SemaChecking.cpp revision da51f0d136df131a0137c0ec1069fb586d8a296a
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 "Sema.h" 16#include "clang/Analysis/CFG.h" 17#include "clang/Analysis/AnalysisContext.h" 18#include "clang/Analysis/Analyses/PrintfFormatString.h" 19#include "clang/AST/ASTContext.h" 20#include "clang/AST/CharUnits.h" 21#include "clang/AST/DeclObjC.h" 22#include "clang/AST/ExprCXX.h" 23#include "clang/AST/ExprObjC.h" 24#include "clang/AST/DeclObjC.h" 25#include "clang/AST/StmtCXX.h" 26#include "clang/AST/StmtObjC.h" 27#include "clang/Lex/LiteralSupport.h" 28#include "clang/Lex/Preprocessor.h" 29#include "llvm/ADT/BitVector.h" 30#include "llvm/ADT/STLExtras.h" 31#include <limits> 32#include <queue> 33using namespace clang; 34 35/// getLocationOfStringLiteralByte - Return a source location that points to the 36/// specified byte of the specified string literal. 37/// 38/// Strings are amazingly complex. They can be formed from multiple tokens and 39/// can have escape sequences in them in addition to the usual trigraph and 40/// escaped newline business. This routine handles this complexity. 41/// 42SourceLocation Sema::getLocationOfStringLiteralByte(const StringLiteral *SL, 43 unsigned ByteNo) const { 44 assert(!SL->isWide() && "This doesn't work for wide strings yet"); 45 46 // Loop over all of the tokens in this string until we find the one that 47 // contains the byte we're looking for. 48 unsigned TokNo = 0; 49 while (1) { 50 assert(TokNo < SL->getNumConcatenated() && "Invalid byte number!"); 51 SourceLocation StrTokLoc = SL->getStrTokenLoc(TokNo); 52 53 // Get the spelling of the string so that we can get the data that makes up 54 // the string literal, not the identifier for the macro it is potentially 55 // expanded through. 56 SourceLocation StrTokSpellingLoc = SourceMgr.getSpellingLoc(StrTokLoc); 57 58 // Re-lex the token to get its length and original spelling. 59 std::pair<FileID, unsigned> LocInfo = 60 SourceMgr.getDecomposedLoc(StrTokSpellingLoc); 61 std::pair<const char *,const char *> Buffer = 62 SourceMgr.getBufferData(LocInfo.first); 63 const char *StrData = Buffer.first+LocInfo.second; 64 65 // Create a langops struct and enable trigraphs. This is sufficient for 66 // relexing tokens. 67 LangOptions LangOpts; 68 LangOpts.Trigraphs = true; 69 70 // Create a lexer starting at the beginning of this token. 71 Lexer TheLexer(StrTokSpellingLoc, LangOpts, Buffer.first, StrData, 72 Buffer.second); 73 Token TheTok; 74 TheLexer.LexFromRawLexer(TheTok); 75 76 // Use the StringLiteralParser to compute the length of the string in bytes. 77 StringLiteralParser SLP(&TheTok, 1, PP); 78 unsigned TokNumBytes = SLP.GetStringLength(); 79 80 // If the byte is in this token, return the location of the byte. 81 if (ByteNo < TokNumBytes || 82 (ByteNo == TokNumBytes && TokNo == SL->getNumConcatenated())) { 83 unsigned Offset = 84 StringLiteralParser::getOffsetOfStringByte(TheTok, ByteNo, PP); 85 86 // Now that we know the offset of the token in the spelling, use the 87 // preprocessor to get the offset in the original source. 88 return PP.AdvanceToTokenCharacter(StrTokLoc, Offset); 89 } 90 91 // Move to the next string token. 92 ++TokNo; 93 ByteNo -= TokNumBytes; 94 } 95} 96 97/// CheckablePrintfAttr - does a function call have a "printf" attribute 98/// and arguments that merit checking? 99bool Sema::CheckablePrintfAttr(const FormatAttr *Format, CallExpr *TheCall) { 100 if (Format->getType() == "printf") return true; 101 if (Format->getType() == "printf0") { 102 // printf0 allows null "format" string; if so don't check format/args 103 unsigned format_idx = Format->getFormatIdx() - 1; 104 // Does the index refer to the implicit object argument? 105 if (isa<CXXMemberCallExpr>(TheCall)) { 106 if (format_idx == 0) 107 return false; 108 --format_idx; 109 } 110 if (format_idx < TheCall->getNumArgs()) { 111 Expr *Format = TheCall->getArg(format_idx)->IgnoreParenCasts(); 112 if (!Format->isNullPointerConstant(Context, Expr::NPC_ValueDependentIsNull)) 113 return true; 114 } 115 } 116 return false; 117} 118 119Action::OwningExprResult 120Sema::CheckBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) { 121 OwningExprResult TheCallResult(Owned(TheCall)); 122 123 switch (BuiltinID) { 124 case Builtin::BI__builtin___CFStringMakeConstantString: 125 assert(TheCall->getNumArgs() == 1 && 126 "Wrong # arguments to builtin CFStringMakeConstantString"); 127 if (CheckObjCString(TheCall->getArg(0))) 128 return ExprError(); 129 break; 130 case Builtin::BI__builtin_stdarg_start: 131 case Builtin::BI__builtin_va_start: 132 if (SemaBuiltinVAStart(TheCall)) 133 return ExprError(); 134 break; 135 case Builtin::BI__builtin_isgreater: 136 case Builtin::BI__builtin_isgreaterequal: 137 case Builtin::BI__builtin_isless: 138 case Builtin::BI__builtin_islessequal: 139 case Builtin::BI__builtin_islessgreater: 140 case Builtin::BI__builtin_isunordered: 141 if (SemaBuiltinUnorderedCompare(TheCall)) 142 return ExprError(); 143 break; 144 case Builtin::BI__builtin_isfinite: 145 case Builtin::BI__builtin_isinf: 146 case Builtin::BI__builtin_isinf_sign: 147 case Builtin::BI__builtin_isnan: 148 case Builtin::BI__builtin_isnormal: 149 if (SemaBuiltinUnaryFP(TheCall)) 150 return ExprError(); 151 break; 152 case Builtin::BI__builtin_return_address: 153 case Builtin::BI__builtin_frame_address: 154 if (SemaBuiltinStackAddress(TheCall)) 155 return ExprError(); 156 break; 157 case Builtin::BI__builtin_eh_return_data_regno: 158 if (SemaBuiltinEHReturnDataRegNo(TheCall)) 159 return ExprError(); 160 break; 161 case Builtin::BI__builtin_shufflevector: 162 return SemaBuiltinShuffleVector(TheCall); 163 // TheCall will be freed by the smart pointer here, but that's fine, since 164 // SemaBuiltinShuffleVector guts it, but then doesn't release it. 165 case Builtin::BI__builtin_prefetch: 166 if (SemaBuiltinPrefetch(TheCall)) 167 return ExprError(); 168 break; 169 case Builtin::BI__builtin_object_size: 170 if (SemaBuiltinObjectSize(TheCall)) 171 return ExprError(); 172 break; 173 case Builtin::BI__builtin_longjmp: 174 if (SemaBuiltinLongjmp(TheCall)) 175 return ExprError(); 176 break; 177 case Builtin::BI__sync_fetch_and_add: 178 case Builtin::BI__sync_fetch_and_sub: 179 case Builtin::BI__sync_fetch_and_or: 180 case Builtin::BI__sync_fetch_and_and: 181 case Builtin::BI__sync_fetch_and_xor: 182 case Builtin::BI__sync_fetch_and_nand: 183 case Builtin::BI__sync_add_and_fetch: 184 case Builtin::BI__sync_sub_and_fetch: 185 case Builtin::BI__sync_and_and_fetch: 186 case Builtin::BI__sync_or_and_fetch: 187 case Builtin::BI__sync_xor_and_fetch: 188 case Builtin::BI__sync_nand_and_fetch: 189 case Builtin::BI__sync_val_compare_and_swap: 190 case Builtin::BI__sync_bool_compare_and_swap: 191 case Builtin::BI__sync_lock_test_and_set: 192 case Builtin::BI__sync_lock_release: 193 if (SemaBuiltinAtomicOverloaded(TheCall)) 194 return ExprError(); 195 break; 196 } 197 198 return move(TheCallResult); 199} 200 201/// CheckFunctionCall - Check a direct function call for various correctness 202/// and safety properties not strictly enforced by the C type system. 203bool Sema::CheckFunctionCall(FunctionDecl *FDecl, CallExpr *TheCall) { 204 // Get the IdentifierInfo* for the called function. 205 IdentifierInfo *FnInfo = FDecl->getIdentifier(); 206 207 // None of the checks below are needed for functions that don't have 208 // simple names (e.g., C++ conversion functions). 209 if (!FnInfo) 210 return false; 211 212 // FIXME: This mechanism should be abstracted to be less fragile and 213 // more efficient. For example, just map function ids to custom 214 // handlers. 215 216 // Printf checking. 217 if (const FormatAttr *Format = FDecl->getAttr<FormatAttr>()) { 218 if (CheckablePrintfAttr(Format, TheCall)) { 219 bool HasVAListArg = Format->getFirstArg() == 0; 220 if (!HasVAListArg) { 221 if (const FunctionProtoType *Proto 222 = FDecl->getType()->getAs<FunctionProtoType>()) 223 HasVAListArg = !Proto->isVariadic(); 224 } 225 CheckPrintfArguments(TheCall, HasVAListArg, Format->getFormatIdx() - 1, 226 HasVAListArg ? 0 : Format->getFirstArg() - 1); 227 } 228 } 229 230 for (const NonNullAttr *NonNull = FDecl->getAttr<NonNullAttr>(); NonNull; 231 NonNull = NonNull->getNext<NonNullAttr>()) 232 CheckNonNullArguments(NonNull, TheCall); 233 234 return false; 235} 236 237bool Sema::CheckBlockCall(NamedDecl *NDecl, CallExpr *TheCall) { 238 // Printf checking. 239 const FormatAttr *Format = NDecl->getAttr<FormatAttr>(); 240 if (!Format) 241 return false; 242 243 const VarDecl *V = dyn_cast<VarDecl>(NDecl); 244 if (!V) 245 return false; 246 247 QualType Ty = V->getType(); 248 if (!Ty->isBlockPointerType()) 249 return false; 250 251 if (!CheckablePrintfAttr(Format, TheCall)) 252 return false; 253 254 bool HasVAListArg = Format->getFirstArg() == 0; 255 if (!HasVAListArg) { 256 const FunctionType *FT = 257 Ty->getAs<BlockPointerType>()->getPointeeType()->getAs<FunctionType>(); 258 if (const FunctionProtoType *Proto = dyn_cast<FunctionProtoType>(FT)) 259 HasVAListArg = !Proto->isVariadic(); 260 } 261 CheckPrintfArguments(TheCall, HasVAListArg, Format->getFormatIdx() - 1, 262 HasVAListArg ? 0 : Format->getFirstArg() - 1); 263 264 return false; 265} 266 267/// SemaBuiltinAtomicOverloaded - We have a call to a function like 268/// __sync_fetch_and_add, which is an overloaded function based on the pointer 269/// type of its first argument. The main ActOnCallExpr routines have already 270/// promoted the types of arguments because all of these calls are prototyped as 271/// void(...). 272/// 273/// This function goes through and does final semantic checking for these 274/// builtins, 275bool Sema::SemaBuiltinAtomicOverloaded(CallExpr *TheCall) { 276 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 277 FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl()); 278 279 // Ensure that we have at least one argument to do type inference from. 280 if (TheCall->getNumArgs() < 1) 281 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 282 << 0 << TheCall->getCallee()->getSourceRange(); 283 284 // Inspect the first argument of the atomic builtin. This should always be 285 // a pointer type, whose element is an integral scalar or pointer type. 286 // Because it is a pointer type, we don't have to worry about any implicit 287 // casts here. 288 Expr *FirstArg = TheCall->getArg(0); 289 if (!FirstArg->getType()->isPointerType()) 290 return Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer) 291 << FirstArg->getType() << FirstArg->getSourceRange(); 292 293 QualType ValType = FirstArg->getType()->getAs<PointerType>()->getPointeeType(); 294 if (!ValType->isIntegerType() && !ValType->isPointerType() && 295 !ValType->isBlockPointerType()) 296 return Diag(DRE->getLocStart(), 297 diag::err_atomic_builtin_must_be_pointer_intptr) 298 << FirstArg->getType() << FirstArg->getSourceRange(); 299 300 // We need to figure out which concrete builtin this maps onto. For example, 301 // __sync_fetch_and_add with a 2 byte object turns into 302 // __sync_fetch_and_add_2. 