 // CHECK: format string is not a string literal. 1055 // 1056 // Dynamically generated format strings are difficult to 1057 // automatically vet at compile time. Requiring that format strings 1058 // are string literals: (1) permits the checking of format strings by 1059 // the compiler and thereby (2) can practically remove the source of 1060 // many format string exploits. 1061 1062 // Format string can be either ObjC string (e.g. @"%d") or 1063 // C string (e.g. "%d") 1064 // ObjC string uses the same format specifiers as C string, so we can use 1065 // the same format string checking logic for both ObjC and C strings. 1066 if (SemaCheckStringLiteral(OrigFormatExpr, TheCall, HasVAListArg, format_idx, 1067 firstDataArg, isPrintf)) 1068 return; // Literal format string found, check done! 1069 1070 // If there are no arguments specified, warn with -Wformat-security, otherwise 1071 // warn only with -Wformat-nonliteral. 1072 if (TheCall->getNumArgs() == format_idx+1) 1073 Diag(TheCall->getArg(format_idx)->getLocStart(), 1074 diag::warn_format_nonliteral_noargs) 1075 << OrigFormatExpr->getSourceRange(); 1076 else 1077 Diag(TheCall->getArg(format_idx)->getLocStart(), 1078 diag::warn_format_nonliteral) 1079 << OrigFormatExpr->getSourceRange(); 1080} 1081 1082namespace { 1083class CheckFormatHandler : public analyze_format_string::FormatStringHandler { 1084protected: 1085 Sema &S; 1086 const StringLiteral *FExpr; 1087 const Expr *OrigFormatExpr; 1088 const unsigned FirstDataArg; 1089 const unsigned NumDataArgs; 1090 const bool IsObjCLiteral; 1091 const char *Beg; // Start of format string. 1092 const bool HasVAListArg; 1093 const CallExpr *TheCall; 1094 unsigned FormatIdx; 1095 llvm::BitVector CoveredArgs; 1096 bool usesPositionalArgs; 1097 bool atFirstArg; 1098public: 1099 CheckFormatHandler(Sema &s, const StringLiteral *fexpr, 1100 const Expr *origFormatExpr, unsigned firstDataArg, 1101 unsigned numDataArgs, bool isObjCLiteral, 1102 const char *beg, bool hasVAListArg, 1103 const CallExpr *theCall, unsigned formatIdx) 1104 : S(s), FExpr(fexpr), OrigFormatExpr(origFormatExpr), 1105 FirstDataArg(firstDataArg), 1106 NumDataArgs(numDataArgs), 1107 IsObjCLiteral(isObjCLiteral), Beg(beg), 1108 HasVAListArg(hasVAListArg), 1109 TheCall(theCall), FormatIdx(formatIdx), 1110 usesPositionalArgs(false), atFirstArg(true) { 1111 CoveredArgs.resize(numDataArgs); 1112 CoveredArgs.reset(); 1113 } 1114 1115 void DoneProcessing(); 1116 1117 void HandleIncompleteSpecifier(const char *startSpecifier, 1118 unsigned specifierLen); 1119 1120 virtual void HandleInvalidPosition(const char *startSpecifier, 1121 unsigned specifierLen, 1122 analyze_format_string::PositionContext p); 1123 1124 virtual void HandleZeroPosition(const char *startPos, unsigned posLen); 1125 1126 void HandleNullChar(const char *nullCharacter); 1127 1128protected: 1129 bool HandleInvalidConversionSpecifier(unsigned argIndex, SourceLocation Loc, 1130 const char *startSpec, 1131 unsigned specifierLen, 1132 const char *csStart, unsigned csLen); 1133 1134 SourceRange getFormatStringRange(); 1135 CharSourceRange getSpecifierRange(const char *startSpecifier, 1136 unsigned specifierLen); 1137 SourceLocation getLocationOfByte(const char *x); 1138 1139 const Expr *getDataArg(unsigned i) const; 1140 1141 bool CheckNumArgs(const analyze_format_string::FormatSpecifier &FS, 1142 const analyze_format_string::ConversionSpecifier &CS, 1143 const char *startSpecifier, unsigned specifierLen, 1144 unsigned argIndex); 1145}; 1146} 1147 1148SourceRange CheckFormatHandler::getFormatStringRange() { 1149 return OrigFormatExpr->getSourceRange(); 1150} 1151 1152CharSourceRange CheckFormatHandler:: 1153getSpecifierRange(const char *startSpecifier, unsigned specifierLen) { 1154 SourceLocation Start = getLocationOfByte(startSpecifier); 1155 SourceLocation End = getLocationOfByte(startSpecifier + specifierLen - 1); 1156 1157 // Advance the end SourceLocation by one due to half-open ranges. 1158 End = End.getFileLocWithOffset(1); 1159 1160 return CharSourceRange::getCharRange(Start, End); 1161} 1162 1163SourceLocation CheckFormatHandler::getLocationOfByte(const char *x) { 1164 return S.getLocationOfStringLiteralByte(FExpr, x - Beg); 1165} 1166 1167void CheckFormatHandler::HandleIncompleteSpecifier(const char *startSpecifier, 1168 unsigned specifierLen){ 1169 SourceLocation Loc = getLocationOfByte(startSpecifier); 1170 S.Diag(Loc, diag::warn_printf_incomplete_specifier) 1171 << getSpecifierRange(startSpecifier, specifierLen); 1172} 1173 1174void 1175CheckFormatHandler::HandleInvalidPosition(const char *startPos, unsigned posLen, 1176 analyze_format_string::PositionContext p) { 1177 SourceLocation Loc = getLocationOfByte(startPos); 1178 S.Diag(Loc, diag::warn_format_invalid_positional_specifier) 1179 << (unsigned) p << getSpecifierRange(startPos, posLen); 1180} 1181 1182void CheckFormatHandler::HandleZeroPosition(const char *startPos, 1183 unsigned posLen) { 1184 SourceLocation Loc = getLocationOfByte(startPos); 1185 S.Diag(Loc, diag::warn_format_zero_positional_specifier) 1186 << getSpecifierRange(startPos, posLen); 1187} 1188 1189void CheckFormatHandler::HandleNullChar(const char *nullCharacter) { 1190 // The presence of a null character is likely an error. 1191 S.Diag(getLocationOfByte(nullCharacter), 1192 diag::warn_printf_format_string_contains_null_char) 1193 << getFormatStringRange(); 1194} 1195 1196const Expr *CheckFormatHandler::getDataArg(unsigned i) const { 1197 return TheCall->getArg(FirstDataArg + i); 1198} 1199 1200void CheckFormatHandler::DoneProcessing() { 1201 // Does the number of data arguments exceed the number of 1202 // format conversions in the format string? 1203 if (!HasVAListArg) { 1204 // Find any arguments that weren't covered. 1205 CoveredArgs.flip(); 1206 signed notCoveredArg = CoveredArgs.find_first(); 1207 if (notCoveredArg >= 0) { 1208 assert((unsigned)notCoveredArg < NumDataArgs); 1209 S.Diag(getDataArg((unsigned) notCoveredArg)->getLocStart(), 1210 diag::warn_printf_data_arg_not_used) 1211 << getFormatStringRange(); 1212 } 1213 } 1214} 1215 1216bool 1217CheckFormatHandler::HandleInvalidConversionSpecifier(unsigned argIndex, 1218 SourceLocation Loc, 1219 const char *startSpec, 1220 unsigned specifierLen, 1221 const char *csStart, 1222 unsigned csLen) { 1223 1224 bool keepGoing = true; 1225 if (argIndex < NumDataArgs) { 1226 // Consider the argument coverered, even though the specifier doesn't 1227 // make sense. 1228 CoveredArgs.set(argIndex); 1229 } 1230 else { 1231 // If argIndex exceeds the number of data arguments we 1232 // don't issue a warning because that is just a cascade of warnings (and 1233 // they may have intended '%%' anyway). We don't want to continue processing 1234 // the format string after this point, however, as we will like just get 1235 // gibberish when trying to match arguments. 1236 keepGoing = false; 1237 } 1238 1239 S.Diag(Loc, diag::warn_format_invalid_conversion) 1240 << llvm::StringRef(csStart, csLen) 1241 << getSpecifierRange(startSpec, specifierLen); 1242 1243 return keepGoing; 1244} 1245 1246bool 1247CheckFormatHandler::CheckNumArgs( 1248 const analyze_format_string::FormatSpecifier &FS, 1249 const analyze_format_string::ConversionSpecifier &CS, 1250 const char *startSpecifier, unsigned specifierLen, unsigned argIndex) { 1251 1252 if (argIndex >= NumDataArgs) { 1253 if (FS.usesPositionalArg()) { 1254 S.Diag(getLocationOfByte(CS.getStart()), 1255 diag::warn_printf_positional_arg_exceeds_data_args) 1256 << (argIndex+1) << NumDataArgs 1257 << getSpecifierRange(startSpecifier, specifierLen); 1258 } 1259 else { 1260 S.Diag(getLocationOfByte(CS.getStart()), 1261 diag::warn_printf_insufficient_data_args) 1262 << getSpecifierRange(startSpecifier, specifierLen); 1263 } 1264 1265 return false; 1266 } 1267 return true; 1268} 1269 1270//===--- CHECK: Printf format string checking ------------------------------===// 1271 1272namespace { 1273class CheckPrintfHandler : public CheckFormatHandler { 1274public: 1275 CheckPrintfHandler(Sema &s, const StringLiteral *fexpr, 1276 const Expr *origFormatExpr, unsigned firstDataArg, 1277 unsigned numDataArgs, bool isObjCLiteral, 1278 const char *beg, bool hasVAListArg, 1279 const CallExpr *theCall, unsigned formatIdx) 1280 : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg, 1281 numDataArgs, isObjCLiteral, beg, hasVAListArg, 1282 theCall, formatIdx) {} 1283 1284 1285 bool HandleInvalidPrintfConversionSpecifier( 1286 const analyze_printf::PrintfSpecifier &FS, 1287 const char *startSpecifier, 1288 unsigned specifierLen); 1289 1290 bool HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier &FS, 1291 const char *startSpecifier, 1292 unsigned specifierLen); 1293 1294 bool HandleAmount(const analyze_format_string::OptionalAmount &Amt, unsigned k, 1295 const char *startSpecifier, unsigned specifierLen); 1296 void HandleInvalidAmount(const analyze_printf::PrintfSpecifier &FS, 1297 const analyze_printf::OptionalAmount &Amt, 1298 unsigned type, 1299 const char *startSpecifier, unsigned specifierLen); 1300 void HandleFlag(const analyze_printf::PrintfSpecifier &FS, 1301 const analyze_printf::OptionalFlag &flag, 1302 const char *startSpecifier, unsigned specifierLen); 1303 void HandleIgnoredFlag(const analyze_printf::PrintfSpecifier &FS, 1304 const analyze_printf::OptionalFlag &ignoredFlag, 1305 const analyze_printf::OptionalFlag &flag, 1306 const char *startSpecifier, unsigned specifierLen); 1307}; 1308} 1309 1310bool CheckPrintfHandler::HandleInvalidPrintfConversionSpecifier( 1311 const analyze_printf::PrintfSpecifier &FS, 1312 const char *startSpecifier, 1313 unsigned specifierLen) { 1314 const analyze_printf::PrintfConversionSpecifier &CS = 1315 FS.getConversionSpecifier(); 1316 1317 return HandleInvalidConversionSpecifier(FS.getArgIndex(), 1318 getLocationOfByte(CS.getStart()), 1319 startSpecifier, specifierLen, 1320 CS.getStart(), CS.getLength()); 1321} 1322 1323bool CheckPrintfHandler::HandleAmount( 1324 const analyze_format_string::OptionalAmount &Amt, 1325 unsigned k, const char *startSpecifier, 1326 unsigned specifierLen) { 1327 1328 if (Amt.hasDataArgument()) { 1329 if (!HasVAListArg) { 1330 unsigned argIndex = Amt.getArgIndex(); 1331 if (argIndex >= NumDataArgs) { 1332 S.Diag(getLocationOfByte(Amt.getStart()), 1333 diag::warn_printf_asterisk_missing_arg) 1334 << k << getSpecifierRange(startSpecifier, specifierLen); 1335 // Don't do any more checking. We will just emit 1336 // spurious errors. 