303#define BUILTIN_ROW(x) \ 304 { Builtin::BI##x##_1, Builtin::BI##x##_2, Builtin::BI##x##_4, \ 305 Builtin::BI##x##_8, Builtin::BI##x##_16 } 306 307 static const unsigned BuiltinIndices[][5] = { 308 BUILTIN_ROW(__sync_fetch_and_add), 309 BUILTIN_ROW(__sync_fetch_and_sub), 310 BUILTIN_ROW(__sync_fetch_and_or), 311 BUILTIN_ROW(__sync_fetch_and_and), 312 BUILTIN_ROW(__sync_fetch_and_xor), 313 BUILTIN_ROW(__sync_fetch_and_nand), 314 315 BUILTIN_ROW(__sync_add_and_fetch), 316 BUILTIN_ROW(__sync_sub_and_fetch), 317 BUILTIN_ROW(__sync_and_and_fetch), 318 BUILTIN_ROW(__sync_or_and_fetch), 319 BUILTIN_ROW(__sync_xor_and_fetch), 320 BUILTIN_ROW(__sync_nand_and_fetch), 321 322 BUILTIN_ROW(__sync_val_compare_and_swap), 323 BUILTIN_ROW(__sync_bool_compare_and_swap), 324 BUILTIN_ROW(__sync_lock_test_and_set), 325 BUILTIN_ROW(__sync_lock_release) 326 }; 327#undef BUILTIN_ROW 328 329 // Determine the index of the size. 330 unsigned SizeIndex; 331 switch (Context.getTypeSizeInChars(ValType).getQuantity()) { 332 case 1: SizeIndex = 0; break; 333 case 2: SizeIndex = 1; break; 334 case 4: SizeIndex = 2; break; 335 case 8: SizeIndex = 3; break; 336 case 16: SizeIndex = 4; break; 337 default: 338 return Diag(DRE->getLocStart(), diag::err_atomic_builtin_pointer_size) 339 << FirstArg->getType() << FirstArg->getSourceRange(); 340 } 341 342 // Each of these builtins has one pointer argument, followed by some number of 343 // values (0, 1 or 2) followed by a potentially empty varags list of stuff 344 // that we ignore. Find out which row of BuiltinIndices to read from as well 345 // as the number of fixed args. 346 unsigned BuiltinID = FDecl->getBuiltinID(); 347 unsigned BuiltinIndex, NumFixed = 1; 348 switch (BuiltinID) { 349 default: assert(0 && "Unknown overloaded atomic builtin!"); 350 case Builtin::BI__sync_fetch_and_add: BuiltinIndex = 0; break; 351 case Builtin::BI__sync_fetch_and_sub: BuiltinIndex = 1; break; 352 case Builtin::BI__sync_fetch_and_or: BuiltinIndex = 2; break; 353 case Builtin::BI__sync_fetch_and_and: BuiltinIndex = 3; break; 354 case Builtin::BI__sync_fetch_and_xor: BuiltinIndex = 4; break; 355 case Builtin::BI__sync_fetch_and_nand:BuiltinIndex = 5; break; 356 357 case Builtin::BI__sync_add_and_fetch: BuiltinIndex = 6; break; 358 case Builtin::BI__sync_sub_and_fetch: BuiltinIndex = 7; break; 359 case Builtin::BI__sync_and_and_fetch: BuiltinIndex = 8; break; 360 case Builtin::BI__sync_or_and_fetch: BuiltinIndex = 9; break; 361 case Builtin::BI__sync_xor_and_fetch: BuiltinIndex =10; break; 362 case Builtin::BI__sync_nand_and_fetch:BuiltinIndex =11; break; 363 364 case Builtin::BI__sync_val_compare_and_swap: 365 BuiltinIndex = 12; 366 NumFixed = 2; 367 break; 368 case Builtin::BI__sync_bool_compare_and_swap: 369 BuiltinIndex = 13; 370 NumFixed = 2; 371 break; 372 case Builtin::BI__sync_lock_test_and_set: BuiltinIndex = 14; break; 373 case Builtin::BI__sync_lock_release: 374 BuiltinIndex = 15; 375 NumFixed = 0; 376 break; 377 } 378 379 // Now that we know how many fixed arguments we expect, first check that we 380 // have at least that many. 381 if (TheCall->getNumArgs() < 1+NumFixed) 382 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 383 << 0 << TheCall->getCallee()->getSourceRange(); 384 385 386 // Get the decl for the concrete builtin from this, we can tell what the 387 // concrete integer type we should convert to is. 388 unsigned NewBuiltinID = BuiltinIndices[BuiltinIndex][SizeIndex]; 389 const char *NewBuiltinName = Context.BuiltinInfo.GetName(NewBuiltinID); 390 IdentifierInfo *NewBuiltinII = PP.getIdentifierInfo(NewBuiltinName); 391 FunctionDecl *NewBuiltinDecl = 392 cast<FunctionDecl>(LazilyCreateBuiltin(NewBuiltinII, NewBuiltinID, 393 TUScope, false, DRE->getLocStart())); 394 const FunctionProtoType *BuiltinFT = 395 NewBuiltinDecl->getType()->getAs<FunctionProtoType>(); 396 ValType = BuiltinFT->getArgType(0)->getAs<PointerType>()->getPointeeType(); 397 398 // If the first type needs to be converted (e.g. void** -> int*), do it now. 399 if (BuiltinFT->getArgType(0) != FirstArg->getType()) { 400 ImpCastExprToType(FirstArg, BuiltinFT->getArgType(0), CastExpr::CK_BitCast); 401 TheCall->setArg(0, FirstArg); 402 } 403 404 // Next, walk the valid ones promoting to the right type. 405 for (unsigned i = 0; i != NumFixed; ++i) { 406 Expr *Arg = TheCall->getArg(i+1); 407 408 // If the argument is an implicit cast, then there was a promotion due to 409 // "...", just remove it now. 410 if (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(Arg)) { 411 Arg = ICE->getSubExpr(); 412 ICE->setSubExpr(0); 413 ICE->Destroy(Context); 414 TheCall->setArg(i+1, Arg); 415 } 416 417 // GCC does an implicit conversion to the pointer or integer ValType. This 418 // can fail in some cases (1i -> int**), check for this error case now. 419 CastExpr::CastKind Kind = CastExpr::CK_Unknown; 420 CXXMethodDecl *ConversionDecl = 0; 421 if (CheckCastTypes(Arg->getSourceRange(), ValType, Arg, Kind, 422 ConversionDecl)) 423 return true; 424 425 // Okay, we have something that *can* be converted to the right type. Check 426 // to see if there is a potentially weird extension going on here. This can 427 // happen when you do an atomic operation on something like an char* and 428 // pass in 42. The 42 gets converted to char. This is even more strange 429 // for things like 45.123 -> char, etc. 430 // FIXME: Do this check. 431 ImpCastExprToType(Arg, ValType, Kind, /*isLvalue=*/false); 432 TheCall->setArg(i+1, Arg); 433 } 434 435 // Switch the DeclRefExpr to refer to the new decl. 436 DRE->setDecl(NewBuiltinDecl); 437 DRE->setType(NewBuiltinDecl->getType()); 438 439 // Set the callee in the CallExpr. 440 // FIXME: This leaks the original parens and implicit casts. 441 Expr *PromotedCall = DRE; 442 UsualUnaryConversions(PromotedCall); 443 TheCall->setCallee(PromotedCall); 444 445 446 // Change the result type of the call to match the result type of the decl. 447 TheCall->setType(NewBuiltinDecl->getResultType()); 448 return false; 449} 450 451 452/// CheckObjCString - Checks that the argument to the builtin 453/// CFString constructor is correct 454/// FIXME: GCC currently emits the following warning: 455/// "warning: input conversion stopped due to an input byte that does not 456/// belong to the input codeset UTF-8" 457/// Note: It might also make sense to do the UTF-16 conversion here (would 458/// simplify the backend). 459bool Sema::CheckObjCString(Expr *Arg) { 460 Arg = Arg->IgnoreParenCasts(); 461 StringLiteral *Literal = dyn_cast<StringLiteral>(Arg); 462 463 if (!Literal || Literal->isWide()) { 464 Diag(Arg->getLocStart(), diag::err_cfstring_literal_not_string_constant) 465 << Arg->getSourceRange(); 466 return true; 467 } 468 469 const char *Data = Literal->getStrData(); 470 unsigned Length = Literal->getByteLength(); 471 472 for (unsigned i = 0; i < Length; ++i) { 473 if (!Data[i]) { 474 Diag(getLocationOfStringLiteralByte(Literal, i), 475 diag::warn_cfstring_literal_contains_nul_character) 476 << Arg->getSourceRange(); 477 break; 478 } 479 } 480 481 return false; 482} 483 484/// SemaBuiltinVAStart - Check the arguments to __builtin_va_start for validity. 485/// Emit an error and return true on failure, return false on success. 486bool Sema::SemaBuiltinVAStart(CallExpr *TheCall) { 487 Expr *Fn = TheCall->getCallee(); 488 if (TheCall->getNumArgs() > 2) { 489 Diag(TheCall->getArg(2)->getLocStart(), 490 diag::err_typecheck_call_too_many_args) 491 << 0 /*function call*/ << Fn->getSourceRange() 492 << SourceRange(TheCall->getArg(2)->getLocStart(), 493 (*(TheCall->arg_end()-1))->getLocEnd()); 494 return true; 495 } 496 497 if (TheCall->getNumArgs() < 2) { 498 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 499 << 0 /*function call*/; 500 } 501 502 // Determine whether the current function is variadic or not. 503 bool isVariadic; 504 if (CurBlock) 505 isVariadic = CurBlock->isVariadic; 506 else if (getCurFunctionDecl()) { 507 if (FunctionProtoType* FTP = 508 dyn_cast<FunctionProtoType>(getCurFunctionDecl()->getType())) 509 isVariadic = FTP->isVariadic(); 510 else 511 isVariadic = false; 512 } else { 513 isVariadic = getCurMethodDecl()->isVariadic(); 514 } 515 516 if (!isVariadic) { 517 Diag(Fn->getLocStart(), diag::err_va_start_used_in_non_variadic_function); 518 return true; 519 } 520 521 // Verify that the second argument to the builtin is the last argument of the 522 // current function or method. 523 bool SecondArgIsLastNamedArgument = false; 524 const Expr *Arg = TheCall->getArg(1)->IgnoreParenCasts(); 525 526 if (const DeclRefExpr *DR = dyn_cast<DeclRefExpr>(Arg)) { 527 if (const ParmVarDecl *PV = dyn_cast<ParmVarDecl>(DR->getDecl())) { 528 // FIXME: This isn't correct for methods (results in bogus warning). 529 // Get the last formal in the current function. 530 const ParmVarDecl *LastArg; 531 if (CurBlock) 532 LastArg = *(CurBlock->TheDecl->param_end()-1); 533 else if (FunctionDecl *FD = getCurFunctionDecl()) 534 LastArg = *(FD->param_end()-1); 535 else 536 LastArg = *(getCurMethodDecl()->param_end()-1); 537 SecondArgIsLastNamedArgument = PV == LastArg; 538 } 539 } 540 541 if (!SecondArgIsLastNamedArgument) 542 Diag(TheCall->getArg(1)->getLocStart(), 543 diag::warn_second_parameter_of_va_start_not_last_named_argument); 544 return false; 545} 546 547/// SemaBuiltinUnorderedCompare - Handle functions like __builtin_isgreater and 548/// friends. This is declared to take (...), so we have to check everything. 549bool Sema::SemaBuiltinUnorderedCompare(CallExpr *TheCall) { 550 if (TheCall->getNumArgs() < 2) 551 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 552 << 0 /*function call*/; 553 if (TheCall->getNumArgs() > 2) 554 return Diag(TheCall->getArg(2)->getLocStart(), 555 diag::err_typecheck_call_too_many_args) 556 << 0 /*function call*/ 557 << SourceRange(TheCall->getArg(2)->getLocStart(), 558 (*(TheCall->arg_end()-1))->getLocEnd()); 559 560 Expr *OrigArg0 = TheCall->getArg(0); 561 Expr *OrigArg1 = TheCall->getArg(1); 562 563 // Do standard promotions between the two arguments, returning their common 564 // type. 565 QualType Res = UsualArithmeticConversions(OrigArg0, OrigArg1, false); 566 567 // Make sure any conversions are pushed back into the call; this is 568 // type safe since unordered compare builtins are declared as "_Bool 569 // foo(...)". 570 TheCall->setArg(0, OrigArg0); 571 TheCall->setArg(1, OrigArg1); 572 573 if (OrigArg0->isTypeDependent() || OrigArg1->isTypeDependent()) 574 return false; 575 576 // If the common type isn't a real floating type, then the arguments were 577 // invalid for this operation. 