1337 return false; 1338 } 1339 1340 // Type check the data argument. It should be an 'int'. 1341 // Although not in conformance with C99, we also allow the argument to be 1342 // an 'unsigned int' as that is a reasonably safe case. GCC also 1343 // doesn't emit a warning for that case. 1344 CoveredArgs.set(argIndex); 1345 const Expr *Arg = getDataArg(argIndex); 1346 QualType T = Arg->getType(); 1347 1348 const analyze_printf::ArgTypeResult &ATR = Amt.getArgType(S.Context); 1349 assert(ATR.isValid()); 1350 1351 if (!ATR.matchesType(S.Context, T)) { 1352 S.Diag(getLocationOfByte(Amt.getStart()), 1353 diag::warn_printf_asterisk_wrong_type) 1354 << k 1355 << ATR.getRepresentativeType(S.Context) << T 1356 << getSpecifierRange(startSpecifier, specifierLen) 1357 << Arg->getSourceRange(); 1358 // Don't do any more checking. We will just emit 1359 // spurious errors. 1360 return false; 1361 } 1362 } 1363 } 1364 return true; 1365} 1366 1367void CheckPrintfHandler::HandleInvalidAmount( 1368 const analyze_printf::PrintfSpecifier &FS, 1369 const analyze_printf::OptionalAmount &Amt, 1370 unsigned type, 1371 const char *startSpecifier, 1372 unsigned specifierLen) { 1373 const analyze_printf::PrintfConversionSpecifier &CS = 1374 FS.getConversionSpecifier(); 1375 switch (Amt.getHowSpecified()) { 1376 case analyze_printf::OptionalAmount::Constant: 1377 S.Diag(getLocationOfByte(Amt.getStart()), 1378 diag::warn_printf_nonsensical_optional_amount) 1379 << type 1380 << CS.toString() 1381 << getSpecifierRange(startSpecifier, specifierLen) 1382 << FixItHint::CreateRemoval(getSpecifierRange(Amt.getStart(), 1383 Amt.getConstantLength())); 1384 break; 1385 1386 default: 1387 S.Diag(getLocationOfByte(Amt.getStart()), 1388 diag::warn_printf_nonsensical_optional_amount) 1389 << type 1390 << CS.toString() 1391 << getSpecifierRange(startSpecifier, specifierLen); 1392 break; 1393 } 1394} 1395 1396void CheckPrintfHandler::HandleFlag(const analyze_printf::PrintfSpecifier &FS, 1397 const analyze_printf::OptionalFlag &flag, 1398 const char *startSpecifier, 1399 unsigned specifierLen) { 1400 // Warn about pointless flag with a fixit removal. 1401 const analyze_printf::PrintfConversionSpecifier &CS = 1402 FS.getConversionSpecifier(); 1403 S.Diag(getLocationOfByte(flag.getPosition()), 1404 diag::warn_printf_nonsensical_flag) 1405 << flag.toString() << CS.toString() 1406 << getSpecifierRange(startSpecifier, specifierLen) 1407 << FixItHint::CreateRemoval(getSpecifierRange(flag.getPosition(), 1)); 1408} 1409 1410void CheckPrintfHandler::HandleIgnoredFlag( 1411 const analyze_printf::PrintfSpecifier &FS, 1412 const analyze_printf::OptionalFlag &ignoredFlag, 1413 const analyze_printf::OptionalFlag &flag, 1414 const char *startSpecifier, 1415 unsigned specifierLen) { 1416 // Warn about ignored flag with a fixit removal. 1417 S.Diag(getLocationOfByte(ignoredFlag.getPosition()), 1418 diag::warn_printf_ignored_flag) 1419 << ignoredFlag.toString() << flag.toString() 1420 << getSpecifierRange(startSpecifier, specifierLen) 1421 << FixItHint::CreateRemoval(getSpecifierRange( 1422 ignoredFlag.getPosition(), 1)); 1423} 1424 1425bool 1426CheckPrintfHandler::HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier 1427 &FS, 1428 const char *startSpecifier, 1429 unsigned specifierLen) { 1430 1431 using namespace analyze_format_string; 1432 using namespace analyze_printf; 1433 const PrintfConversionSpecifier &CS = FS.getConversionSpecifier(); 1434 1435 if (FS.consumesDataArgument()) { 1436 if (atFirstArg) { 1437 atFirstArg = false; 1438 usesPositionalArgs = FS.usesPositionalArg(); 1439 } 1440 else if (usesPositionalArgs != FS.usesPositionalArg()) { 1441 // Cannot mix-and-match positional and non-positional arguments. 1442 S.Diag(getLocationOfByte(CS.getStart()), 1443 diag::warn_format_mix_positional_nonpositional_args) 1444 << getSpecifierRange(startSpecifier, specifierLen); 1445 return false; 1446 } 1447 } 1448 1449 // First check if the field width, precision, and conversion specifier 1450 // have matching data arguments. 1451 if (!HandleAmount(FS.getFieldWidth(), /* field width */ 0, 1452 startSpecifier, specifierLen)) { 1453 return false; 1454 } 1455 1456 if (!HandleAmount(FS.getPrecision(), /* precision */ 1, 1457 startSpecifier, specifierLen)) { 1458 return false; 1459 } 1460 1461 if (!CS.consumesDataArgument()) { 1462 // FIXME: Technically specifying a precision or field width here 1463 // makes no sense. Worth issuing a warning at some point. 1464 return true; 1465 } 1466 1467 // Consume the argument. 1468 unsigned argIndex = FS.getArgIndex(); 1469 if (argIndex < NumDataArgs) { 1470 // The check to see if the argIndex is valid will come later. 1471 // We set the bit here because we may exit early from this 1472 // function if we encounter some other error. 1473 CoveredArgs.set(argIndex); 1474 } 1475 1476 // Check for using an Objective-C specific conversion specifier 1477 // in a non-ObjC literal. 1478 if (!IsObjCLiteral && CS.isObjCArg()) { 1479 return HandleInvalidPrintfConversionSpecifier(FS, startSpecifier, 1480 specifierLen); 1481 } 1482 1483 // Check for invalid use of field width 1484 if (!FS.hasValidFieldWidth()) { 1485 HandleInvalidAmount(FS, FS.getFieldWidth(), /* field width */ 0, 1486 startSpecifier, specifierLen); 1487 } 1488 1489 // Check for invalid use of precision 1490 if (!FS.hasValidPrecision()) { 1491 HandleInvalidAmount(FS, FS.getPrecision(), /* precision */ 1, 1492 startSpecifier, specifierLen); 1493 } 1494 1495 // Check each flag does not conflict with any other component. 1496 if (!FS.hasValidLeadingZeros()) 1497 HandleFlag(FS, FS.hasLeadingZeros(), startSpecifier, specifierLen); 1498 if (!FS.hasValidPlusPrefix()) 1499 HandleFlag(FS, FS.hasPlusPrefix(), startSpecifier, specifierLen); 1500 if (!FS.hasValidSpacePrefix()) 1501 HandleFlag(FS, FS.hasSpacePrefix(), startSpecifier, specifierLen); 1502 if (!FS.hasValidAlternativeForm()) 1503 HandleFlag(FS, FS.hasAlternativeForm(), startSpecifier, specifierLen); 1504 if (!FS.hasValidLeftJustified()) 1505 HandleFlag(FS, FS.isLeftJustified(), startSpecifier, specifierLen); 1506 1507 // Check that flags are not ignored by another flag 1508 if (FS.hasSpacePrefix() && FS.hasPlusPrefix()) // ' ' ignored by '+' 1509 HandleIgnoredFlag(FS, FS.hasSpacePrefix(), FS.hasPlusPrefix(), 1510 startSpecifier, specifierLen); 1511 if (FS.hasLeadingZeros() && FS.isLeftJustified()) // '0' ignored by '-' 1512 HandleIgnoredFlag(FS, FS.hasLeadingZeros(), FS.isLeftJustified(), 1513 startSpecifier, specifierLen); 1514 1515 // Check the length modifier is valid with the given conversion specifier. 1516 const LengthModifier &LM = FS.getLengthModifier(); 1517 if (!FS.hasValidLengthModifier()) 1518 S.Diag(getLocationOfByte(LM.getStart()), 1519 diag::warn_format_nonsensical_length) 1520 << LM.toString() << CS.toString() 1521 << getSpecifierRange(startSpecifier, specifierLen) 1522 << FixItHint::CreateRemoval(getSpecifierRange(LM.getStart(), 1523 LM.getLength())); 1524 1525 // Are we using '%n'? 1526 if (CS.getKind() == ConversionSpecifier::nArg) { 1527 // Issue a warning about this being a possible security issue. 1528 S.Diag(getLocationOfByte(CS.getStart()), diag::warn_printf_write_back) 1529 << getSpecifierRange(startSpecifier, specifierLen); 1530 // Continue checking the other format specifiers. 1531 return true; 1532 } 1533 1534 // The remaining checks depend on the data arguments. 1535 if (HasVAListArg) 1536 return true; 1537 1538 if (!CheckNumArgs(FS, CS, startSpecifier, specifierLen, argIndex)) 1539 return false; 1540 1541 // Now type check the data expression that matches the 1542 // format specifier. 1543 const Expr *Ex = getDataArg(argIndex); 1544 const analyze_printf::ArgTypeResult &ATR = FS.getArgType(S.Context); 1545 if (ATR.isValid() && !ATR.matchesType(S.Context, Ex->getType())) { 1546 // Check if we didn't match because of an implicit cast from a 'char' 1547 // or 'short' to an 'int'. This is done because printf is a varargs 1548 // function. 1549 if (const ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(Ex)) 1550 if (ICE->getType() == S.Context.IntTy) { 1551 // All further checking is done on the subexpression. 