578 if (!Res->isRealFloatingType()) 579 return Diag(OrigArg0->getLocStart(), 580 diag::err_typecheck_call_invalid_ordered_compare) 581 << OrigArg0->getType() << OrigArg1->getType() 582 << SourceRange(OrigArg0->getLocStart(), OrigArg1->getLocEnd()); 583 584 return false; 585} 586 587/// SemaBuiltinUnorderedCompare - Handle functions like __builtin_isnan and 588/// friends. This is declared to take (...), so we have to check everything. 589bool Sema::SemaBuiltinUnaryFP(CallExpr *TheCall) { 590 if (TheCall->getNumArgs() < 1) 591 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 592 << 0 /*function call*/; 593 if (TheCall->getNumArgs() > 1) 594 return Diag(TheCall->getArg(1)->getLocStart(), 595 diag::err_typecheck_call_too_many_args) 596 << 0 /*function call*/ 597 << SourceRange(TheCall->getArg(1)->getLocStart(), 598 (*(TheCall->arg_end()-1))->getLocEnd()); 599 600 Expr *OrigArg = TheCall->getArg(0); 601 602 if (OrigArg->isTypeDependent()) 603 return false; 604 605 // This operation requires a floating-point number 606 if (!OrigArg->getType()->isRealFloatingType()) 607 return Diag(OrigArg->getLocStart(), 608 diag::err_typecheck_call_invalid_unary_fp) 609 << OrigArg->getType() << OrigArg->getSourceRange(); 610 611 return false; 612} 613 614bool Sema::SemaBuiltinStackAddress(CallExpr *TheCall) { 615 // The signature for these builtins is exact; the only thing we need 616 // to check is that the argument is a constant. 617 SourceLocation Loc; 618 if (!TheCall->getArg(0)->isTypeDependent() && 619 !TheCall->getArg(0)->isValueDependent() && 620 !TheCall->getArg(0)->isIntegerConstantExpr(Context, &Loc)) 621 return Diag(Loc, diag::err_stack_const_level) << TheCall->getSourceRange(); 622 623 return false; 624} 625 626/// SemaBuiltinShuffleVector - Handle __builtin_shufflevector. 627// This is declared to take (...), so we have to check everything. 628Action::OwningExprResult Sema::SemaBuiltinShuffleVector(CallExpr *TheCall) { 629 if (TheCall->getNumArgs() < 3) 630 return ExprError(Diag(TheCall->getLocEnd(), 631 diag::err_typecheck_call_too_few_args) 632 << 0 /*function call*/ << TheCall->getSourceRange()); 633 634 unsigned numElements = std::numeric_limits<unsigned>::max(); 635 if (!TheCall->getArg(0)->isTypeDependent() && 636 !TheCall->getArg(1)->isTypeDependent()) { 637 QualType FAType = TheCall->getArg(0)->getType(); 638 QualType SAType = TheCall->getArg(1)->getType(); 639 640 if (!FAType->isVectorType() || !SAType->isVectorType()) { 641 Diag(TheCall->getLocStart(), diag::err_shufflevector_non_vector) 642 << SourceRange(TheCall->getArg(0)->getLocStart(), 643 TheCall->getArg(1)->getLocEnd()); 644 return ExprError(); 645 } 646 647 if (!Context.hasSameUnqualifiedType(FAType, SAType)) { 648 Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector) 649 << SourceRange(TheCall->getArg(0)->getLocStart(), 650 TheCall->getArg(1)->getLocEnd()); 651 return ExprError(); 652 } 653 654 numElements = FAType->getAs<VectorType>()->getNumElements(); 655 if (TheCall->getNumArgs() != numElements+2) { 656 if (TheCall->getNumArgs() < numElements+2) 657 return ExprError(Diag(TheCall->getLocEnd(), 658 diag::err_typecheck_call_too_few_args) 659 << 0 /*function call*/ << TheCall->getSourceRange()); 660 return ExprError(Diag(TheCall->getLocEnd(), 661 diag::err_typecheck_call_too_many_args) 662 << 0 /*function call*/ << TheCall->getSourceRange()); 663 } 664 } 665 666 for (unsigned i = 2; i < TheCall->getNumArgs(); i++) { 667 if (TheCall->getArg(i)->isTypeDependent() || 668 TheCall->getArg(i)->isValueDependent()) 669 continue; 670 671 llvm::APSInt Result(32); 672 if (!TheCall->getArg(i)->isIntegerConstantExpr(Result, Context)) 673 return ExprError(Diag(TheCall->getLocStart(), 674 diag::err_shufflevector_nonconstant_argument) 675 << TheCall->getArg(i)->getSourceRange()); 676 677 if (Result.getActiveBits() > 64 || Result.getZExtValue() >= numElements*2) 678 return ExprError(Diag(TheCall->getLocStart(), 679 diag::err_shufflevector_argument_too_large) 680 << TheCall->getArg(i)->getSourceRange()); 681 } 682 683 llvm::SmallVector<Expr*, 32> exprs; 684 685 for (unsigned i = 0, e = TheCall->getNumArgs(); i != e; i++) { 686 exprs.push_back(TheCall->getArg(i)); 687 TheCall->setArg(i, 0); 688 } 689 690 return Owned(new (Context) ShuffleVectorExpr(Context, exprs.begin(), 691 exprs.size(), exprs[0]->getType(), 692 TheCall->getCallee()->getLocStart(), 693 TheCall->getRParenLoc())); 694} 695 696/// SemaBuiltinPrefetch - Handle __builtin_prefetch. 697// This is declared to take (const void*, ...) and can take two 698// optional constant int args. 699bool Sema::SemaBuiltinPrefetch(CallExpr *TheCall) { 700 unsigned NumArgs = TheCall->getNumArgs(); 701 702 if (NumArgs > 3) 703 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_many_args) 704 << 0 /*function call*/ << TheCall->getSourceRange(); 705 706 // Argument 0 is checked for us and the remaining arguments must be 707 // constant integers. 708 for (unsigned i = 1; i != NumArgs; ++i) { 709 Expr *Arg = TheCall->getArg(i); 710 if (Arg->isTypeDependent()) 711 continue; 712 713 if (!Arg->getType()->isIntegralType()) 714 return Diag(TheCall->getLocStart(), diag::err_prefetch_invalid_arg_type) 715 << Arg->getSourceRange(); 716 717 ImpCastExprToType(Arg, Context.IntTy, CastExpr::CK_IntegralCast); 718 TheCall->setArg(i, Arg); 719 720 if (Arg->isValueDependent()) 721 continue; 722 723 llvm::APSInt Result; 724 if (!Arg->isIntegerConstantExpr(Result, Context)) 725 return Diag(TheCall->getLocStart(), diag::err_prefetch_invalid_arg_ice) 726 << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 727 728 // FIXME: gcc issues a warning and rewrites these to 0. These 729 // seems especially odd for the third argument since the default 730 // is 3. 731 if (i == 1) { 732 if (Result.getLimitedValue() > 1) 733 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 734 << "0" << "1" << Arg->getSourceRange(); 735 } else { 736 if (Result.getLimitedValue() > 3) 737 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 738 << "0" << "3" << Arg->getSourceRange(); 739 } 740 } 741 742 return false; 743} 744 745/// SemaBuiltinEHReturnDataRegNo - Handle __builtin_eh_return_data_regno, the 746/// operand must be an integer constant. 747bool Sema::SemaBuiltinEHReturnDataRegNo(CallExpr *TheCall) { 748 llvm::APSInt Result; 749 if (!TheCall->getArg(0)->isIntegerConstantExpr(Result, Context)) 750 return Diag(TheCall->getLocStart(), diag::err_expr_not_ice) 751 << TheCall->getArg(0)->getSourceRange(); 752 753 return false; 754} 755 756 757/// SemaBuiltinObjectSize - Handle __builtin_object_size(void *ptr, 758/// int type). This simply type checks that type is one of the defined 759/// constants (0-3). 760// For compatability check 0-3, llvm only handles 0 and 2. 761bool Sema::SemaBuiltinObjectSize(CallExpr *TheCall) { 762 Expr *Arg = TheCall->getArg(1); 763 if (Arg->isTypeDependent()) 764 return false; 765 766 QualType ArgType = Arg->getType(); 767 const BuiltinType *BT = ArgType->getAs<BuiltinType>(); 768 llvm::APSInt Result(32); 769 if (!BT || BT->getKind() != BuiltinType::Int) 770 return Diag(TheCall->getLocStart(), diag::err_object_size_invalid_argument) 771 << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 772 773 if (Arg->isValueDependent()) 774 return false; 775 776 if (!Arg->isIntegerConstantExpr(Result, Context)) { 777 return Diag(TheCall->getLocStart(), diag::err_object_size_invalid_argument) 778 << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 779 } 780 781 if (Result.getSExtValue() < 0 || Result.getSExtValue() > 3) { 782 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 783 << "0" << "3" << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 784 } 785 786 return false; 787} 788 789/// SemaBuiltinLongjmp - Handle __builtin_longjmp(void *env[5], int val). 790/// This checks that val is a constant 1. 791bool Sema::SemaBuiltinLongjmp(CallExpr *TheCall) { 792 Expr *Arg = TheCall->getArg(1); 793 if (Arg->isTypeDependent() || Arg->isValueDependent()) 794 return false; 795 796 llvm::APSInt Result(32); 797 if (!Arg->isIntegerConstantExpr(Result, Context) || Result != 1) 798 return Diag(TheCall->getLocStart(), diag::err_builtin_longjmp_invalid_val) 799 << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 800 801 return false; 802} 803 804// Handle i > 1 ? "x" : "y", recursivelly 805bool Sema::SemaCheckStringLiteral(const Expr *E, const CallExpr *TheCall, 806 bool HasVAListArg, 807 unsigned format_idx, unsigned firstDataArg) { 808 if (E->isTypeDependent() || E->isValueDependent()) 809 return false; 810 811 switch (E->getStmtClass()) { 812 case Stmt::ConditionalOperatorClass: { 813 const ConditionalOperator *C = cast<ConditionalOperator>(E); 814 return SemaCheckStringLiteral(C->getTrueExpr(), TheCall, 815 HasVAListArg, format_idx, firstDataArg) 816 && SemaCheckStringLiteral(C->getRHS(), TheCall, 817 HasVAListArg, format_idx, firstDataArg); 818 } 819 820 case Stmt::ImplicitCastExprClass: { 821 const ImplicitCastExpr *Expr = cast<ImplicitCastExpr>(E); 822 return SemaCheckStringLiteral(Expr->getSubExpr(), TheCall, HasVAListArg, 823 format_idx, firstDataArg); 824 } 825 826 case Stmt::ParenExprClass: { 827 const ParenExpr *Expr = cast<ParenExpr>(E); 828 return SemaCheckStringLiteral(Expr->getSubExpr(), TheCall, HasVAListArg, 829 format_idx, firstDataArg); 830 } 831 832 case Stmt::DeclRefExprClass: { 833 const DeclRefExpr *DR = cast<DeclRefExpr>(E); 834 835 // As an exception, do not flag errors for variables binding to 836 // const string literals. 837 if (const VarDecl *VD = dyn_cast<VarDecl>(DR->getDecl())) { 838 bool isConstant = false; 839 QualType T = DR->getType(); 840 841 if (const ArrayType *AT = Context.getAsArrayType(T)) { 842 isConstant = AT->getElementType().isConstant(Context); 843 } else if (const PointerType *PT = T->getAs<PointerType>()) { 844 isConstant = T.isConstant(Context) && 845 PT->getPointeeType().isConstant(Context); 846 } 847 848 if (isConstant) { 849 const VarDecl *Def = 0; 850 if (const Expr *Init = VD->getDefinition(Def)) 851 return SemaCheckStringLiteral(Init, TheCall, 852 HasVAListArg, format_idx, firstDataArg); 853 } 854 855 // For vprintf* functions (i.e., HasVAListArg==true), we add a 856 // special check to see if the format string is a function parameter 857 // of the function calling the printf function. If the function 858 // has an attribute indicating it is a printf-like function, then we 859 // should suppress warnings concerning non-literals being used in a call 860 // to a vprintf function. For example: 861 // 862 // void 863 // logmessage(char const *fmt __attribute__ (format (printf, 1, 2)), ...){ 864 // va_list ap; 865 // va_start(ap, fmt); 866 // vprintf(fmt, ap); // Do NOT emit a warning about "fmt". 867 // ... 868 // 869 // 870 // FIXME: We don't have full attribute support yet, so just check to see 871 // if the argument is a DeclRefExpr that references a parameter. We'll 872 // add proper support for checking the attribute later. 