1552 Ex = ICE->getSubExpr(); 1553 if (ATR.matchesType(S.Context, Ex->getType())) 1554 return true; 1555 } 1556 1557 // We may be able to offer a FixItHint if it is a supported type. 1558 PrintfSpecifier fixedFS = FS; 1559 bool success = fixedFS.fixType(Ex->getType()); 1560 1561 if (success) { 1562 // Get the fix string from the fixed format specifier 1563 llvm::SmallString<128> buf; 1564 llvm::raw_svector_ostream os(buf); 1565 fixedFS.toString(os); 1566 1567 // FIXME: getRepresentativeType() perhaps should return a string 1568 // instead of a QualType to better handle when the representative 1569 // type is 'wint_t' (which is defined in the system headers). 1570 S.Diag(getLocationOfByte(CS.getStart()), 1571 diag::warn_printf_conversion_argument_type_mismatch) 1572 << ATR.getRepresentativeType(S.Context) << Ex->getType() 1573 << getSpecifierRange(startSpecifier, specifierLen) 1574 << Ex->getSourceRange() 1575 << FixItHint::CreateReplacement( 1576 getSpecifierRange(startSpecifier, specifierLen), 1577 os.str()); 1578 } 1579 else { 1580 S.Diag(getLocationOfByte(CS.getStart()), 1581 diag::warn_printf_conversion_argument_type_mismatch) 1582 << ATR.getRepresentativeType(S.Context) << Ex->getType() 1583 << getSpecifierRange(startSpecifier, specifierLen) 1584 << Ex->getSourceRange(); 1585 } 1586 } 1587 1588 return true; 1589} 1590 1591//===--- CHECK: Scanf format string checking ------------------------------===// 1592 1593namespace { 1594class CheckScanfHandler : public CheckFormatHandler { 1595public: 1596 CheckScanfHandler(Sema &s, const StringLiteral *fexpr, 1597 const Expr *origFormatExpr, unsigned firstDataArg, 1598 unsigned numDataArgs, bool isObjCLiteral, 1599 const char *beg, bool hasVAListArg, 1600 const CallExpr *theCall, unsigned formatIdx) 1601 : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg, 1602 numDataArgs, isObjCLiteral, beg, hasVAListArg, 1603 theCall, formatIdx) {} 1604 1605 bool HandleScanfSpecifier(const analyze_scanf::ScanfSpecifier &FS, 1606 const char *startSpecifier, 1607 unsigned specifierLen); 1608 1609 bool HandleInvalidScanfConversionSpecifier( 1610 const analyze_scanf::ScanfSpecifier &FS, 1611 const char *startSpecifier, 1612 unsigned specifierLen); 1613 1614 void HandleIncompleteScanList(const char *start, const char *end); 1615}; 1616} 1617 1618void CheckScanfHandler::HandleIncompleteScanList(const char *start, 1619 const char *end) { 1620 S.Diag(getLocationOfByte(end), diag::warn_scanf_scanlist_incomplete) 1621 << getSpecifierRange(start, end - start); 1622} 1623 1624bool CheckScanfHandler::HandleInvalidScanfConversionSpecifier( 1625 const analyze_scanf::ScanfSpecifier &FS, 1626 const char *startSpecifier, 1627 unsigned specifierLen) { 1628 1629 const analyze_scanf::ScanfConversionSpecifier &CS = 1630 FS.getConversionSpecifier(); 1631 1632 return HandleInvalidConversionSpecifier(FS.getArgIndex(), 1633 getLocationOfByte(CS.getStart()), 1634 startSpecifier, specifierLen, 1635 CS.getStart(), CS.getLength()); 1636} 1637 1638bool CheckScanfHandler::HandleScanfSpecifier( 1639 const analyze_scanf::ScanfSpecifier &FS, 1640 const char *startSpecifier, 1641 unsigned specifierLen) { 1642 1643 using namespace analyze_scanf; 1644 using namespace analyze_format_string; 1645 1646 const ScanfConversionSpecifier &CS = FS.getConversionSpecifier(); 1647 1648 // Handle case where '%' and '*' don't consume an argument. These shouldn't 1649 // be used to decide if we are using positional arguments consistently. 1650 if (FS.consumesDataArgument()) { 1651 if (atFirstArg) { 1652 atFirstArg = false; 1653 usesPositionalArgs = FS.usesPositionalArg(); 1654 } 1655 else if (usesPositionalArgs != FS.usesPositionalArg()) { 1656 // Cannot mix-and-match positional and non-positional arguments. 1657 S.Diag(getLocationOfByte(CS.getStart()), 1658 diag::warn_format_mix_positional_nonpositional_args) 1659 << getSpecifierRange(startSpecifier, specifierLen); 1660 return false; 1661 } 1662 } 1663 1664 // Check if the field with is non-zero. 1665 const OptionalAmount &Amt = FS.getFieldWidth(); 1666 if (Amt.getHowSpecified() == OptionalAmount::Constant) { 1667 if (Amt.getConstantAmount() == 0) { 1668 const CharSourceRange &R = getSpecifierRange(Amt.getStart(), 1669 Amt.getConstantLength()); 1670 S.Diag(getLocationOfByte(Amt.getStart()), 1671 diag::warn_scanf_nonzero_width) 1672 << R << FixItHint::CreateRemoval(R); 1673 } 1674 } 1675 1676 if (!FS.consumesDataArgument()) { 1677 // FIXME: Technically specifying a precision or field width here 1678 // makes no sense. Worth issuing a warning at some point. 1679 return true; 1680 } 1681 1682 // Consume the argument. 1683 unsigned argIndex = FS.getArgIndex(); 1684 if (argIndex < NumDataArgs) { 1685 // The check to see if the argIndex is valid will come later. 1686 // We set the bit here because we may exit early from this 1687 // function if we encounter some other error. 1688 CoveredArgs.set(argIndex); 1689 } 1690 1691 // Check the length modifier is valid with the given conversion specifier. 1692 const LengthModifier &LM = FS.getLengthModifier(); 1693 if (!FS.hasValidLengthModifier()) { 1694 S.Diag(getLocationOfByte(LM.getStart()), 1695 diag::warn_format_nonsensical_length) 1696 << LM.toString() << CS.toString() 1697 << getSpecifierRange(startSpecifier, specifierLen) 1698 << FixItHint::CreateRemoval(getSpecifierRange(LM.getStart(), 1699 LM.getLength())); 1700 } 1701 1702 // The remaining checks depend on the data arguments. 1703 if (HasVAListArg) 1704 return true; 1705 1706 if (!CheckNumArgs(FS, CS, startSpecifier, specifierLen, argIndex)) 1707 return false; 1708 1709 // FIXME: Check that the argument type matches the format specifier. 1710 1711 return true; 1712} 1713 1714void Sema::CheckFormatString(const StringLiteral *FExpr, 1715 const Expr *OrigFormatExpr, 1716 const CallExpr *TheCall, bool HasVAListArg, 1717 unsigned format_idx, unsigned firstDataArg, 1718 bool isPrintf) { 1719 1720 // CHECK: is the format string a wide literal? 1721 if (FExpr->isWide()) { 1722 Diag(FExpr->getLocStart(), 1723 diag::warn_format_string_is_wide_literal) 1724 << OrigFormatExpr->getSourceRange(); 1725 return; 1726 } 1727 1728 // Str - The format string. NOTE: this is NOT null-terminated! 1729 llvm::StringRef StrRef = FExpr->getString(); 1730 const char *Str = StrRef.data(); 1731 unsigned StrLen = StrRef.size(); 1732 1733 // CHECK: empty format string? 1734 if (StrLen == 0) { 1735 Diag(FExpr->getLocStart(), diag::warn_empty_format_string) 1736 << OrigFormatExpr->getSourceRange(); 1737 return; 1738 } 1739 1740 if (isPrintf) { 1741 CheckPrintfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg, 1742 TheCall->getNumArgs() - firstDataArg, 1743 isa<ObjCStringLiteral>(OrigFormatExpr), Str, 1744 HasVAListArg, TheCall, format_idx); 1745 1746 if (!analyze_format_string::ParsePrintfString(H, Str, Str + StrLen)) 1747 H.DoneProcessing(); 1748 } 1749 else { 1750 CheckScanfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg, 1751 TheCall->getNumArgs() - firstDataArg, 1752 isa<ObjCStringLiteral>(OrigFormatExpr), Str, 1753 HasVAListArg, TheCall, format_idx); 1754 1755 if (!analyze_format_string::ParseScanfString(H, Str, Str + StrLen)) 1756 H.DoneProcessing(); 1757 } 1758} 1759 1760//===--- CHECK: Return Address of Stack Variable --------------------------===// 1761 1762static Expr *EvalVal(Expr *E, llvm::SmallVectorImpl<DeclRefExpr *> &refVars); 1763static Expr *EvalAddr(Expr* E, llvm::SmallVectorImpl<DeclRefExpr *> &refVars); 1764 1765/// CheckReturnStackAddr - Check if a return statement returns the address 1766/// of a stack variable. 1767void 1768Sema::CheckReturnStackAddr(Expr *RetValExp, QualType lhsType, 1769 SourceLocation ReturnLoc) { 1770 1771 Expr *stackE = 0; 1772 llvm::SmallVector<DeclRefExpr *, 8> refVars; 1773 1774 // Perform checking for returned stack addresses, local blocks, 1775 // label addresses or references to temporaries. 1776 if (lhsType->isPointerType() || lhsType->isBlockPointerType()) { 1777 stackE = EvalAddr(RetValExp, refVars); 1778 } else if (lhsType->isReferenceType()) { 1779 stackE = EvalVal(RetValExp, refVars); 1780 } 1781 1782 if (stackE == 0) 1783 return; // Nothing suspicious was found. 1784 1785 SourceLocation diagLoc; 1786 SourceRange diagRange; 1787 if (refVars.empty()) { 1788 diagLoc = stackE->getLocStart(); 1789 diagRange = stackE->getSourceRange(); 1790 } else { 1791 // We followed through a reference variable. 'stackE' contains the 1792 // problematic expression but we will warn at the return statement pointing 1793 // at the reference variable. We will later display the "trail" of 1794 // reference variables using notes. 1795 diagLoc = refVars[0]->getLocStart(); 1796 diagRange = refVars[0]->getSourceRange(); 1797 } 1798 1799 if (DeclRefExpr *DR = dyn_cast<DeclRefExpr>(stackE)) { //address of local var. 1800 Diag(diagLoc, lhsType->isReferenceType() ? diag::warn_ret_stack_ref 1801 : diag::warn_ret_stack_addr) 1802 << DR->getDecl()->getDeclName() << diagRange; 1803 } else if (isa<BlockExpr>(stackE)) { // local block. 1804 Diag(diagLoc, diag::err_ret_local_block) << diagRange; 1805 } else if (isa<AddrLabelExpr>(stackE)) { // address of label. 1806 Diag(diagLoc, diag::warn_ret_addr_label) << diagRange; 1807 } else { // local temporary. 1808 Diag(diagLoc, lhsType->isReferenceType() ? diag::warn_ret_local_temp_ref 1809 : diag::warn_ret_local_temp_addr) 1810 << diagRange; 1811 } 1812 1813 // Display the "trail" of reference variables that we followed until we 1814 // found the problematic expression using notes. 1815 for (unsigned i = 0, e = refVars.size(); i != e; ++i) { 1816 VarDecl *VD = cast<VarDecl>(refVars[i]->getDecl()); 1817 // If this var binds to another reference var, show the range of the next 1818 // var, otherwise the var binds to the problematic expression, in which case 1819 // show the range of the expression. 