873 if (HasVAListArg) 874 if (isa<ParmVarDecl>(VD)) 875 return true; 876 } 877 878 return false; 879 } 880 881 case Stmt::CallExprClass: { 882 const CallExpr *CE = cast<CallExpr>(E); 883 if (const ImplicitCastExpr *ICE 884 = dyn_cast<ImplicitCastExpr>(CE->getCallee())) { 885 if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(ICE->getSubExpr())) { 886 if (const FunctionDecl *FD = dyn_cast<FunctionDecl>(DRE->getDecl())) { 887 if (const FormatArgAttr *FA = FD->getAttr<FormatArgAttr>()) { 888 unsigned ArgIndex = FA->getFormatIdx(); 889 const Expr *Arg = CE->getArg(ArgIndex - 1); 890 891 return SemaCheckStringLiteral(Arg, TheCall, HasVAListArg, 892 format_idx, firstDataArg); 893 } 894 } 895 } 896 } 897 898 return false; 899 } 900 case Stmt::ObjCStringLiteralClass: 901 case Stmt::StringLiteralClass: { 902 const StringLiteral *StrE = NULL; 903 904 if (const ObjCStringLiteral *ObjCFExpr = dyn_cast<ObjCStringLiteral>(E)) 905 StrE = ObjCFExpr->getString(); 906 else 907 StrE = cast<StringLiteral>(E); 908 909 if (StrE) { 910 CheckPrintfString(StrE, E, TheCall, HasVAListArg, format_idx, 911 firstDataArg); 912 return true; 913 } 914 915 return false; 916 } 917 918 default: 919 return false; 920 } 921} 922 923void 924Sema::CheckNonNullArguments(const NonNullAttr *NonNull, 925 const CallExpr *TheCall) { 926 for (NonNullAttr::iterator i = NonNull->begin(), e = NonNull->end(); 927 i != e; ++i) { 928 const Expr *ArgExpr = TheCall->getArg(*i); 929 if (ArgExpr->isNullPointerConstant(Context, 930 Expr::NPC_ValueDependentIsNotNull)) 931 Diag(TheCall->getCallee()->getLocStart(), diag::warn_null_arg) 932 << ArgExpr->getSourceRange(); 933 } 934} 935 936/// CheckPrintfArguments - Check calls to printf (and similar functions) for 937/// correct use of format strings. 938/// 939/// HasVAListArg - A predicate indicating whether the printf-like 940/// function is passed an explicit va_arg argument (e.g., vprintf) 941/// 942/// format_idx - The index into Args for the format string. 943/// 944/// Improper format strings to functions in the printf family can be 945/// the source of bizarre bugs and very serious security holes. A 946/// good source of information is available in the following paper 947/// (which includes additional references): 948/// 949/// FormatGuard: Automatic Protection From printf Format String 950/// Vulnerabilities, Proceedings of the 10th USENIX Security Symposium, 2001. 951/// 952/// Functionality implemented: 953/// 954/// We can statically check the following properties for string 955/// literal format strings for non v.*printf functions (where the 956/// arguments are passed directly): 957// 958/// (1) Are the number of format conversions equal to the number of 959/// data arguments? 960/// 961/// (2) Does each format conversion correctly match the type of the 962/// corresponding data argument? (TODO) 963/// 964/// Moreover, for all printf functions we can: 965/// 966/// (3) Check for a missing format string (when not caught by type checking). 967/// 968/// (4) Check for no-operation flags; e.g. using "#" with format 969/// conversion 'c' (TODO) 970/// 971/// (5) Check the use of '%n', a major source of security holes. 972/// 973/// (6) Check for malformed format conversions that don't specify anything. 974/// 975/// (7) Check for empty format strings. e.g: printf(""); 976/// 977/// (8) Check that the format string is a wide literal. 978/// 979/// All of these checks can be done by parsing the format string. 980/// 981/// For now, we ONLY do (1), (3), (5), (6), (7), and (8). 982void 983Sema::CheckPrintfArguments(const CallExpr *TheCall, bool HasVAListArg, 984 unsigned format_idx, unsigned firstDataArg) { 985 const Expr *Fn = TheCall->getCallee(); 986 987 // The way the format attribute works in GCC, the implicit this argument 988 // of member functions is counted. However, it doesn't appear in our own 989 // lists, so decrement format_idx in that case. 990 if (isa<CXXMemberCallExpr>(TheCall)) { 991 // Catch a format attribute mistakenly referring to the object argument. 992 if (format_idx == 0) 993 return; 994 --format_idx; 995 if(firstDataArg != 0) 996 --firstDataArg; 997 } 998 999 // CHECK: printf-like function is called with no format string. 1000 if (format_idx >= TheCall->getNumArgs()) { 1001 Diag(TheCall->getRParenLoc(), diag::warn_printf_missing_format_string) 1002 << Fn->getSourceRange(); 1003 return; 1004 } 1005 1006 const Expr *OrigFormatExpr = TheCall->getArg(format_idx)->IgnoreParenCasts(); 1007 1008 // CHECK: format string is not a string literal. 1009 // 1010 // Dynamically generated format strings are difficult to 1011 // automatically vet at compile time. Requiring that format strings 1012 // are string literals: (1) permits the checking of format strings by 1013 // the compiler and thereby (2) can practically remove the source of 1014 // many format string exploits. 1015 1016 // Format string can be either ObjC string (e.g. @"%d") or 1017 // C string (e.g. "%d") 1018 // ObjC string uses the same format specifiers as C string, so we can use 1019 // the same format string checking logic for both ObjC and C strings. 1020 if (SemaCheckStringLiteral(OrigFormatExpr, TheCall, HasVAListArg, format_idx, 1021 firstDataArg)) 1022 return; // Literal format string found, check done! 1023 1024 // If there are no arguments specified, warn with -Wformat-security, otherwise 1025 // warn only with -Wformat-nonliteral. 1026 if (TheCall->getNumArgs() == format_idx+1) 1027 Diag(TheCall->getArg(format_idx)->getLocStart(), 1028 diag::warn_printf_nonliteral_noargs) 1029 << OrigFormatExpr->getSourceRange(); 1030 else 1031 Diag(TheCall->getArg(format_idx)->getLocStart(), 1032 diag::warn_printf_nonliteral) 1033 << OrigFormatExpr->getSourceRange(); 1034} 1035 1036void Sema::CheckPrintfString(const StringLiteral *FExpr, 1037 const Expr *OrigFormatExpr, 1038 const CallExpr *TheCall, bool HasVAListArg, 1039 unsigned format_idx, unsigned firstDataArg) { 1040 1041 static bool UseAlternatePrintfChecking = false; 1042 if (UseAlternatePrintfChecking) { 1043 AlternateCheckPrintfString(FExpr, OrigFormatExpr, TheCall, 1044 HasVAListArg, format_idx, firstDataArg); 1045 return; 1046 } 1047 1048 1049 const ObjCStringLiteral *ObjCFExpr = 1050 dyn_cast<ObjCStringLiteral>(OrigFormatExpr); 1051 1052 // CHECK: is the format string a wide literal? 1053 if (FExpr->isWide()) { 1054 Diag(FExpr->getLocStart(), 1055 diag::warn_printf_format_string_is_wide_literal) 1056 << OrigFormatExpr->getSourceRange(); 1057 return; 1058 } 1059 1060 // Str - The format string. NOTE: this is NOT null-terminated! 1061 const char *Str = FExpr->getStrData(); 1062 1063 // CHECK: empty format string? 1064 unsigned StrLen = FExpr->getByteLength(); 1065 1066 if (StrLen == 0) { 1067 Diag(FExpr->getLocStart(), diag::warn_printf_empty_format_string) 1068 << OrigFormatExpr->getSourceRange(); 1069 return; 1070 } 1071 1072 // We process the format string using a binary state machine. The 1073 // current state is stored in CurrentState. 1074 enum { 1075 state_OrdChr, 1076 state_Conversion 1077 } CurrentState = state_OrdChr; 1078 1079 // numConversions - The number of conversions seen so far. This is 1080 // incremented as we traverse the format string. 1081 unsigned numConversions = 0; 1082 1083 // numDataArgs - The number of data arguments after the format 1084 // string. This can only be determined for non vprintf-like 1085 // functions. For those functions, this value is 1 (the sole 1086 // va_arg argument). 1087 unsigned numDataArgs = TheCall->getNumArgs()-firstDataArg; 1088 1089 // Inspect the format string. 1090 unsigned StrIdx = 0; 1091 1092 // LastConversionIdx - Index within the format string where we last saw 1093 // a '%' character that starts a new format conversion. 1094 unsigned LastConversionIdx = 0; 1095 1096 for (; StrIdx < StrLen; ++StrIdx) { 1097 1098 // Is the number of detected conversion conversions greater than 1099 // the number of matching data arguments? If so, stop. 1100 if (!HasVAListArg && numConversions > numDataArgs) break; 1101 1102 // Handle "\0" 1103 if (Str[StrIdx] == '\0') { 1104 // The string returned by getStrData() is not null-terminated, 1105 // so the presence of a null character is likely an error. 1106 Diag(getLocationOfStringLiteralByte(FExpr, StrIdx), 1107 diag::warn_printf_format_string_contains_null_char) 1108 << OrigFormatExpr->getSourceRange(); 1109 return; 1110 } 1111 1112 // Ordinary characters (not processing a format conversion). 1113 if (CurrentState == state_OrdChr) { 1114 if (Str[StrIdx] == '%') { 1115 CurrentState = state_Conversion; 1116 LastConversionIdx = StrIdx; 1117 } 1118 continue; 1119 } 1120 1121 // Seen '%'. Now processing a format conversion. 1122 switch (Str[StrIdx]) { 1123 // Handle dynamic precision or width specifier. 1124 case '*': { 1125 ++numConversions; 1126 1127 if (!HasVAListArg) { 1128 if (numConversions > numDataArgs) { 1129 SourceLocation Loc = getLocationOfStringLiteralByte(FExpr, StrIdx); 1130 1131 if (Str[StrIdx-1] == '.') 1132 Diag(Loc, diag::warn_printf_asterisk_precision_missing_arg) 1133 << OrigFormatExpr->getSourceRange(); 1134 else 1135 Diag(Loc, diag::warn_printf_asterisk_width_missing_arg) 1136 << OrigFormatExpr->getSourceRange(); 1137 1138 // Don't do any more checking. We'll just emit spurious errors. 1139 return; 1140 } 1141 1142 // Perform type checking on width/precision specifier. 1143 const Expr *E = TheCall->getArg(format_idx+numConversions); 1144 if (const BuiltinType *BT = E->getType()->getAs<BuiltinType>()) 1145 if (BT->getKind() == BuiltinType::Int) 1146 break; 1147 1148 SourceLocation Loc = getLocationOfStringLiteralByte(FExpr, StrIdx); 1149 1150 if (Str[StrIdx-1] == '.') 1151 Diag(Loc, diag::warn_printf_asterisk_precision_wrong_type) 1152 << E->getType() << E->getSourceRange(); 1153 else 1154 Diag(Loc, diag::warn_printf_asterisk_width_wrong_type) 1155 << E->getType() << E->getSourceRange(); 1156 1157 break; 1158 } 1159 } 1160 1161 // Characters which can terminate a format conversion 1162 // (e.g. "%d"). Characters that specify length modifiers or 1163 // other flags are handled by the default case below. 1164 // 1165 // FIXME: additional checks will go into the following cases. 1166 case 'i': 1167 case 'd': 1168 case 'o': 1169 case 'u': 1170 case 'x': 1171 case 'X': 1172 case 'e': 1173 case 'E': 1174 case 'f': 1175 case 'F': 1176 case 'g': 1177 case 'G': 1178 case 'a': 1179 case 'A': 1180 case 'c': 1181 case 's': 1182 case 'p': 1183 ++numConversions; 1184 CurrentState = state_OrdChr; 1185 break; 1186 1187 case 'm': 1188 // FIXME: Warn in situations where this isn't supported! 1189 CurrentState = state_OrdChr; 1190 break; 1191 1192 // CHECK: Are we using "%n"? Issue a warning. 