1820 SourceRange range = (i < e-1) ? refVars[i+1]->getSourceRange() 1821 : stackE->getSourceRange(); 1822 Diag(VD->getLocation(), diag::note_ref_var_local_bind) 1823 << VD->getDeclName() << range; 1824 } 1825} 1826 1827/// EvalAddr - EvalAddr and EvalVal are mutually recursive functions that 1828/// check if the expression in a return statement evaluates to an address 1829/// to a location on the stack, a local block, an address of a label, or a 1830/// reference to local temporary. The recursion is used to traverse the 1831/// AST of the return expression, with recursion backtracking when we 1832/// encounter a subexpression that (1) clearly does not lead to one of the 1833/// above problematic expressions (2) is something we cannot determine leads to 1834/// a problematic expression based on such local checking. 1835/// 1836/// Both EvalAddr and EvalVal follow through reference variables to evaluate 1837/// the expression that they point to. Such variables are added to the 1838/// 'refVars' vector so that we know what the reference variable "trail" was. 1839/// 1840/// EvalAddr processes expressions that are pointers that are used as 1841/// references (and not L-values). EvalVal handles all other values. 1842/// At the base case of the recursion is a check for the above problematic 1843/// expressions. 1844/// 1845/// This implementation handles: 1846/// 1847/// * pointer-to-pointer casts 1848/// * implicit conversions from array references to pointers 1849/// * taking the address of fields 1850/// * arbitrary interplay between "&" and "*" operators 1851/// * pointer arithmetic from an address of a stack variable 1852/// * taking the address of an array element where the array is on the stack 1853static Expr *EvalAddr(Expr *E, llvm::SmallVectorImpl<DeclRefExpr *> &refVars) { 1854 if (E->isTypeDependent()) 1855 return NULL; 1856 1857 // We should only be called for evaluating pointer expressions. 1858 assert((E->getType()->isAnyPointerType() || 1859 E->getType()->isBlockPointerType() || 1860 E->getType()->isObjCQualifiedIdType()) && 1861 "EvalAddr only works on pointers"); 1862 1863 // Our "symbolic interpreter" is just a dispatch off the currently 1864 // viewed AST node. We then recursively traverse the AST by calling 1865 // EvalAddr and EvalVal appropriately. 1866 switch (E->getStmtClass()) { 1867 case Stmt::ParenExprClass: 1868 // Ignore parentheses. 1869 return EvalAddr(cast<ParenExpr>(E)->getSubExpr(), refVars); 1870 1871 case Stmt::DeclRefExprClass: { 1872 DeclRefExpr *DR = cast<DeclRefExpr>(E); 1873 1874 if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl())) 1875 // If this is a reference variable, follow through to the expression that 1876 // it points to. 1877 if (V->hasLocalStorage() && 1878 V->getType()->isReferenceType() && V->hasInit()) { 1879 // Add the reference variable to the "trail". 1880 refVars.push_back(DR); 1881 return EvalAddr(V->getInit(), refVars); 1882 } 1883 1884 return NULL; 1885 } 1886 1887 case Stmt::UnaryOperatorClass: { 1888 // The only unary operator that make sense to handle here 1889 // is AddrOf. All others don't make sense as pointers. 1890 UnaryOperator *U = cast<UnaryOperator>(E); 1891 1892 if (U->getOpcode() == UO_AddrOf) 1893 return EvalVal(U->getSubExpr(), refVars); 1894 else 1895 return NULL; 1896 } 1897 1898 case Stmt::BinaryOperatorClass: { 1899 // Handle pointer arithmetic. All other binary operators are not valid 1900 // in this context. 1901 BinaryOperator *B = cast<BinaryOperator>(E); 1902 BinaryOperatorKind op = B->getOpcode(); 1903 1904 if (op != BO_Add && op != BO_Sub) 1905 return NULL; 1906 1907 Expr *Base = B->getLHS(); 1908 1909 // Determine which argument is the real pointer base. It could be 1910 // the RHS argument instead of the LHS. 1911 if (!Base->getType()->isPointerType()) Base = B->getRHS(); 1912 1913 assert (Base->getType()->isPointerType()); 1914 return EvalAddr(Base, refVars); 1915 } 1916 1917 // For conditional operators we need to see if either the LHS or RHS are 1918 // valid DeclRefExpr*s. If one of them is valid, we return it. 1919 case Stmt::ConditionalOperatorClass: { 1920 ConditionalOperator *C = cast<ConditionalOperator>(E); 1921 1922 // Handle the GNU extension for missing LHS. 1923 if (Expr *lhsExpr = C->getLHS()) { 1924 // In C++, we can have a throw-expression, which has 'void' type. 1925 if (!lhsExpr->getType()->isVoidType()) 1926 if (Expr* LHS = EvalAddr(lhsExpr, refVars)) 1927 return LHS; 1928 } 1929 1930 // In C++, we can have a throw-expression, which has 'void' type. 1931 if (C->getRHS()->getType()->isVoidType()) 1932 return NULL; 1933 1934 return EvalAddr(C->getRHS(), refVars); 1935 } 1936 1937 case Stmt::BlockExprClass: 1938 if (cast<BlockExpr>(E)->hasBlockDeclRefExprs()) 1939 return E; // local block. 1940 return NULL; 1941 1942 case Stmt::AddrLabelExprClass: 1943 return E; // address of label. 1944 1945 // For casts, we need to handle conversions from arrays to 1946 // pointer values, and pointer-to-pointer conversions. 1947 case Stmt::ImplicitCastExprClass: 1948 case Stmt::CStyleCastExprClass: 1949 case Stmt::CXXFunctionalCastExprClass: { 1950 Expr* SubExpr = cast<CastExpr>(E)->getSubExpr(); 1951 QualType T = SubExpr->getType(); 1952 1953 if (SubExpr->getType()->isPointerType() || 1954 SubExpr->getType()->isBlockPointerType() || 1955 SubExpr->getType()->isObjCQualifiedIdType()) 1956 return EvalAddr(SubExpr, refVars); 1957 else if (T->isArrayType()) 1958 return EvalVal(SubExpr, refVars); 1959 else 1960 return 0; 1961 } 1962 1963 // C++ casts. For dynamic casts, static casts, and const casts, we 1964 // are always converting from a pointer-to-pointer, so we just blow 1965 // through the cast. In the case the dynamic cast doesn't fail (and 1966 // return NULL), we take the conservative route and report cases 1967 // where we return the address of a stack variable. For Reinterpre 1968 // FIXME: The comment about is wrong; we're not always converting 1969 // from pointer to pointer. I'm guessing that this code should also 1970 // handle references to objects. 1971 case Stmt::CXXStaticCastExprClass: 1972 case Stmt::CXXDynamicCastExprClass: 1973 case Stmt::CXXConstCastExprClass: 1974 case Stmt::CXXReinterpretCastExprClass: { 1975 Expr *S = cast<CXXNamedCastExpr>(E)->getSubExpr(); 1976 if (S->getType()->isPointerType() || S->getType()->isBlockPointerType()) 1977 return EvalAddr(S, refVars); 1978 else 1979 return NULL; 1980 } 1981 1982 // Everything else: we simply don't reason about them. 1983 default: 1984 return NULL; 1985 } 1986} 1987 1988 1989/// EvalVal - This function is complements EvalAddr in the mutual recursion. 1990/// See the comments for EvalAddr for more details. 1991static Expr *EvalVal(Expr *E, llvm::SmallVectorImpl<DeclRefExpr *> &refVars) { 1992do { 1993 // We should only be called for evaluating non-pointer expressions, or 1994 // expressions with a pointer type that are not used as references but instead 1995 // are l-values (e.g., DeclRefExpr with a pointer type). 1996 1997 // Our "symbolic interpreter" is just a dispatch off the currently 1998 // viewed AST node. We then recursively traverse the AST by calling 1999 // EvalAddr and EvalVal appropriately. 2000 switch (E->getStmtClass()) { 2001 case Stmt::ImplicitCastExprClass: { 2002 ImplicitCastExpr *IE = cast<ImplicitCastExpr>(E); 2003 if (IE->getValueKind() == VK_LValue) { 2004 E = IE->getSubExpr(); 2005 continue; 2006 } 2007 return NULL; 2008 } 2009 2010 case Stmt::DeclRefExprClass: { 2011 // When we hit a DeclRefExpr we are looking at code that refers to a 2012 // variable's name. If it's not a reference variable we check if it has 2013 // local storage within the function, and if so, return the expression. 2014 DeclRefExpr *DR = cast<DeclRefExpr>(E); 2015 2016 if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl())) 2017 if (V->hasLocalStorage()) { 2018 if (!V->getType()->isReferenceType()) 2019 return DR; 2020 2021 // Reference variable, follow through to the expression that 2022 // it points to. 2023 if (V->hasInit()) { 2024 // Add the reference variable to the "trail". 2025 refVars.push_back(DR); 2026 return EvalVal(V->getInit(), refVars); 2027 } 2028 } 2029 2030 return NULL; 2031 } 2032 2033 case Stmt::ParenExprClass: { 2034 // Ignore parentheses. 2035 E = cast<ParenExpr>(E)->getSubExpr(); 2036 continue; 2037 } 2038 2039 case Stmt::UnaryOperatorClass: { 2040 // The only unary operator that make sense to handle here 2041 // is Deref. All others don't resolve to a "name." This includes 2042 // handling all sorts of rvalues passed to a unary operator. 2043 UnaryOperator *U = cast<UnaryOperator>(E); 2044 2045 if (U->getOpcode() == UO_Deref) 2046 return EvalAddr(U->getSubExpr(), refVars); 2047 2048 return NULL; 2049 } 2050 2051 case Stmt::ArraySubscriptExprClass: { 2052 // Array subscripts are potential references to data on the stack. We 2053 // retrieve the DeclRefExpr* for the array variable if it indeed 2054 // has local storage. 2055 return EvalAddr(cast<ArraySubscriptExpr>(E)->getBase(), refVars); 2056 } 2057 2058 case Stmt::ConditionalOperatorClass: { 2059 // For conditional operators we need to see if either the LHS or RHS are 2060 // non-NULL Expr's. If one is non-NULL, we return it. 2061 ConditionalOperator *C = cast<ConditionalOperator>(E); 2062 2063 // Handle the GNU extension for missing LHS. 2064 if (Expr *lhsExpr = C->getLHS()) 2065 if (Expr *LHS = EvalVal(lhsExpr, refVars)) 2066 return LHS; 2067 2068 return EvalVal(C->getRHS(), refVars); 2069 } 2070 2071 // Accesses to members are potential references to data on the stack. 