1193 case 'n': { 1194 ++numConversions; 1195 CurrentState = state_OrdChr; 1196 SourceLocation Loc = getLocationOfStringLiteralByte(FExpr, 1197 LastConversionIdx); 1198 1199 Diag(Loc, diag::warn_printf_write_back)<<OrigFormatExpr->getSourceRange(); 1200 break; 1201 } 1202 1203 // Handle "%@" 1204 case '@': 1205 // %@ is allowed in ObjC format strings only. 1206 if (ObjCFExpr != NULL) 1207 CurrentState = state_OrdChr; 1208 else { 1209 // Issue a warning: invalid format conversion. 1210 SourceLocation Loc = 1211 getLocationOfStringLiteralByte(FExpr, LastConversionIdx); 1212 1213 Diag(Loc, diag::warn_printf_invalid_conversion) 1214 << std::string(Str+LastConversionIdx, 1215 Str+std::min(LastConversionIdx+2, StrLen)) 1216 << OrigFormatExpr->getSourceRange(); 1217 } 1218 ++numConversions; 1219 break; 1220 1221 // Handle "%%" 1222 case '%': 1223 // Sanity check: Was the first "%" character the previous one? 1224 // If not, we will assume that we have a malformed format 1225 // conversion, and that the current "%" character is the start 1226 // of a new conversion. 1227 if (StrIdx - LastConversionIdx == 1) 1228 CurrentState = state_OrdChr; 1229 else { 1230 // Issue a warning: invalid format conversion. 1231 SourceLocation Loc = 1232 getLocationOfStringLiteralByte(FExpr, LastConversionIdx); 1233 1234 Diag(Loc, diag::warn_printf_invalid_conversion) 1235 << std::string(Str+LastConversionIdx, Str+StrIdx) 1236 << OrigFormatExpr->getSourceRange(); 1237 1238 // This conversion is broken. Advance to the next format 1239 // conversion. 1240 LastConversionIdx = StrIdx; 1241 ++numConversions; 1242 } 1243 break; 1244 1245 default: 1246 // This case catches all other characters: flags, widths, etc. 1247 // We should eventually process those as well. 1248 break; 1249 } 1250 } 1251 1252 if (CurrentState == state_Conversion) { 1253 // Issue a warning: invalid format conversion. 1254 SourceLocation Loc = 1255 getLocationOfStringLiteralByte(FExpr, LastConversionIdx); 1256 1257 Diag(Loc, diag::warn_printf_invalid_conversion) 1258 << std::string(Str+LastConversionIdx, 1259 Str+std::min(LastConversionIdx+2, StrLen)) 1260 << OrigFormatExpr->getSourceRange(); 1261 return; 1262 } 1263 1264 if (!HasVAListArg) { 1265 // CHECK: Does the number of format conversions exceed the number 1266 // of data arguments? 1267 if (numConversions > numDataArgs) { 1268 SourceLocation Loc = 1269 getLocationOfStringLiteralByte(FExpr, LastConversionIdx); 1270 1271 Diag(Loc, diag::warn_printf_insufficient_data_args) 1272 << OrigFormatExpr->getSourceRange(); 1273 } 1274 // CHECK: Does the number of data arguments exceed the number of 1275 // format conversions in the format string? 1276 else if (numConversions < numDataArgs) 1277 Diag(TheCall->getArg(format_idx+numConversions+1)->getLocStart(), 1278 diag::warn_printf_too_many_data_args) 1279 << OrigFormatExpr->getSourceRange(); 1280 } 1281} 1282 1283 1284namespace { 1285class CheckPrintfHandler : public analyze_printf::FormatStringHandler { 1286 Sema &S; 1287 const StringLiteral *FExpr; 1288 const Expr *OrigFormatExpr; 1289 unsigned NumConversions; 1290 const unsigned NumDataArgs; 1291 const bool IsObjCLiteral; 1292 const char *Beg; // Start of format string. 1293 const bool HasVAListArg; 1294 const CallExpr *TheCall; 1295 unsigned FormatIdx; 1296public: 1297 CheckPrintfHandler(Sema &s, const StringLiteral *fexpr, 1298 const Expr *origFormatExpr, 1299 unsigned numDataArgs, bool isObjCLiteral, 1300 const char *beg, bool hasVAListArg, 1301 const CallExpr *theCall, unsigned formatIdx) 1302 : S(s), FExpr(fexpr), OrigFormatExpr(origFormatExpr), 1303 NumConversions(0), NumDataArgs(numDataArgs), 1304 IsObjCLiteral(isObjCLiteral), Beg(beg), 1305 HasVAListArg(hasVAListArg), 1306 TheCall(theCall), FormatIdx(formatIdx) {} 1307 1308 void HandleNullChar(const char *nullCharacter); 1309 1310 bool HandleFormatSpecifier(const analyze_printf::FormatSpecifier &FS, 1311 const char *startSpecifier, 1312 unsigned specifierLen); 1313private: 1314 SourceRange getFormatRange(); 1315 SourceLocation getLocationOfByte(const char *x); 1316 1317 bool HandleAmount(const analyze_printf::OptionalAmount &Amt, 1318 unsigned MissingArgDiag, unsigned BadTypeDiag); 1319 1320 const Expr *getDataArg(unsigned i) const; 1321}; 1322} 1323 1324SourceRange CheckPrintfHandler::getFormatRange() { 1325 return OrigFormatExpr->getSourceRange(); 1326} 1327 1328SourceLocation CheckPrintfHandler::getLocationOfByte(const char *x) { 1329 return S.getLocationOfStringLiteralByte(FExpr, x - Beg); 1330} 1331 1332void CheckPrintfHandler::HandleNullChar(const char *nullCharacter) { 1333 // The presence of a null character is likely an error. 1334 S.Diag(getLocationOfByte(nullCharacter), 1335 diag::warn_printf_format_string_contains_null_char) 1336 << getFormatRange(); 1337} 1338 1339const Expr *CheckPrintfHandler::getDataArg(unsigned i) const { 1340 return TheCall->getArg(FormatIdx + i); 1341} 1342 1343bool 1344CheckPrintfHandler::HandleAmount(const analyze_printf::OptionalAmount &Amt, 1345 unsigned MissingArgDiag, 1346 unsigned BadTypeDiag) { 1347 1348 if (Amt.hasDataArgument()) { 1349 ++NumConversions; 1350 if (!HasVAListArg) { 1351 if (NumConversions > NumDataArgs) { 1352 S.Diag(getLocationOfByte(Amt.getStart()), MissingArgDiag) 1353 << getFormatRange(); 1354 // Don't do any more checking. We will just emit 1355 // spurious errors. 1356 return false; 1357 } 1358 1359 // Type check the data argument. It should be an 'int'. 1360 const Expr *Arg = getDataArg(NumConversions); 1361 QualType T = Arg->getType(); 1362 const BuiltinType *BT = T->getAs<BuiltinType>(); 1363 if (!BT || BT->getKind() != BuiltinType::Int) { 1364 S.Diag(getLocationOfByte(Amt.getStart()), BadTypeDiag) 1365 << T << getFormatRange() << Arg->getSourceRange(); 1366 // Don't do any more checking. We will just emit 1367 // spurious errors. 1368 return false; 1369 } 1370 } 1371 } 1372 return true; 1373} 1374 1375 1376bool 1377CheckPrintfHandler::HandleFormatSpecifier(const analyze_printf::FormatSpecifier &FS, 1378 const char *startSpecifier, 1379 unsigned specifierLen) { 1380 1381 using namespace analyze_printf; 1382 const ConversionSpecifier &CS = FS.getConversionSpecifier(); 1383 1384 // First check if the field width, precision, and conversion specifier 1385 // have matching data arguments. 1386 if (!HandleAmount(FS.getFieldWidth(), 1387 diag::warn_printf_asterisk_width_missing_arg, 1388 diag::warn_printf_asterisk_width_wrong_type)) { 1389 return false; 1390 } 1391 1392 if (!HandleAmount(FS.getPrecision(), 1393 diag::warn_printf_asterisk_precision_missing_arg, 1394 diag::warn_printf_asterisk_precision_wrong_type)) { 1395 return false; 1396 } 1397 1398 ++NumConversions; 1399 1400 // Check for using an Objective-C specific conversion specifier 1401 // in a non-ObjC literal. 1402 if (!IsObjCLiteral && CS.isObjCArg()) { 1403 SourceLocation Loc = getLocationOfByte(CS.getStart()); 1404 S.Diag(Loc, diag::warn_printf_invalid_conversion) 1405 << llvm::StringRef(startSpecifier, specifierLen) 1406 << getFormatRange(); 1407 1408 // Continue checking the other format specifiers. 1409 return true; 1410 } 1411 1412 // Are we using '%n'? Issue a warning about this being 1413 // a possible security issue. 1414 if (CS.getKind() == ConversionSpecifier::OutIntPtrArg) { 1415 S.Diag(getLocationOfByte(CS.getStart()), diag::warn_printf_write_back) 1416 << getFormatRange(); 1417 // Continue checking the other format specifiers. 1418 return true; 1419 } 1420 1421 1422 // The remaining checks depend on the data arguments. 1423 if (HasVAListArg) 1424 return true; 1425 1426 if (NumConversions > NumDataArgs) { 1427 S.Diag(getLocationOfByte(CS.getStart()), 1428 diag::warn_printf_insufficient_data_args) 1429 << getFormatRange(); 1430 // Don't do any more checking. 1431 return false; 1432 } 1433 1434 return true; 1435} 1436 1437 1438void 1439Sema::AlternateCheckPrintfString(const StringLiteral *FExpr, 1440 const Expr *OrigFormatExpr, 1441 const CallExpr *TheCall, bool HasVAListArg, 1442 unsigned format_idx, unsigned firstDataArg) { 1443 1444 // CHECK: is the format string a wide literal? 1445 if (FExpr->isWide()) { 1446 Diag(FExpr->getLocStart(), 1447 diag::warn_printf_format_string_is_wide_literal) 1448 << OrigFormatExpr->getSourceRange(); 1449 return; 1450 } 1451 1452 // Str - The format string. NOTE: this is NOT null-terminated! 1453 const char *Str = FExpr->getStrData(); 1454 1455 // CHECK: empty format string? 1456 unsigned StrLen = FExpr->getByteLength(); 1457 1458 if (StrLen == 0) { 1459 Diag(FExpr->getLocStart(), diag::warn_printf_empty_format_string) 1460 << OrigFormatExpr->getSourceRange(); 1461 return; 1462 } 1463 1464 CheckPrintfHandler H(*this, FExpr, OrigFormatExpr, 1465 TheCall->getNumArgs() - firstDataArg, 1466 isa<ObjCStringLiteral>(OrigFormatExpr), Str, 1467 HasVAListArg, TheCall, format_idx); 1468 1469 analyze_printf::ParseFormatString(H, Str, Str + StrLen); 1470} 1471 1472//===--- CHECK: Return Address of Stack Variable --------------------------===// 1473 1474static DeclRefExpr* EvalVal(Expr *E); 1475static DeclRefExpr* EvalAddr(Expr* E); 1476 1477/// CheckReturnStackAddr - Check if a return statement returns the address 1478/// of a stack variable. 1479void 1480Sema::CheckReturnStackAddr(Expr *RetValExp, QualType lhsType, 1481 SourceLocation ReturnLoc) { 1482 1483 // Perform checking for returned stack addresses. 1484 if (lhsType->isPointerType() || lhsType->isBlockPointerType()) { 1485 if (DeclRefExpr *DR = EvalAddr(RetValExp)) 1486 Diag(DR->getLocStart(), diag::warn_ret_stack_addr) 1487 << DR->getDecl()->getDeclName() << RetValExp->getSourceRange(); 1488 1489 // Skip over implicit cast expressions when checking for block expressions. 1490 RetValExp = RetValExp->IgnoreParenCasts(); 1491 1492 if (BlockExpr *C = dyn_cast<BlockExpr>(RetValExp)) 1493 if (C->hasBlockDeclRefExprs()) 1494 Diag(C->getLocStart(), diag::err_ret_local_block) 1495 << C->getSourceRange(); 1496 1497 if (AddrLabelExpr *ALE = dyn_cast<AddrLabelExpr>(RetValExp)) 1498 Diag(ALE->getLocStart(), diag::warn_ret_addr_label) 1499 << ALE->getSourceRange(); 1500 1501 } else if (lhsType->isReferenceType()) { 1502 // Perform checking for stack values returned by reference. 1503 // Check for a reference to the stack 1504 if (DeclRefExpr *DR = EvalVal(RetValExp)) 1505 Diag(DR->getLocStart(), diag::warn_ret_stack_ref) 1506 << DR->getDecl()->getDeclName() << RetValExp->getSourceRange(); 1507 } 1508} 1509 1510/// EvalAddr - EvalAddr and EvalVal are mutually recursive functions that 1511/// check if the expression in a return statement evaluates to an address 1512/// to a location on the stack. The recursion is used to traverse the 1513/// AST of the return expression, with recursion backtracking when we 1514/// encounter a subexpression that (1) clearly does not lead to the address 1515/// of a stack variable or (2) is something we cannot determine leads to 1516/// the address of a stack variable based on such local checking. 