2072 case Stmt::MemberExprClass: { 2073 MemberExpr *M = cast<MemberExpr>(E); 2074 2075 // Check for indirect access. We only want direct field accesses. 2076 if (M->isArrow()) 2077 return NULL; 2078 2079 // Check whether the member type is itself a reference, in which case 2080 // we're not going to refer to the member, but to what the member refers to. 2081 if (M->getMemberDecl()->getType()->isReferenceType()) 2082 return NULL; 2083 2084 return EvalVal(M->getBase(), refVars); 2085 } 2086 2087 default: 2088 // Check that we don't return or take the address of a reference to a 2089 // temporary. This is only useful in C++. 2090 if (!E->isTypeDependent() && E->isRValue()) 2091 return E; 2092 2093 // Everything else: we simply don't reason about them. 2094 return NULL; 2095 } 2096} while (true); 2097} 2098 2099//===--- CHECK: Floating-Point comparisons (-Wfloat-equal) ---------------===// 2100 2101/// Check for comparisons of floating point operands using != and ==. 2102/// Issue a warning if these are no self-comparisons, as they are not likely 2103/// to do what the programmer intended. 2104void Sema::CheckFloatComparison(SourceLocation loc, Expr* lex, Expr *rex) { 2105 bool EmitWarning = true; 2106 2107 Expr* LeftExprSansParen = lex->IgnoreParens(); 2108 Expr* RightExprSansParen = rex->IgnoreParens(); 2109 2110 // Special case: check for x == x (which is OK). 2111 // Do not emit warnings for such cases. 2112 if (DeclRefExpr* DRL = dyn_cast<DeclRefExpr>(LeftExprSansParen)) 2113 if (DeclRefExpr* DRR = dyn_cast<DeclRefExpr>(RightExprSansParen)) 2114 if (DRL->getDecl() == DRR->getDecl()) 2115 EmitWarning = false; 2116 2117 2118 // Special case: check for comparisons against literals that can be exactly 2119 // represented by APFloat. In such cases, do not emit a warning. This 2120 // is a heuristic: often comparison against such literals are used to 2121 // detect if a value in a variable has not changed. This clearly can 2122 // lead to false negatives. 2123 if (EmitWarning) { 2124 if (FloatingLiteral* FLL = dyn_cast<FloatingLiteral>(LeftExprSansParen)) { 2125 if (FLL->isExact()) 2126 EmitWarning = false; 2127 } else 2128 if (FloatingLiteral* FLR = dyn_cast<FloatingLiteral>(RightExprSansParen)){ 2129 if (FLR->isExact()) 2130 EmitWarning = false; 2131 } 2132 } 2133 2134 // Check for comparisons with builtin types. 2135 if (EmitWarning) 2136 if (CallExpr* CL = dyn_cast<CallExpr>(LeftExprSansParen)) 2137 if (CL->isBuiltinCall(Context)) 2138 EmitWarning = false; 2139 2140 if (EmitWarning) 2141 if (CallExpr* CR = dyn_cast<CallExpr>(RightExprSansParen)) 2142 if (CR->isBuiltinCall(Context)) 2143 EmitWarning = false; 2144 2145 // Emit the diagnostic. 2146 if (EmitWarning) 2147 Diag(loc, diag::warn_floatingpoint_eq) 2148 << lex->getSourceRange() << rex->getSourceRange(); 2149} 2150 2151//===--- CHECK: Integer mixed-sign comparisons (-Wsign-compare) --------===// 2152//===--- CHECK: Lossy implicit conversions (-Wconversion) --------------===// 2153 2154namespace { 2155 2156/// Structure recording the 'active' range of an integer-valued 2157/// expression. 2158struct IntRange { 2159 /// The number of bits active in the int. 2160 unsigned Width; 2161 2162 /// True if the int is known not to have negative values. 2163 bool NonNegative; 2164 2165 IntRange(unsigned Width, bool NonNegative) 2166 : Width(Width), NonNegative(NonNegative) 2167 {} 2168 2169 /// Returns the range of the bool type. 2170 static IntRange forBoolType() { 2171 return IntRange(1, true); 2172 } 2173 2174 /// Returns the range of an opaque value of the given integral type. 2175 static IntRange forValueOfType(ASTContext &C, QualType T) { 2176 return forValueOfCanonicalType(C, 2177 T->getCanonicalTypeInternal().getTypePtr()); 2178 } 2179 2180 /// Returns the range of an opaque value of a canonical integral type. 2181 static IntRange forValueOfCanonicalType(ASTContext &C, const Type *T) { 2182 assert(T->isCanonicalUnqualified()); 2183 2184 if (const VectorType *VT = dyn_cast<VectorType>(T)) 2185 T = VT->getElementType().getTypePtr(); 2186 if (const ComplexType *CT = dyn_cast<ComplexType>(T)) 2187 T = CT->getElementType().getTypePtr(); 2188 2189 // For enum types, use the known bit width of the enumerators. 2190 if (const EnumType *ET = dyn_cast<EnumType>(T)) { 2191 EnumDecl *Enum = ET->getDecl(); 2192 if (!Enum->isDefinition()) 2193 return IntRange(C.getIntWidth(QualType(T, 0)), false); 2194 2195 unsigned NumPositive = Enum->getNumPositiveBits(); 2196 unsigned NumNegative = Enum->getNumNegativeBits(); 2197 2198 return IntRange(std::max(NumPositive, NumNegative), NumNegative == 0); 2199 } 2200 2201 const BuiltinType *BT = cast<BuiltinType>(T); 2202 assert(BT->isInteger()); 2203 2204 return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger()); 2205 } 2206 2207 /// Returns the "target" range of a canonical integral type, i.e. 2208 /// the range of values expressible in the type. 2209 /// 2210 /// This matches forValueOfCanonicalType except that enums have the 2211 /// full range of their type, not the range of their enumerators. 2212 static IntRange forTargetOfCanonicalType(ASTContext &C, const Type *T) { 2213 assert(T->isCanonicalUnqualified()); 2214 2215 if (const VectorType *VT = dyn_cast<VectorType>(T)) 2216 T = VT->getElementType().getTypePtr(); 2217 if (const ComplexType *CT = dyn_cast<ComplexType>(T)) 2218 T = CT->getElementType().getTypePtr(); 2219 if (const EnumType *ET = dyn_cast<EnumType>(T)) 2220 T = ET->getDecl()->getIntegerType().getTypePtr(); 2221 2222 const BuiltinType *BT = cast<BuiltinType>(T); 2223 assert(BT->isInteger()); 2224 2225 return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger()); 2226 } 2227 2228 /// Returns the supremum of two ranges: i.e. their conservative merge. 2229 static IntRange join(IntRange L, IntRange R) { 2230 return IntRange(std::max(L.Width, R.Width), 2231 L.NonNegative && R.NonNegative); 2232 } 2233 2234 /// Returns the infinum of two ranges: i.e. their aggressive merge. 2235 static IntRange meet(IntRange L, IntRange R) { 2236 return IntRange(std::min(L.Width, R.Width), 2237 L.NonNegative || R.NonNegative); 2238 } 2239}; 2240 2241IntRange GetValueRange(ASTContext &C, llvm::APSInt &value, unsigned MaxWidth) { 2242 if (value.isSigned() && value.isNegative()) 2243 return IntRange(value.getMinSignedBits(), false); 2244 2245 if (value.getBitWidth() > MaxWidth) 2246 value.trunc(MaxWidth); 2247 2248 // isNonNegative() just checks the sign bit without considering 2249 // signedness. 2250 return IntRange(value.getActiveBits(), true); 2251} 2252 2253IntRange GetValueRange(ASTContext &C, APValue &result, QualType Ty, 2254 unsigned MaxWidth) { 2255 if (result.isInt()) 2256 return GetValueRange(C, result.getInt(), MaxWidth); 2257 2258 if (result.isVector()) { 2259 IntRange R = GetValueRange(C, result.getVectorElt(0), Ty, MaxWidth); 2260 for (unsigned i = 1, e = result.getVectorLength(); i != e; ++i) { 2261 IntRange El = GetValueRange(C, result.getVectorElt(i), Ty, MaxWidth); 2262 R = IntRange::join(R, El); 2263 } 2264 return R; 2265 } 2266 2267 if (result.isComplexInt()) { 2268 IntRange R = GetValueRange(C, result.getComplexIntReal(), MaxWidth); 2269 IntRange I = GetValueRange(C, result.getComplexIntImag(), MaxWidth); 2270 return IntRange::join(R, I); 2271 } 2272 2273 // This can happen with lossless casts to intptr_t of "based" lvalues. 2274 // Assume it might use arbitrary bits. 2275 // FIXME: The only reason we need to pass the type in here is to get 2276 // the sign right on this one case. It would be nice if APValue 2277 // preserved this. 2278 assert(result.isLValue()); 2279 return IntRange(MaxWidth, Ty->isUnsignedIntegerType()); 2280} 2281 2282/// Pseudo-evaluate the given integer expression, estimating the 2283/// range of values it might take. 2284/// 2285/// \param MaxWidth - the width to which the value will be truncated 2286IntRange GetExprRange(ASTContext &C, Expr *E, unsigned MaxWidth) { 2287 E = E->IgnoreParens(); 2288 2289 // Try a full evaluation first. 2290 Expr::EvalResult result; 2291 if (E->Evaluate(result, C)) 2292 return GetValueRange(C, result.Val, E->getType(), MaxWidth); 2293 2294 // I think we only want to look through implicit casts here; if the 2295 // user has an explicit widening cast, we should treat the value as 2296 // being of the new, wider type. 2297 if (ImplicitCastExpr *CE = dyn_cast<ImplicitCastExpr>(E)) { 2298 if (CE->getCastKind() == CK_NoOp) 2299 return GetExprRange(C, CE->getSubExpr(), MaxWidth); 2300 2301 IntRange OutputTypeRange = IntRange::forValueOfType(C, CE->getType()); 2302 2303 bool isIntegerCast = (CE->getCastKind() == CK_IntegralCast); 2304 2305 // Assume that non-integer casts can span the full range of the type. 2306 if (!isIntegerCast) 2307 return OutputTypeRange; 2308 2309 IntRange SubRange 2310 = GetExprRange(C, CE->getSubExpr(), 2311 std::min(MaxWidth, OutputTypeRange.Width)); 2312 2313 // Bail out if the subexpr's range is as wide as the cast type. 2314 if (SubRange.Width >= OutputTypeRange.Width) 2315 return OutputTypeRange; 2316 2317 // Otherwise, we take the smaller width, and we're non-negative if 2318 // either the output type or the subexpr is. 2319 return IntRange(SubRange.Width, 2320 SubRange.NonNegative || OutputTypeRange.NonNegative); 2321 } 2322 2323 if (ConditionalOperator *CO = dyn_cast<ConditionalOperator>(E)) { 2324 // If we can fold the condition, just take that operand. 