1517/// 1518/// EvalAddr processes expressions that are pointers that are used as 1519/// references (and not L-values). EvalVal handles all other values. 1520/// At the base case of the recursion is a check for a DeclRefExpr* in 1521/// the refers to a stack variable. 1522/// 1523/// This implementation handles: 1524/// 1525/// * pointer-to-pointer casts 1526/// * implicit conversions from array references to pointers 1527/// * taking the address of fields 1528/// * arbitrary interplay between "&" and "*" operators 1529/// * pointer arithmetic from an address of a stack variable 1530/// * taking the address of an array element where the array is on the stack 1531static DeclRefExpr* EvalAddr(Expr *E) { 1532 // We should only be called for evaluating pointer expressions. 1533 assert((E->getType()->isAnyPointerType() || 1534 E->getType()->isBlockPointerType() || 1535 E->getType()->isObjCQualifiedIdType()) && 1536 "EvalAddr only works on pointers"); 1537 1538 // Our "symbolic interpreter" is just a dispatch off the currently 1539 // viewed AST node. We then recursively traverse the AST by calling 1540 // EvalAddr and EvalVal appropriately. 1541 switch (E->getStmtClass()) { 1542 case Stmt::ParenExprClass: 1543 // Ignore parentheses. 1544 return EvalAddr(cast<ParenExpr>(E)->getSubExpr()); 1545 1546 case Stmt::UnaryOperatorClass: { 1547 // The only unary operator that make sense to handle here 1548 // is AddrOf. All others don't make sense as pointers. 1549 UnaryOperator *U = cast<UnaryOperator>(E); 1550 1551 if (U->getOpcode() == UnaryOperator::AddrOf) 1552 return EvalVal(U->getSubExpr()); 1553 else 1554 return NULL; 1555 } 1556 1557 case Stmt::BinaryOperatorClass: { 1558 // Handle pointer arithmetic. All other binary operators are not valid 1559 // in this context. 1560 BinaryOperator *B = cast<BinaryOperator>(E); 1561 BinaryOperator::Opcode op = B->getOpcode(); 1562 1563 if (op != BinaryOperator::Add && op != BinaryOperator::Sub) 1564 return NULL; 1565 1566 Expr *Base = B->getLHS(); 1567 1568 // Determine which argument is the real pointer base. It could be 1569 // the RHS argument instead of the LHS. 1570 if (!Base->getType()->isPointerType()) Base = B->getRHS(); 1571 1572 assert (Base->getType()->isPointerType()); 1573 return EvalAddr(Base); 1574 } 1575 1576 // For conditional operators we need to see if either the LHS or RHS are 1577 // valid DeclRefExpr*s. If one of them is valid, we return it. 1578 case Stmt::ConditionalOperatorClass: { 1579 ConditionalOperator *C = cast<ConditionalOperator>(E); 1580 1581 // Handle the GNU extension for missing LHS. 1582 if (Expr *lhsExpr = C->getLHS()) 1583 if (DeclRefExpr* LHS = EvalAddr(lhsExpr)) 1584 return LHS; 1585 1586 return EvalAddr(C->getRHS()); 1587 } 1588 1589 // For casts, we need to handle conversions from arrays to 1590 // pointer values, and pointer-to-pointer conversions. 1591 case Stmt::ImplicitCastExprClass: 1592 case Stmt::CStyleCastExprClass: 1593 case Stmt::CXXFunctionalCastExprClass: { 1594 Expr* SubExpr = cast<CastExpr>(E)->getSubExpr(); 1595 QualType T = SubExpr->getType(); 1596 1597 if (SubExpr->getType()->isPointerType() || 1598 SubExpr->getType()->isBlockPointerType() || 1599 SubExpr->getType()->isObjCQualifiedIdType()) 1600 return EvalAddr(SubExpr); 1601 else if (T->isArrayType()) 1602 return EvalVal(SubExpr); 1603 else 1604 return 0; 1605 } 1606 1607 // C++ casts. For dynamic casts, static casts, and const casts, we 1608 // are always converting from a pointer-to-pointer, so we just blow 1609 // through the cast. In the case the dynamic cast doesn't fail (and 1610 // return NULL), we take the conservative route and report cases 1611 // where we return the address of a stack variable. For Reinterpre 1612 // FIXME: The comment about is wrong; we're not always converting 1613 // from pointer to pointer. I'm guessing that this code should also 1614 // handle references to objects. 1615 case Stmt::CXXStaticCastExprClass: 1616 case Stmt::CXXDynamicCastExprClass: 1617 case Stmt::CXXConstCastExprClass: 1618 case Stmt::CXXReinterpretCastExprClass: { 1619 Expr *S = cast<CXXNamedCastExpr>(E)->getSubExpr(); 1620 if (S->getType()->isPointerType() || S->getType()->isBlockPointerType()) 1621 return EvalAddr(S); 1622 else 1623 return NULL; 1624 } 1625 1626 // Everything else: we simply don't reason about them. 1627 default: 1628 return NULL; 1629 } 1630} 1631 1632 1633/// EvalVal - This function is complements EvalAddr in the mutual recursion. 1634/// See the comments for EvalAddr for more details. 1635static DeclRefExpr* EvalVal(Expr *E) { 1636 1637 // We should only be called for evaluating non-pointer expressions, or 1638 // expressions with a pointer type that are not used as references but instead 1639 // are l-values (e.g., DeclRefExpr with a pointer type). 1640 1641 // Our "symbolic interpreter" is just a dispatch off the currently 1642 // viewed AST node. We then recursively traverse the AST by calling 1643 // EvalAddr and EvalVal appropriately. 1644 switch (E->getStmtClass()) { 1645 case Stmt::DeclRefExprClass: { 1646 // DeclRefExpr: the base case. When we hit a DeclRefExpr we are looking 1647 // at code that refers to a variable's name. We check if it has local 1648 // storage within the function, and if so, return the expression. 1649 DeclRefExpr *DR = cast<DeclRefExpr>(E); 1650 1651 if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl())) 1652 if (V->hasLocalStorage() && !V->getType()->isReferenceType()) return DR; 1653 1654 return NULL; 1655 } 1656 1657 case Stmt::ParenExprClass: 1658 // Ignore parentheses. 1659 return EvalVal(cast<ParenExpr>(E)->getSubExpr()); 1660 1661 case Stmt::UnaryOperatorClass: { 1662 // The only unary operator that make sense to handle here 1663 // is Deref. All others don't resolve to a "name." This includes 1664 // handling all sorts of rvalues passed to a unary operator. 1665 UnaryOperator *U = cast<UnaryOperator>(E); 1666 1667 if (U->getOpcode() == UnaryOperator::Deref) 1668 return EvalAddr(U->getSubExpr()); 1669 1670 return NULL; 1671 } 1672 1673 case Stmt::ArraySubscriptExprClass: { 1674 // Array subscripts are potential references to data on the stack. We 1675 // retrieve the DeclRefExpr* for the array variable if it indeed 1676 // has local storage. 1677 return EvalAddr(cast<ArraySubscriptExpr>(E)->getBase()); 1678 } 1679 1680 case Stmt::ConditionalOperatorClass: { 1681 // For conditional operators we need to see if either the LHS or RHS are 1682 // non-NULL DeclRefExpr's. If one is non-NULL, we return it. 1683 ConditionalOperator *C = cast<ConditionalOperator>(E); 1684 1685 // Handle the GNU extension for missing LHS. 1686 if (Expr *lhsExpr = C->getLHS()) 1687 if (DeclRefExpr *LHS = EvalVal(lhsExpr)) 1688 return LHS; 1689 1690 return EvalVal(C->getRHS()); 1691 } 1692 1693 // Accesses to members are potential references to data on the stack. 1694 case Stmt::MemberExprClass: { 1695 MemberExpr *M = cast<MemberExpr>(E); 1696 1697 // Check for indirect access. We only want direct field accesses. 1698 if (!M->isArrow()) 1699 return EvalVal(M->getBase()); 1700 else 1701 return NULL; 1702 } 1703 1704 // Everything else: we simply don't reason about them. 1705 default: 1706 return NULL; 1707 } 1708} 1709 1710//===--- CHECK: Floating-Point comparisons (-Wfloat-equal) ---------------===// 1711 1712/// Check for comparisons of floating point operands using != and ==. 1713/// Issue a warning if these are no self-comparisons, as they are not likely 1714/// to do what the programmer intended. 1715void Sema::CheckFloatComparison(SourceLocation loc, Expr* lex, Expr *rex) { 1716 bool EmitWarning = true; 1717 1718 Expr* LeftExprSansParen = lex->IgnoreParens(); 1719 Expr* RightExprSansParen = rex->IgnoreParens(); 1720 1721 // Special case: check for x == x (which is OK). 1722 // Do not emit warnings for such cases. 1723 if (DeclRefExpr* DRL = dyn_cast<DeclRefExpr>(LeftExprSansParen)) 1724 if (DeclRefExpr* DRR = dyn_cast<DeclRefExpr>(RightExprSansParen)) 1725 if (DRL->getDecl() == DRR->getDecl()) 1726 EmitWarning = false; 1727 1728 1729 // Special case: check for comparisons against literals that can be exactly 1730 // represented by APFloat. In such cases, do not emit a warning. This 1731 // is a heuristic: often comparison against such literals are used to 1732 // detect if a value in a variable has not changed. This clearly can 1733 // lead to false negatives. 1734 if (EmitWarning) { 1735 if (FloatingLiteral* FLL = dyn_cast<FloatingLiteral>(LeftExprSansParen)) { 1736 if (FLL->isExact()) 1737 EmitWarning = false; 1738 } else 1739 if (FloatingLiteral* FLR = dyn_cast<FloatingLiteral>(RightExprSansParen)){ 1740 if (FLR->isExact()) 1741 EmitWarning = false; 1742 } 1743 } 1744 1745 // Check for comparisons with builtin types. 1746 if (EmitWarning) 1747 if (CallExpr* CL = dyn_cast<CallExpr>(LeftExprSansParen)) 1748 if (CL->isBuiltinCall(Context)) 1749 EmitWarning = false; 1750 1751 if (EmitWarning) 1752 if (CallExpr* CR = dyn_cast<CallExpr>(RightExprSansParen)) 1753 if (CR->isBuiltinCall(Context)) 1754 EmitWarning = false; 1755 1756 // Emit the diagnostic. 1757 if (EmitWarning) 1758 Diag(loc, diag::warn_floatingpoint_eq) 1759 << lex->getSourceRange() << rex->getSourceRange(); 1760} 1761 1762//===--- CHECK: Integer mixed-sign comparisons (-Wsign-compare) --------===// 1763//===--- CHECK: Lossy implicit conversions (-Wconversion) --------------===// 1764 1765namespace { 1766 1767/// Structure recording the 'active' range of an integer-valued 1768/// expression. 1769struct IntRange { 1770 /// The number of bits active in the int. 1771 unsigned Width; 1772 1773 /// True if the int is known not to have negative values. 1774 bool NonNegative; 1775 1776 IntRange() {} 1777 IntRange(unsigned Width, bool NonNegative) 1778 : Width(Width), NonNegative(NonNegative) 1779 {} 1780 1781 // Returns the range of the bool type. 1782 static IntRange forBoolType() { 1783 return IntRange(1, true); 1784 } 1785 1786 // Returns the range of an integral type. 1787 static IntRange forType(ASTContext &C, QualType T) { 1788 return forCanonicalType(C, T->getCanonicalTypeInternal().getTypePtr()); 1789 } 1790 1791 // Returns the range of an integeral type based on its canonical 1792 // representation. 1793 static IntRange forCanonicalType(ASTContext &C, const Type *T) { 1794 assert(T->isCanonicalUnqualified()); 1795 1796 if (const VectorType *VT = dyn_cast<VectorType>(T)) 1797 T = VT->getElementType().getTypePtr(); 1798 if (const ComplexType *CT = dyn_cast<ComplexType>(T)) 1799 T = CT->getElementType().getTypePtr(); 1800 if (const EnumType *ET = dyn_cast<EnumType>(T)) 1801 T = ET->getDecl()->getIntegerType().getTypePtr(); 1802 1803 const BuiltinType *BT = cast<BuiltinType>(T); 1804 assert(BT->isInteger()); 1805 1806 return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger()); 1807 } 1808 1809 // Returns the supremum of two ranges: i.e. their conservative merge. 1810 static IntRange join(const IntRange &L, const IntRange &R) { 1811 return IntRange(std::max(L.Width, R.Width), 1812 L.NonNegative && R.NonNegative); 1813 } 1814 1815 // Returns the infinum of two ranges: i.e. their aggressive merge. 1816 static IntRange meet(const IntRange &L, const IntRange &R) { 1817 return IntRange(std::min(L.Width, R.Width), 1818 L.NonNegative || R.NonNegative); 1819 } 1820}; 1821 1822IntRange GetValueRange(ASTContext &C, llvm::APSInt &value, unsigned MaxWidth) { 1823 if (value.isSigned() && value.isNegative()) 1824 return IntRange(value.getMinSignedBits(), false); 1825 1826 if (value.getBitWidth() > MaxWidth) 1827 value.trunc(MaxWidth); 1828 1829 // isNonNegative() just checks the sign bit without considering 1830 // signedness. 1831 return IntRange(value.getActiveBits(), true); 1832} 1833 1834IntRange GetValueRange(ASTContext &C, APValue &result, QualType Ty, 1835 unsigned MaxWidth) { 1836 if (result.isInt()) 1837 return GetValueRange(C, result.getInt(), MaxWidth); 1838 1839 if (result.isVector()) { 1840 IntRange R = GetValueRange(C, result.getVectorElt(0), Ty, MaxWidth); 1841 for (unsigned i = 1, e = result.getVectorLength(); i != e; ++i) { 1842 IntRange El = GetValueRange(C, result.getVectorElt(i), Ty, MaxWidth); 1843 R = IntRange::join(R, El); 1844 } 1845 return R; 1846 } 1847 1848 if (result.isComplexInt()) { 1849 IntRange R = GetValueRange(C, result.getComplexIntReal(), MaxWidth); 1850 IntRange I = GetValueRange(C, result.getComplexIntImag(), MaxWidth); 1851 return IntRange::join(R, I); 1852 } 1853 1854 // This can happen with lossless casts to intptr_t of "based" lvalues. 1855 // Assume it might use arbitrary bits. 1856 // FIXME: The only reason we need to pass the type in here is to get 1857 // the sign right on this one case. It would be nice if APValue 1858 // preserved this. 1859 assert(result.isLValue()); 1860 return IntRange(MaxWidth, Ty->isUnsignedIntegerType()); 1861} 1862 1863/// Pseudo-evaluate the given integer expression, estimating the 1864/// range of values it might take. 1865/// 1866/// \param MaxWidth - the width to which the value will be truncated 1867IntRange GetExprRange(ASTContext &C, Expr *E, unsigned MaxWidth) { 1868 E = E->IgnoreParens(); 1869 1870 // Try a full evaluation first. 1871 Expr::EvalResult result; 1872 if (E->Evaluate(result, C)) 1873 return GetValueRange(C, result.Val, E->getType(), MaxWidth); 1874 1875 // I think we only want to look through implicit casts here; if the 1876 // user has an explicit widening cast, we should treat the value as 1877 // being of the new, wider type. 1878 if (ImplicitCastExpr *CE = dyn_cast<ImplicitCastExpr>(E)) { 1879 if (CE->getCastKind() == CastExpr::CK_NoOp) 1880 return GetExprRange(C, CE->getSubExpr(), MaxWidth); 1881 1882 IntRange OutputTypeRange = IntRange::forType(C, CE->getType()); 1883 1884 bool isIntegerCast = (CE->getCastKind() == CastExpr::CK_IntegralCast); 1885 if (!isIntegerCast && CE->getCastKind() == CastExpr::CK_Unknown) 1886 isIntegerCast = CE->getSubExpr()->getType()->isIntegerType(); 1887 1888 // Assume that non-integer casts can span the full range of the type. 1889 if (!isIntegerCast) 1890 return OutputTypeRange; 1891 1892 IntRange SubRange 1893 = GetExprRange(C, CE->getSubExpr(), 1894 std::min(MaxWidth, OutputTypeRange.Width)); 1895 1896 // Bail out if the subexpr's range is as wide as the cast type. 1897 if (SubRange.Width >= OutputTypeRange.Width) 1898 return OutputTypeRange; 1899 1900 // Otherwise, we take the smaller width, and we're non-negative if 1901 // either the output type or the subexpr is. 1902 return IntRange(SubRange.Width, 1903 SubRange.NonNegative || OutputTypeRange.NonNegative); 1904 } 1905 1906 if (ConditionalOperator *CO = dyn_cast<ConditionalOperator>(E)) { 1907 // If we can fold the condition, just take that operand. 1908 bool CondResult; 1909 if (CO->getCond()->EvaluateAsBooleanCondition(CondResult, C)) 1910 return GetExprRange(C, CondResult ? CO->getTrueExpr() 1911 : CO->getFalseExpr(), 1912 MaxWidth); 1913 1914 // Otherwise, conservatively merge. 1915 IntRange L = GetExprRange(C, CO->getTrueExpr(), MaxWidth); 1916 IntRange R = GetExprRange(C, CO->getFalseExpr(), MaxWidth); 1917 return IntRange::join(L, R); 1918 } 1919 1920 if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) { 1921 switch (BO->getOpcode()) { 1922 1923 // Boolean-valued operations are single-bit and positive. 1924 case BinaryOperator::LAnd: 1925 case BinaryOperator::LOr: 1926 case BinaryOperator::LT: 1927 case BinaryOperator::GT: 1928 case BinaryOperator::LE: 1929 case BinaryOperator::GE: 1930 case BinaryOperator::EQ: 1931 case BinaryOperator::NE: 1932 return IntRange::forBoolType(); 1933 1934 // Operations with opaque sources are black-listed. 1935 case BinaryOperator::PtrMemD: 1936 case BinaryOperator::PtrMemI: 1937 return IntRange::forType(C, E->getType()); 1938 1939 // Bitwise-and uses the *infinum* of the two source ranges. 1940 case BinaryOperator::And: 1941 return IntRange::meet(GetExprRange(C, BO->getLHS(), MaxWidth), 1942 GetExprRange(C, BO->getRHS(), MaxWidth)); 1943 1944 // Left shift gets black-listed based on a judgement call. 1945 case BinaryOperator::Shl: 1946 return IntRange::forType(C, E->getType()); 1947 1948 // Right shift by a constant can narrow its left argument. 1949 case BinaryOperator::Shr: { 1950 IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth); 1951 1952 // If the shift amount is a positive constant, drop the width by 1953 // that much. 1954 llvm::APSInt shift; 1955 if (BO->getRHS()->isIntegerConstantExpr(shift, C) && 1956 shift.isNonNegative()) { 1957 unsigned zext = shift.getZExtValue(); 1958 if (zext >= L.Width) 1959 L.Width = (L.NonNegative ? 0 : 1); 1960 else 1961 L.Width -= zext; 1962 } 1963 1964 return L; 1965 } 1966 1967 // Comma acts as its right operand. 1968 case BinaryOperator::Comma: 1969 return GetExprRange(C, BO->getRHS(), MaxWidth); 1970 1971 // Black-list pointer subtractions. 1972 case BinaryOperator::Sub: 1973 if (BO->getLHS()->getType()->isPointerType()) 1974 return IntRange::forType(C, E->getType()); 1975 // fallthrough 1976 1977 default: 1978 break; 1979 } 1980 1981 // Treat every other operator as if it were closed on the 1982 // narrowest type that encompasses both operands. 1983 IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth); 1984 IntRange R = GetExprRange(C, BO->getRHS(), MaxWidth); 1985 return IntRange::join(L, R); 1986 } 1987 1988 if (UnaryOperator *UO = dyn_cast<UnaryOperator>(E)) { 1989 switch (UO->getOpcode()) { 1990 // Boolean-valued operations are white-listed. 1991 case UnaryOperator::LNot: 1992 return IntRange::forBoolType(); 1993 1994 // Operations with opaque sources are black-listed. 1995 case UnaryOperator::Deref: 1996 case UnaryOperator::AddrOf: // should be impossible 1997 case UnaryOperator::OffsetOf: 1998 return IntRange::forType(C, E->getType()); 1999 2000 default: 2001 return GetExprRange(C, UO->getSubExpr(), MaxWidth); 2002 } 2003 } 2004 2005 FieldDecl *BitField = E->getBitField(); 2006 if (BitField) { 2007 llvm::APSInt BitWidthAP = BitField->getBitWidth()->EvaluateAsInt(C); 2008 unsigned BitWidth = BitWidthAP.getZExtValue(); 2009 2010 return IntRange(BitWidth, BitField->getType()->isUnsignedIntegerType()); 2011 } 2012 2013 return IntRange::forType(C, E->getType()); 2014} 2015 2016/// Checks whether the given value, which currently has the given 2017/// source semantics, has the same value when coerced through the 2018/// target semantics. 2019bool IsSameFloatAfterCast(const llvm::APFloat &value, 2020 const llvm::fltSemantics &Src, 2021 const llvm::fltSemantics &Tgt) { 2022 llvm::APFloat truncated = value; 2023 2024 bool ignored; 2025 truncated.convert(Src, llvm::APFloat::rmNearestTiesToEven, &ignored); 2026 truncated.convert(Tgt, llvm::APFloat::rmNearestTiesToEven, &ignored); 2027 2028 return truncated.bitwiseIsEqual(value); 2029} 2030 2031/// Checks whether the given value, which currently has the given 2032/// source semantics, has the same value when coerced through the 2033/// target semantics. 2034/// 2035/// The value might be a vector of floats (or a complex number). 2036bool IsSameFloatAfterCast(const APValue &value, 2037 const llvm::fltSemantics &Src, 2038 const llvm::fltSemantics &Tgt) { 2039 if (value.isFloat()) 2040 return IsSameFloatAfterCast(value.getFloat(), Src, Tgt); 2041 2042 if (value.isVector()) { 2043 for (unsigned i = 0, e = value.getVectorLength(); i != e; ++i) 2044 if (!IsSameFloatAfterCast(value.getVectorElt(i), Src, Tgt)) 2045 return false; 2046 return true; 2047 } 2048 2049 assert(value.isComplexFloat()); 2050 return (IsSameFloatAfterCast(value.getComplexFloatReal(), Src, Tgt) && 2051 IsSameFloatAfterCast(value.getComplexFloatImag(), Src, Tgt)); 2052} 2053 2054} // end anonymous namespace 2055 2056/// \brief Implements -Wsign-compare. 2057/// 2058/// \param lex the left-hand expression 2059/// \param rex the right-hand expression 2060/// \param OpLoc the location of the joining operator 2061/// \param Equality whether this is an "equality-like" join, which 2062/// suppresses the warning in some cases 2063void Sema::CheckSignCompare(Expr *lex, Expr *rex, SourceLocation OpLoc, 2064 const PartialDiagnostic &PD, bool Equality) { 2065 // Don't warn if we're in an unevaluated context. 2066 if (ExprEvalContexts.back().Context == Unevaluated) 2067 return; 2068 2069 // If either expression is value-dependent, don't warn. We'll get another 2070 // chance at instantiation time. 2071 if (lex->isValueDependent() || rex->isValueDependent()) 2072 return; 2073 2074 QualType lt = lex->getType(), rt = rex->getType(); 2075 2076 // Only warn if both operands are integral. 2077 if (!lt->isIntegerType() || !rt->isIntegerType()) 2078 return; 2079 2080 // In C, the width of a bitfield determines its type, and the 2081 // declared type only contributes the signedness. This duplicates 2082 // the work that will later be done by UsualUnaryConversions. 2083 // Eventually, this check will be reorganized in a way that avoids 2084 // this duplication. 2085 if (!getLangOptions().CPlusPlus) { 2086 QualType tmp; 2087 tmp = Context.isPromotableBitField(lex); 2088 if (!tmp.isNull()) lt = tmp; 2089 tmp = Context.isPromotableBitField(rex); 2090 if (!tmp.isNull()) rt = tmp; 2091 } 2092 2093 // The rule is that the signed operand becomes unsigned, so isolate the 2094 // signed operand. 2095 Expr *signedOperand = lex, *unsignedOperand = rex; 2096 QualType signedType = lt, unsignedType = rt; 2097 if (lt->isSignedIntegerType()) { 2098 if (rt->isSignedIntegerType()) return; 2099 } else { 2100 if (!rt->isSignedIntegerType()) return; 2101 std::swap(signedOperand, unsignedOperand); 2102 std::swap(signedType, unsignedType); 2103 } 2104 2105 unsigned unsignedWidth = Context.getIntWidth(unsignedType); 2106 unsigned signedWidth = Context.getIntWidth(signedType); 2107 2108 // If the unsigned type is strictly smaller than the signed type, 2109 // then (1) the result type will be signed and (2) the unsigned 2110 // value will fit fully within the signed type, and thus the result 2111 // of the comparison will be exact. 