2325 bool CondResult; 2326 if (CO->getCond()->EvaluateAsBooleanCondition(CondResult, C)) 2327 return GetExprRange(C, CondResult ? CO->getTrueExpr() 2328 : CO->getFalseExpr(), 2329 MaxWidth); 2330 2331 // Otherwise, conservatively merge. 2332 IntRange L = GetExprRange(C, CO->getTrueExpr(), MaxWidth); 2333 IntRange R = GetExprRange(C, CO->getFalseExpr(), MaxWidth); 2334 return IntRange::join(L, R); 2335 } 2336 2337 if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) { 2338 switch (BO->getOpcode()) { 2339 2340 // Boolean-valued operations are single-bit and positive. 2341 case BO_LAnd: 2342 case BO_LOr: 2343 case BO_LT: 2344 case BO_GT: 2345 case BO_LE: 2346 case BO_GE: 2347 case BO_EQ: 2348 case BO_NE: 2349 return IntRange::forBoolType(); 2350 2351 // The type of these compound assignments is the type of the LHS, 2352 // so the RHS is not necessarily an integer. 2353 case BO_MulAssign: 2354 case BO_DivAssign: 2355 case BO_RemAssign: 2356 case BO_AddAssign: 2357 case BO_SubAssign: 2358 return IntRange::forValueOfType(C, E->getType()); 2359 2360 // Operations with opaque sources are black-listed. 2361 case BO_PtrMemD: 2362 case BO_PtrMemI: 2363 return IntRange::forValueOfType(C, E->getType()); 2364 2365 // Bitwise-and uses the *infinum* of the two source ranges. 2366 case BO_And: 2367 case BO_AndAssign: 2368 return IntRange::meet(GetExprRange(C, BO->getLHS(), MaxWidth), 2369 GetExprRange(C, BO->getRHS(), MaxWidth)); 2370 2371 // Left shift gets black-listed based on a judgement call. 2372 case BO_Shl: 2373 // ...except that we want to treat '1 << (blah)' as logically 2374 // positive. It's an important idiom. 2375 if (IntegerLiteral *I 2376 = dyn_cast<IntegerLiteral>(BO->getLHS()->IgnoreParenCasts())) { 2377 if (I->getValue() == 1) { 2378 IntRange R = IntRange::forValueOfType(C, E->getType()); 2379 return IntRange(R.Width, /*NonNegative*/ true); 2380 } 2381 } 2382 // fallthrough 2383 2384 case BO_ShlAssign: 2385 return IntRange::forValueOfType(C, E->getType()); 2386 2387 // Right shift by a constant can narrow its left argument. 2388 case BO_Shr: 2389 case BO_ShrAssign: { 2390 IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth); 2391 2392 // If the shift amount is a positive constant, drop the width by 2393 // that much. 2394 llvm::APSInt shift; 2395 if (BO->getRHS()->isIntegerConstantExpr(shift, C) && 2396 shift.isNonNegative()) { 2397 unsigned zext = shift.getZExtValue(); 2398 if (zext >= L.Width) 2399 L.Width = (L.NonNegative ? 0 : 1); 2400 else 2401 L.Width -= zext; 2402 } 2403 2404 return L; 2405 } 2406 2407 // Comma acts as its right operand. 2408 case BO_Comma: 2409 return GetExprRange(C, BO->getRHS(), MaxWidth); 2410 2411 // Black-list pointer subtractions. 2412 case BO_Sub: 2413 if (BO->getLHS()->getType()->isPointerType()) 2414 return IntRange::forValueOfType(C, E->getType()); 2415 // fallthrough 2416 2417 default: 2418 break; 2419 } 2420 2421 // Treat every other operator as if it were closed on the 2422 // narrowest type that encompasses both operands. 2423 IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth); 2424 IntRange R = GetExprRange(C, BO->getRHS(), MaxWidth); 2425 return IntRange::join(L, R); 2426 } 2427 2428 if (UnaryOperator *UO = dyn_cast<UnaryOperator>(E)) { 2429 switch (UO->getOpcode()) { 2430 // Boolean-valued operations are white-listed. 2431 case UO_LNot: 2432 return IntRange::forBoolType(); 2433 2434 // Operations with opaque sources are black-listed. 2435 case UO_Deref: 2436 case UO_AddrOf: // should be impossible 2437 return IntRange::forValueOfType(C, E->getType()); 2438 2439 default: 2440 return GetExprRange(C, UO->getSubExpr(), MaxWidth); 2441 } 2442 } 2443 2444 if (dyn_cast<OffsetOfExpr>(E)) { 2445 IntRange::forValueOfType(C, E->getType()); 2446 } 2447 2448 FieldDecl *BitField = E->getBitField(); 2449 if (BitField) { 2450 llvm::APSInt BitWidthAP = BitField->getBitWidth()->EvaluateAsInt(C); 2451 unsigned BitWidth = BitWidthAP.getZExtValue(); 2452 2453 return IntRange(BitWidth, BitField->getType()->isUnsignedIntegerType()); 2454 } 2455 2456 return IntRange::forValueOfType(C, E->getType()); 2457} 2458 2459IntRange GetExprRange(ASTContext &C, Expr *E) { 2460 return GetExprRange(C, E, C.getIntWidth(E->getType())); 2461} 2462 2463/// Checks whether the given value, which currently has the given 2464/// source semantics, has the same value when coerced through the 2465/// target semantics. 2466bool IsSameFloatAfterCast(const llvm::APFloat &value, 2467 const llvm::fltSemantics &Src, 2468 const llvm::fltSemantics &Tgt) { 2469 llvm::APFloat truncated = value; 2470 2471 bool ignored; 2472 truncated.convert(Src, llvm::APFloat::rmNearestTiesToEven, &ignored); 2473 truncated.convert(Tgt, llvm::APFloat::rmNearestTiesToEven, &ignored); 2474 2475 return truncated.bitwiseIsEqual(value); 2476} 2477 2478/// Checks whether the given value, which currently has the given 2479/// source semantics, has the same value when coerced through the 2480/// target semantics. 2481/// 2482/// The value might be a vector of floats (or a complex number). 2483bool IsSameFloatAfterCast(const APValue &value, 2484 const llvm::fltSemantics &Src, 2485 const llvm::fltSemantics &Tgt) { 2486 if (value.isFloat()) 2487 return IsSameFloatAfterCast(value.getFloat(), Src, Tgt); 2488 2489 if (value.isVector()) { 2490 for (unsigned i = 0, e = value.getVectorLength(); i != e; ++i) 2491 if (!IsSameFloatAfterCast(value.getVectorElt(i), Src, Tgt)) 2492 return false; 2493 return true; 2494 } 2495 2496 assert(value.isComplexFloat()); 2497 return (IsSameFloatAfterCast(value.getComplexFloatReal(), Src, Tgt) && 2498 IsSameFloatAfterCast(value.getComplexFloatImag(), Src, Tgt)); 2499} 2500 2501void AnalyzeImplicitConversions(Sema &S, Expr *E, SourceLocation CC); 2502 2503static bool IsZero(Sema &S, Expr *E) { 2504 // Suppress cases where we are comparing against an enum constant. 2505 if (const DeclRefExpr *DR = 2506 dyn_cast<DeclRefExpr>(E->IgnoreParenImpCasts())) 2507 if (isa<EnumConstantDecl>(DR->getDecl())) 2508 return false; 2509 2510 // Suppress cases where the '0' value is expanded from a macro. 2511 if (E->getLocStart().isMacroID()) 2512 return false; 2513 2514 llvm::APSInt Value; 2515 return E->isIntegerConstantExpr(Value, S.Context) && Value == 0; 2516} 2517 2518static bool HasEnumType(Expr *E) { 2519 // Strip off implicit integral promotions. 2520 while (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(E)) { 2521 if (ICE->getCastKind() != CK_IntegralCast && 2522 ICE->getCastKind() != CK_NoOp) 2523 break; 2524 E = ICE->getSubExpr(); 2525 } 2526 2527 return E->getType()->isEnumeralType(); 2528} 2529 2530void CheckTrivialUnsignedComparison(Sema &S, BinaryOperator *E) { 2531 BinaryOperatorKind op = E->getOpcode(); 2532 if (op == BO_LT && IsZero(S, E->getRHS())) { 2533 S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison) 2534 << "< 0" << "false" << HasEnumType(E->getLHS()) 2535 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 2536 } else if (op == BO_GE && IsZero(S, E->getRHS())) { 2537 S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison) 2538 << ">= 0" << "true" << HasEnumType(E->getLHS()) 2539 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 2540 } else if (op == BO_GT && IsZero(S, E->getLHS())) { 2541 S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison) 2542 << "0 >" << "false" << HasEnumType(E->getRHS()) 2543 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 2544 } else if (op == BO_LE && IsZero(S, E->getLHS())) { 2545 S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison) 2546 << "0 <=" << "true" << HasEnumType(E->getRHS()) 2547 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 2548 } 2549} 2550 2551/// Analyze the operands of the given comparison. Implements the 2552/// fallback case from AnalyzeComparison. 2553void AnalyzeImpConvsInComparison(Sema &S, BinaryOperator *E) { 2554 AnalyzeImplicitConversions(S, E->getLHS(), E->getOperatorLoc()); 2555 AnalyzeImplicitConversions(S, E->getRHS(), E->getOperatorLoc()); 2556} 2557 2558/// \brief Implements -Wsign-compare. 2559/// 2560/// \param lex the left-hand expression 2561/// \param rex the right-hand expression 2562/// \param OpLoc the location of the joining operator 2563/// \param BinOpc binary opcode or 0 2564void AnalyzeComparison(Sema &S, BinaryOperator *E) { 2565 // The type the comparison is being performed in. 2566 QualType T = E->getLHS()->getType(); 2567 assert(S.Context.hasSameUnqualifiedType(T, E->getRHS()->getType()) 2568 && "comparison with mismatched types"); 2569 2570 // We don't do anything special if this isn't an unsigned integral 2571 // comparison: we're only interested in integral comparisons, and 2572 // signed comparisons only happen in cases we don't care to warn about. 2573 if (!T->hasUnsignedIntegerRepresentation()) 2574 return AnalyzeImpConvsInComparison(S, E); 2575 2576 Expr *lex = E->getLHS()->IgnoreParenImpCasts(); 2577 Expr *rex = E->getRHS()->IgnoreParenImpCasts(); 2578 2579 // Check to see if one of the (unmodified) operands is of different 2580 // signedness. 2581 Expr *signedOperand, *unsignedOperand; 2582 if (lex->getType()->hasSignedIntegerRepresentation()) { 2583 assert(!rex->getType()->hasSignedIntegerRepresentation() && 2584 "unsigned comparison between two signed integer expressions?"); 2585 signedOperand = lex; 2586 unsignedOperand = rex; 2587 } else if (rex->getType()->hasSignedIntegerRepresentation()) { 2588 signedOperand = rex; 2589 unsignedOperand = lex; 2590 } else { 2591 CheckTrivialUnsignedComparison(S, E); 2592 return AnalyzeImpConvsInComparison(S, E); 2593 } 2594 2595 // Otherwise, calculate the effective range of the signed operand. 2596 IntRange signedRange = GetExprRange(S.Context, signedOperand); 2597 2598 // Go ahead and analyze implicit conversions in the operands. Note 2599 // that we skip the implicit conversions on both sides. 2600 AnalyzeImplicitConversions(S, lex, E->getOperatorLoc()); 2601 AnalyzeImplicitConversions(S, rex, E->getOperatorLoc()); 2602 2603 // If the signed range is non-negative, -Wsign-compare won't fire, 2604 // but we should still check for comparisons which are always true 2605 // or false. 2606 if (signedRange.NonNegative) 2607 return CheckTrivialUnsignedComparison(S, E); 2608 2609 // For (in)equality comparisons, if the unsigned operand is a 2610 // constant which cannot collide with a overflowed signed operand, 2611 // then reinterpreting the signed operand as unsigned will not 2612 // change the result of the comparison. 2613 if (E->isEqualityOp()) { 2614 unsigned comparisonWidth = S.Context.getIntWidth(T); 2615 IntRange unsignedRange = GetExprRange(S.Context, unsignedOperand); 2616 2617 // We should never be unable to prove that the unsigned operand is 2618 // non-negative. 2619 assert(unsignedRange.NonNegative && "unsigned range includes negative?"); 2620 2621 if (unsignedRange.Width < comparisonWidth) 2622 return; 2623 } 2624 2625 S.Diag(E->getOperatorLoc(), diag::warn_mixed_sign_comparison) 2626 << lex->getType() << rex->getType() 2627 << lex->getSourceRange() << rex->getSourceRange(); 2628} 2629 2630/// Analyzes an attempt to assign the given value to a bitfield. 2631/// 2632/// Returns true if there was something fishy about the attempt. 2633bool AnalyzeBitFieldAssignment(Sema &S, FieldDecl *Bitfield, Expr *Init, 2634 SourceLocation InitLoc) { 2635 assert(Bitfield->isBitField()); 2636 if (Bitfield->isInvalidDecl()) 2637 return false; 2638 2639 // White-list bool bitfields. 2640 if (Bitfield->getType()->isBooleanType()) 2641 return false; 2642 2643 Expr *OriginalInit = Init->IgnoreParenImpCasts(); 2644 2645 llvm::APSInt Width(32); 2646 Expr::EvalResult InitValue; 2647 if (!Bitfield->getBitWidth()->isIntegerConstantExpr(Width, S.Context) || 2648 !OriginalInit->Evaluate(InitValue, S.Context) || 2649 !InitValue.Val.isInt()) 2650 return false; 2651 2652 const llvm::APSInt &Value = InitValue.Val.getInt(); 2653 unsigned OriginalWidth = Value.getBitWidth(); 2654 unsigned FieldWidth = Width.getZExtValue(); 2655 2656 if (OriginalWidth <= FieldWidth) 2657 return false; 2658 2659 llvm::APSInt TruncatedValue = Value; 2660 TruncatedValue.trunc(FieldWidth); 2661 2662 // It's fairly common to write values into signed bitfields 2663 // that, if sign-extended, would end up becoming a different 2664 // value. We don't want to warn about that. 2665 if (Value.isSigned() && Value.isNegative()) 2666 TruncatedValue.sext(OriginalWidth); 2667 else 2668 TruncatedValue.zext(OriginalWidth); 2669 2670 if (Value == TruncatedValue) 2671 return false; 2672 2673 std::string PrettyValue = Value.toString(10); 2674 std::string PrettyTrunc = TruncatedValue.toString(10); 2675 2676 S.Diag(InitLoc, diag::warn_impcast_bitfield_precision_constant) 2677 << PrettyValue << PrettyTrunc << OriginalInit->getType() 2678 << Init->getSourceRange(); 2679 2680 return true; 2681} 2682 2683/// Analyze the given simple or compound assignment for warning-worthy 2684/// operations. 2685void AnalyzeAssignment(Sema &S, BinaryOperator *E) { 2686 // Just recurse on the LHS. 2687 AnalyzeImplicitConversions(S, E->getLHS(), E->getOperatorLoc()); 2688 2689 // We want to recurse on the RHS as normal unless we're assigning to 2690 // a bitfield. 2691 if (FieldDecl *Bitfield = E->getLHS()->getBitField()) { 2692 if (AnalyzeBitFieldAssignment(S, Bitfield, E->getRHS(), 2693 E->getOperatorLoc())) { 2694 // Recurse, ignoring any implicit conversions on the RHS. 2695 return AnalyzeImplicitConversions(S, E->getRHS()->IgnoreParenImpCasts(), 2696 E->getOperatorLoc()); 2697 } 2698 } 2699 2700 AnalyzeImplicitConversions(S, E->getRHS(), E->getOperatorLoc()); 2701} 2702 2703/// Diagnose an implicit cast; purely a helper for CheckImplicitConversion. 2704void DiagnoseImpCast(Sema &S, Expr *E, QualType T, SourceLocation CContext, 2705 unsigned diag) { 2706 S.Diag(E->getExprLoc(), diag) 2707 << E->getType() << T << E->getSourceRange() << SourceRange(CContext); 2708} 2709 2710std::string PrettyPrintInRange(const llvm::APSInt &Value, IntRange Range) { 2711 if (!Range.Width) return "0"; 2712 2713 llvm::APSInt ValueInRange = Value; 2714 ValueInRange.setIsSigned(!Range.NonNegative); 2715 ValueInRange.trunc(Range.Width); 2716 return ValueInRange.toString(10); 2717} 2718 2719void CheckImplicitConversion(Sema &S, Expr *E, QualType T, 2720 SourceLocation CC, bool *ICContext = 0) { 2721 if (E->isTypeDependent() || E->isValueDependent()) return; 2722 2723 const Type *Source = S.Context.getCanonicalType(E->getType()).getTypePtr(); 2724 const Type *Target = S.Context.getCanonicalType(T).getTypePtr(); 2725 if (Source == Target) return; 2726 if (Target->isDependentType()) return; 2727 2728 // If the conversion context location is invalid or instantiated 2729 // from a system macro, don't complain. 2730 if (CC.isInvalid() || 2731 (CC.isMacroID() && S.Context.getSourceManager().isInSystemHeader( 2732 S.Context.getSourceManager().getSpellingLoc(CC)))) 2733 return; 2734 2735 // Never diagnose implicit casts to bool. 2736 if (Target->isSpecificBuiltinType(BuiltinType::Bool)) 2737 return; 2738 2739 // Strip vector types. 2740 if (isa<VectorType>(Source)) { 2741 if (!isa<VectorType>(Target)) 2742 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_vector_scalar); 2743 2744 Source = cast<VectorType>(Source)->getElementType().getTypePtr(); 2745 Target = cast<VectorType>(Target)->getElementType().getTypePtr(); 2746 } 2747 2748 // Strip complex types. 2749 if (isa<ComplexType>(Source)) { 2750 if (!isa<ComplexType>(Target)) 2751 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_complex_scalar); 2752 2753 Source = cast<ComplexType>(Source)->getElementType().getTypePtr(); 2754 Target = cast<ComplexType>(Target)->getElementType().getTypePtr(); 2755 } 2756 2757 const BuiltinType *SourceBT = dyn_cast<BuiltinType>(Source); 2758 const BuiltinType *TargetBT = dyn_cast<BuiltinType>(Target); 2759 2760 // If the source is floating point... 2761 if (SourceBT && SourceBT->isFloatingPoint()) { 2762 // ...and the target is floating point... 2763 if (TargetBT && TargetBT->isFloatingPoint()) { 2764 // ...then warn if we're dropping FP rank. 2765 2766 // Builtin FP kinds are ordered by increasing FP rank. 2767 if (SourceBT->getKind() > TargetBT->getKind()) { 2768 // Don't warn about float constants that are precisely 2769 // representable in the target type. 2770 Expr::EvalResult result; 2771 if (E->Evaluate(result, S.Context)) { 2772 // Value might be a float, a float vector, or a float complex. 2773 if (IsSameFloatAfterCast(result.Val, 2774 S.Context.getFloatTypeSemantics(QualType(TargetBT, 0)), 2775 S.Context.getFloatTypeSemantics(QualType(SourceBT, 0)))) 2776 return; 2777 } 2778 2779 DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_float_precision); 2780 } 2781 return; 2782 } 2783 2784 // If the target is integral, always warn. 2785 if ((TargetBT && TargetBT->isInteger())) 2786 // TODO: don't warn for integer values? 2787 DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_float_integer); 2788 2789 return; 2790 } 2791 2792 if (!Source->isIntegerType() || !Target->isIntegerType()) 2793 return; 2794 2795 IntRange SourceRange = GetExprRange(S.Context, E); 2796 IntRange TargetRange = IntRange::forTargetOfCanonicalType(S.Context, Target); 2797 2798 if (SourceRange.Width > TargetRange.Width) { 2799 // If the source is a constant, use a default-on diagnostic. 2800 // TODO: this should happen for bitfield stores, too. 2801 llvm::APSInt Value(32); 2802 if (E->isIntegerConstantExpr(Value, S.Context)) { 2803 std::string PrettySourceValue = Value.toString(10); 2804 std::string PrettyTargetValue = PrettyPrintInRange(Value, TargetRange); 2805 2806 S.Diag(E->getExprLoc(), diag::warn_impcast_integer_precision_constant) 2807 << PrettySourceValue << PrettyTargetValue 2808 << E->getType() << T << E->getSourceRange() << clang::SourceRange(CC); 2809 return; 2810 } 2811 2812 // People want to build with -Wshorten-64-to-32 and not -Wconversion 2813 // and by god we'll let them. 2814 if (SourceRange.Width == 64 && TargetRange.Width == 32) 2815 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_integer_64_32); 2816 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_integer_precision); 2817 } 2818 2819 if ((TargetRange.NonNegative && !SourceRange.NonNegative) || 2820 (!TargetRange.NonNegative && SourceRange.NonNegative && 2821 SourceRange.Width == TargetRange.Width)) { 2822 unsigned DiagID = diag::warn_impcast_integer_sign; 2823 2824 // Traditionally, gcc has warned about this under -Wsign-compare. 2825 // We also want to warn about it in -Wconversion. 2826 // So if -Wconversion is off, use a completely identical diagnostic 2827 // in the sign-compare group. 2828 // The conditional-checking code will 2829 if (ICContext) { 2830 DiagID = diag::warn_impcast_integer_sign_conditional; 2831 *ICContext = true; 2832 } 2833 2834 return DiagnoseImpCast(S, E, T, CC, DiagID); 2835 } 2836 2837 return; 2838} 2839 2840void CheckConditionalOperator(Sema &S, ConditionalOperator *E, QualType T); 2841 2842void CheckConditionalOperand(Sema &S, Expr *E, QualType T, 2843 SourceLocation CC, bool &ICContext) { 2844 E = E->IgnoreParenImpCasts(); 2845 2846 if (isa<ConditionalOperator>(E)) 2847 return CheckConditionalOperator(S, cast<ConditionalOperator>(E), T); 2848 2849 AnalyzeImplicitConversions(S, E, CC); 2850 if (E->getType() != T) 2851 return CheckImplicitConversion(S, E, T, CC, &ICContext); 2852 return; 2853} 2854 2855void CheckConditionalOperator(Sema &S, ConditionalOperator *E, QualType T) { 2856 SourceLocation CC = E->getQuestionLoc(); 2857 2858 AnalyzeImplicitConversions(S, E->getCond(), CC); 2859 2860 bool Suspicious = false; 2861 CheckConditionalOperand(S, E->getTrueExpr(), T, CC, Suspicious); 2862 CheckConditionalOperand(S, E->getFalseExpr(), T, CC, Suspicious); 2863 2864 // If -Wconversion would have warned about either of the candidates 2865 // for a signedness conversion to the context type... 2866 if (!Suspicious) return; 2867 2868 // ...but it's currently ignored... 2869 if (S.Diags.getDiagnosticLevel(diag::warn_impcast_integer_sign_conditional)) 2870 return; 2871 2872 // ...and -Wsign-compare isn't... 2873 if (!S.Diags.getDiagnosticLevel(diag::warn_mixed_sign_conditional)) 2874 return; 2875 2876 // ...then check whether it would have warned about either of the 2877 // candidates for a signedness conversion to the condition type. 2878 if (E->getType() != T) { 2879 Suspicious = false; 2880 CheckImplicitConversion(S, E->getTrueExpr()->IgnoreParenImpCasts(), 2881 E->getType(), CC, &Suspicious); 2882 if (!Suspicious) 2883 CheckImplicitConversion(S, E->getFalseExpr()->IgnoreParenImpCasts(), 2884 E->getType(), CC, &Suspicious); 2885 if (!Suspicious) 2886 return; 2887 } 2888 2889 // If so, emit a diagnostic under -Wsign-compare. 2890 Expr *lex = E->getTrueExpr()->IgnoreParenImpCasts(); 2891 Expr *rex = E->getFalseExpr()->IgnoreParenImpCasts(); 2892 S.Diag(E->getQuestionLoc(), diag::warn_mixed_sign_conditional) 2893 << lex->getType() << rex->getType() 2894 << lex->getSourceRange() << rex->getSourceRange(); 2895} 2896 2897/// AnalyzeImplicitConversions - Find and report any interesting 2898/// implicit conversions in the given expression. There are a couple 2899/// of competing diagnostics here, -Wconversion and -Wsign-compare. 2900void AnalyzeImplicitConversions(Sema &S, Expr *OrigE, SourceLocation CC) { 2901 QualType T = OrigE->getType(); 2902 Expr *E = OrigE->IgnoreParenImpCasts(); 2903 2904 // For conditional operators, we analyze the arguments as if they 2905 // were being fed directly into the output. 2906 if (isa<ConditionalOperator>(E)) { 2907 ConditionalOperator *CO = cast<ConditionalOperator>(E); 2908 CheckConditionalOperator(S, CO, T); 2909 return; 2910 } 2911 2912 // Go ahead and check any implicit conversions we might have skipped. 2913 // The non-canonical typecheck is just an optimization; 2914 // CheckImplicitConversion will filter out dead implicit conversions. 2915 if (E->getType() != T) 2916 CheckImplicitConversion(S, E, T, CC); 2917 2918 // Now continue drilling into this expression. 2919 2920 // Skip past explicit casts. 2921 if (isa<ExplicitCastExpr>(E)) { 2922 E = cast<ExplicitCastExpr>(E)->getSubExpr()->IgnoreParenImpCasts(); 2923 return AnalyzeImplicitConversions(S, E, CC); 2924 } 2925 2926 if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) { 2927 // Do a somewhat different check with comparison operators. 2928 if (BO->isComparisonOp()) 2929 return AnalyzeComparison(S, BO); 2930 2931 // And with assignments and compound assignments. 2932 if (BO->isAssignmentOp()) 2933 return AnalyzeAssignment(S, BO); 2934 } 2935 2936 // These break the otherwise-useful invariant below. Fortunately, 2937 // we don't really need to recurse into them, because any internal 2938 // expressions should have been analyzed already when they were 2939 // built into statements. 2940 if (isa<StmtExpr>(E)) return; 2941 2942 // Don't descend into unevaluated contexts. 2943 if (isa<SizeOfAlignOfExpr>(E)) return; 2944 2945 // Now just recurse over the expression's children. 2946 CC = E->getExprLoc(); 2947 for (Stmt::child_iterator I = E->child_begin(), IE = E->child_end(); 2948 I != IE; ++I) 2949 AnalyzeImplicitConversions(S, cast<Expr>(*I), CC); 2950} 2951 2952} // end anonymous namespace 2953 2954/// Diagnoses "dangerous" implicit conversions within the given 2955/// expression (which is a full expression). Implements -Wconversion 2956/// and -Wsign-compare. 2957/// 2958/// \param CC the "context" location of the implicit conversion, i.e. 2959/// the most location of the syntactic entity requiring the implicit 2960/// conversion 2961void Sema::CheckImplicitConversions(Expr *E, SourceLocation CC) { 2962 // Don't diagnose in unevaluated contexts. 2963 if (ExprEvalContexts.back().Context == Sema::Unevaluated) 2964 return; 2965 2966 // Don't diagnose for value- or type-dependent expressions. 2967 if (E->isTypeDependent() || E->isValueDependent()) 2968 return; 2969 2970 // This is not the right CC for (e.g.) a variable initialization. 2971 AnalyzeImplicitConversions(*this, E, CC); 2972} 2973 2974void Sema::CheckBitFieldInitialization(SourceLocation InitLoc, 2975 FieldDecl *BitField, 2976 Expr *Init) { 2977 (void) AnalyzeBitFieldAssignment(*this, BitField, Init, InitLoc); 2978} 2979 2980/// CheckParmsForFunctionDef - Check that the parameters of the given 2981/// function are appropriate for the definition of a function. This 2982/// takes care of any checks that cannot be performed on the 2983/// declaration itself, e.g., that the types of each of the function 2984/// parameters are complete. 2985bool Sema::CheckParmsForFunctionDef(ParmVarDecl **P, ParmVarDecl **PEnd, 2986 bool CheckParameterNames) { 2987 bool HasInvalidParm = false; 2988 for (; P != PEnd; ++P) { 2989 ParmVarDecl *Param = *P; 2990 2991 // C99 6.7.5.3p4: the parameters in a parameter type list in a 2992 // function declarator that is part of a function definition of 2993 // that function shall not have incomplete type. 2994 // 2995 // This is also C++ [dcl.fct]p6. 2996 if (!Param->isInvalidDecl() && 2997 RequireCompleteType(Param->getLocation(), Param->getType(), 2998 diag::err_typecheck_decl_incomplete_type)) { 2999 Param->setInvalidDecl(); 3000 HasInvalidParm = true; 3001 } 3002 3003 // C99 6.9.1p5: If the declarator includes a parameter type list, the 3004 // declaration of each parameter shall include an identifier. 3005 if (CheckParameterNames && 3006 Param->getIdentifier() == 0 && 3007 !Param->isImplicit() && 3008 !getLangOptions().CPlusPlus) 3009 Diag(Param->getLocation(), diag::err_parameter_name_omitted); 3010 3011 // C99 6.7.5.3p12: 3012 // If the function declarator is not part of a definition of that 3013 // function, parameters may have incomplete type and may use the [*] 3014 // notation in their sequences of declarator specifiers to specify 3015 // variable length array types. 3016 QualType PType = Param->getOriginalType(); 3017 if (const ArrayType *AT = Context.getAsArrayType(PType)) { 3018 if (AT->getSizeModifier() == ArrayType::Star) { 3019 // FIXME: This diagnosic should point the the '[*]' if source-location 3020 // information is added for it. 3021 Diag(Param->getLocation(), diag::err_array_star_in_function_definition); 3022 } 3023 } 3024 } 3025 3026 return HasInvalidParm; 3027} 3028 3029/// CheckCastAlign - Implements -Wcast-align, which warns when a 3030/// pointer cast increases the alignment requirements. 3031void Sema::CheckCastAlign(Expr *Op, QualType T, SourceRange TRange) { 3032 // This is actually a lot of work to potentially be doing on every 3033 // cast; don't do it if we're ignoring -Wcast_align (as is the default). 3034 if (getDiagnostics().getDiagnosticLevel(diag::warn_cast_align) 3035 == Diagnostic::Ignored) 3036 return; 3037 3038 // Ignore dependent types. 3039 if (T->isDependentType() || Op->getType()->isDependentType()) 3040 return; 3041 3042 // Require that the destination be a pointer type. 3043 const PointerType *DestPtr = T->getAs<PointerType>(); 3044 if (!DestPtr) return; 3045 3046 // If the destination has alignment 1, we're done. 3047 QualType DestPointee = DestPtr->getPointeeType(); 3048 if (DestPointee->isIncompleteType()) return; 3049 CharUnits DestAlign = Context.getTypeAlignInChars(DestPointee); 3050 if (DestAlign.isOne()) return; 3051 3052 // Require that the source be a pointer type. 3053 const PointerType *SrcPtr = Op->getType()->getAs<PointerType>(); 3054 if (!SrcPtr) return; 3055 QualType SrcPointee = SrcPtr->getPointeeType(); 3056 3057 // Whitelist casts from cv void*. We already implicitly 3058 // whitelisted casts to cv void*, since they have alignment 1. 3059 // Also whitelist casts involving incomplete types, which implicitly 3060 // includes 'void'. 3061 if (SrcPointee->isIncompleteType()) return; 3062 3063 CharUnits SrcAlign = Context.getTypeAlignInChars(SrcPointee); 3064 if (SrcAlign >= DestAlign) return; 3065 3066 Diag(TRange.getBegin(), diag::warn_cast_align) 3067 << Op->getType() << T 3068 << static_cast<unsigned>(SrcAlign.getQuantity()) 3069 << static_cast<unsigned>(DestAlign.getQuantity()) 3070 << TRange << Op->getSourceRange(); 3071} 3072 3073