2112 if (signedWidth > unsignedWidth) 2113 return; 2114 2115 // Otherwise, calculate the effective ranges. 2116 IntRange signedRange = GetExprRange(Context, signedOperand, signedWidth); 2117 IntRange unsignedRange = GetExprRange(Context, unsignedOperand, unsignedWidth); 2118 2119 // We should never be unable to prove that the unsigned operand is 2120 // non-negative. 2121 assert(unsignedRange.NonNegative && "unsigned range includes negative?"); 2122 2123 // If the signed operand is non-negative, then the signed->unsigned 2124 // conversion won't change it. 2125 if (signedRange.NonNegative) 2126 return; 2127 2128 // For (in)equality comparisons, if the unsigned operand is a 2129 // constant which cannot collide with a overflowed signed operand, 2130 // then reinterpreting the signed operand as unsigned will not 2131 // change the result of the comparison. 2132 if (Equality && unsignedRange.Width < unsignedWidth) 2133 return; 2134 2135 Diag(OpLoc, PD) 2136 << lt << rt << lex->getSourceRange() << rex->getSourceRange(); 2137} 2138 2139/// Diagnose an implicit cast; purely a helper for CheckImplicitConversion. 2140static void DiagnoseImpCast(Sema &S, Expr *E, QualType T, unsigned diag) { 2141 S.Diag(E->getExprLoc(), diag) << E->getType() << T << E->getSourceRange(); 2142} 2143 2144/// Implements -Wconversion. 2145void Sema::CheckImplicitConversion(Expr *E, QualType T) { 2146 // Don't diagnose in unevaluated contexts. 2147 if (ExprEvalContexts.back().Context == Sema::Unevaluated) 2148 return; 2149 2150 // Don't diagnose for value-dependent expressions. 2151 if (E->isValueDependent()) 2152 return; 2153 2154 const Type *Source = Context.getCanonicalType(E->getType()).getTypePtr(); 2155 const Type *Target = Context.getCanonicalType(T).getTypePtr(); 2156 2157 // Never diagnose implicit casts to bool. 2158 if (Target->isSpecificBuiltinType(BuiltinType::Bool)) 2159 return; 2160 2161 // Strip vector types. 2162 if (isa<VectorType>(Source)) { 2163 if (!isa<VectorType>(Target)) 2164 return DiagnoseImpCast(*this, E, T, diag::warn_impcast_vector_scalar); 2165 2166 Source = cast<VectorType>(Source)->getElementType().getTypePtr(); 2167 Target = cast<VectorType>(Target)->getElementType().getTypePtr(); 2168 } 2169 2170 // Strip complex types. 2171 if (isa<ComplexType>(Source)) { 2172 if (!isa<ComplexType>(Target)) 2173 return DiagnoseImpCast(*this, E, T, diag::warn_impcast_complex_scalar); 2174 2175 Source = cast<ComplexType>(Source)->getElementType().getTypePtr(); 2176 Target = cast<ComplexType>(Target)->getElementType().getTypePtr(); 2177 } 2178 2179 const BuiltinType *SourceBT = dyn_cast<BuiltinType>(Source); 2180 const BuiltinType *TargetBT = dyn_cast<BuiltinType>(Target); 2181 2182 // If the source is floating point... 2183 if (SourceBT && SourceBT->isFloatingPoint()) { 2184 // ...and the target is floating point... 2185 if (TargetBT && TargetBT->isFloatingPoint()) { 2186 // ...then warn if we're dropping FP rank. 2187 2188 // Builtin FP kinds are ordered by increasing FP rank. 2189 if (SourceBT->getKind() > TargetBT->getKind()) { 2190 // Don't warn about float constants that are precisely 2191 // representable in the target type. 2192 Expr::EvalResult result; 2193 if (E->Evaluate(result, Context)) { 2194 // Value might be a float, a float vector, or a float complex. 2195 if (IsSameFloatAfterCast(result.Val, 2196 Context.getFloatTypeSemantics(QualType(TargetBT, 0)), 2197 Context.getFloatTypeSemantics(QualType(SourceBT, 0)))) 2198 return; 2199 } 2200 2201 DiagnoseImpCast(*this, E, T, diag::warn_impcast_float_precision); 2202 } 2203 return; 2204 } 2205 2206 // If the target is integral, always warn. 2207 if ((TargetBT && TargetBT->isInteger())) 2208 // TODO: don't warn for integer values? 2209 return DiagnoseImpCast(*this, E, T, diag::warn_impcast_float_integer); 2210 2211 return; 2212 } 2213 2214 if (!Source->isIntegerType() || !Target->isIntegerType()) 2215 return; 2216 2217 IntRange SourceRange = GetExprRange(Context, E, Context.getIntWidth(E->getType())); 2218 IntRange TargetRange = IntRange::forCanonicalType(Context, Target); 2219 2220 // FIXME: also signed<->unsigned? 2221 2222 if (SourceRange.Width > TargetRange.Width) { 2223 // People want to build with -Wshorten-64-to-32 and not -Wconversion 2224 // and by god we'll let them. 2225 if (SourceRange.Width == 64 && TargetRange.Width == 32) 2226 return DiagnoseImpCast(*this, E, T, diag::warn_impcast_integer_64_32); 2227 return DiagnoseImpCast(*this, E, T, diag::warn_impcast_integer_precision); 2228 } 2229 2230 return; 2231} 2232 2233// MarkLive - Mark all the blocks reachable from e as live. Returns the total 2234// number of blocks just marked live. 2235static unsigned MarkLive(CFGBlock *e, llvm::BitVector &live) { 2236 unsigned count = 0; 2237 std::queue<CFGBlock*> workq; 2238 // Prep work queue 2239 live.set(e->getBlockID()); 2240 ++count; 2241 workq.push(e); 2242 // Solve 2243 while (!workq.empty()) { 2244 CFGBlock *item = workq.front(); 2245 workq.pop(); 2246 for (CFGBlock::succ_iterator I=item->succ_begin(), 2247 E=item->succ_end(); 2248 I != E; 2249 ++I) { 2250 if ((*I) && !live[(*I)->getBlockID()]) { 2251 live.set((*I)->getBlockID()); 2252 ++count; 2253 workq.push(*I); 2254 } 2255 } 2256 } 2257 return count; 2258} 2259 2260static SourceLocation GetUnreachableLoc(CFGBlock &b, SourceRange &R1, 2261 SourceRange &R2) { 2262 Stmt *S; 2263 unsigned sn = 0; 2264 R1 = R2 = SourceRange(); 2265 2266 top: 2267 if (sn < b.size()) 2268 S = b[sn].getStmt(); 2269 else if (b.getTerminator()) 2270 S = b.getTerminator(); 2271 else 2272 return SourceLocation(); 2273 2274 switch (S->getStmtClass()) { 2275 case Expr::BinaryOperatorClass: { 2276 BinaryOperator *BO = cast<BinaryOperator>(S); 2277 if (BO->getOpcode() == BinaryOperator::Comma) { 2278 if (sn+1 < b.size()) 2279 return b[sn+1].getStmt()->getLocStart(); 2280 CFGBlock *n = &b; 2281 while (1) { 2282 if (n->getTerminator()) 2283 return n->getTerminator()->getLocStart(); 2284 if (n->succ_size() != 1) 2285 return SourceLocation(); 2286 n = n[0].succ_begin()[0]; 2287 if (n->pred_size() != 1) 2288 return SourceLocation(); 2289 if (!n->empty()) 2290 return n[0][0].getStmt()->getLocStart(); 2291 } 2292 } 2293 R1 = BO->getLHS()->getSourceRange(); 2294 R2 = BO->getRHS()->getSourceRange(); 2295 return BO->getOperatorLoc(); 2296 } 2297 case Expr::UnaryOperatorClass: { 2298 const UnaryOperator *UO = cast<UnaryOperator>(S); 2299 R1 = UO->getSubExpr()->getSourceRange(); 2300 return UO->getOperatorLoc(); 2301 } 2302 case Expr::CompoundAssignOperatorClass: { 2303 const CompoundAssignOperator *CAO = cast<CompoundAssignOperator>(S); 2304 R1 = CAO->getLHS()->getSourceRange(); 2305 R2 = CAO->getRHS()->getSourceRange(); 2306 return CAO->getOperatorLoc(); 2307 } 2308 case Expr::ConditionalOperatorClass: { 2309 const ConditionalOperator *CO = cast<ConditionalOperator>(S); 2310 return CO->getQuestionLoc(); 2311 } 2312 case Expr::MemberExprClass: { 2313 const MemberExpr *ME = cast<MemberExpr>(S); 2314 R1 = ME->getSourceRange(); 2315 return ME->getMemberLoc(); 2316 } 2317 case Expr::ArraySubscriptExprClass: { 2318 const ArraySubscriptExpr *ASE = cast<ArraySubscriptExpr>(S); 2319 R1 = ASE->getLHS()->getSourceRange(); 2320 R2 = ASE->getRHS()->getSourceRange(); 2321 return ASE->getRBracketLoc(); 2322 } 2323 case Expr::CStyleCastExprClass: { 2324 const CStyleCastExpr *CSC = cast<CStyleCastExpr>(S); 2325 R1 = CSC->getSubExpr()->getSourceRange(); 2326 return CSC->getLParenLoc(); 2327 } 2328 case Expr::CXXFunctionalCastExprClass: { 2329 const CXXFunctionalCastExpr *CE = cast <CXXFunctionalCastExpr>(S); 2330 R1 = CE->getSubExpr()->getSourceRange(); 2331 return CE->getTypeBeginLoc(); 2332 } 2333 case Expr::ImplicitCastExprClass: 2334 ++sn; 2335 goto top; 2336 case Stmt::CXXTryStmtClass: { 2337 return cast<CXXTryStmt>(S)->getHandler(0)->getCatchLoc(); 2338 } 2339 default: ; 2340 } 2341 R1 = S->getSourceRange(); 2342 return S->getLocStart(); 2343} 2344 2345static SourceLocation MarkLiveTop(CFGBlock *e, llvm::BitVector &live, 2346 SourceManager &SM) { 2347 std::queue<CFGBlock*> workq; 2348 // Prep work queue 2349 workq.push(e); 2350 SourceRange R1, R2; 2351 SourceLocation top = GetUnreachableLoc(*e, R1, R2); 2352 bool FromMainFile = false; 2353 bool FromSystemHeader = false; 2354 bool TopValid = false; 2355 if (top.isValid()) { 2356 FromMainFile = SM.isFromMainFile(top); 2357 FromSystemHeader = SM.isInSystemHeader(top); 2358 TopValid = true; 2359 } 2360 // Solve 2361 while (!workq.empty()) { 2362 CFGBlock *item = workq.front(); 2363 workq.pop(); 2364 SourceLocation c = GetUnreachableLoc(*item, R1, R2); 2365 if (c.isValid() 2366 && (!TopValid 2367 || (SM.isFromMainFile(c) && !FromMainFile) 2368 || (FromSystemHeader && !SM.isInSystemHeader(c)) 2369 || SM.isBeforeInTranslationUnit(c, top))) { 2370 top = c; 2371 FromMainFile = SM.isFromMainFile(top); 2372 FromSystemHeader = SM.isInSystemHeader(top); 2373 } 2374 live.set(item->getBlockID()); 2375 for (CFGBlock::succ_iterator I=item->succ_begin(), 2376 E=item->succ_end(); 2377 I != E; 2378 ++I) { 2379 if ((*I) && !live[(*I)->getBlockID()]) { 2380 live.set((*I)->getBlockID()); 2381 workq.push(*I); 2382 } 2383 } 2384 } 2385 return top; 2386} 2387 2388static int LineCmp(const void *p1, const void *p2) { 2389 SourceLocation *Line1 = (SourceLocation *)p1; 2390 SourceLocation *Line2 = (SourceLocation *)p2; 2391 return !(*Line1 < *Line2); 2392} 2393 2394namespace { 2395 struct ErrLoc { 2396 SourceLocation Loc; 2397 SourceRange R1; 2398 SourceRange R2; 2399 ErrLoc(SourceLocation l, SourceRange r1, SourceRange r2) 2400 : Loc(l), R1(r1), R2(r2) { } 2401 }; 2402} 2403 2404/// CheckUnreachable - Check for unreachable code. 2405void Sema::CheckUnreachable(AnalysisContext &AC) { 2406 unsigned count; 2407 // We avoid checking when there are errors, as the CFG won't faithfully match 2408 // the user's code. 2409 if (getDiagnostics().hasErrorOccurred()) 2410 return; 2411 if (Diags.getDiagnosticLevel(diag::warn_unreachable) == Diagnostic::Ignored) 2412 return; 2413 2414 CFG *cfg = AC.getCFG(); 2415 if (cfg == 0) 2416 return; 2417 2418 llvm::BitVector live(cfg->getNumBlockIDs()); 2419 // Mark all live things first. 2420 count = MarkLive(&cfg->getEntry(), live); 2421 2422 if (count == cfg->getNumBlockIDs()) 2423 // If there are no dead blocks, we're done. 2424 return; 2425 2426 SourceRange R1, R2; 2427 2428 llvm::SmallVector<ErrLoc, 24> lines; 2429 bool AddEHEdges = AC.getAddEHEdges(); 2430 // First, give warnings for blocks with no predecessors, as they 2431 // can't be part of a loop. 2432 for (CFG::iterator I = cfg->begin(), E = cfg->end(); I != E; ++I) { 2433 CFGBlock &b = **I; 2434 if (!live[b.getBlockID()]) { 2435 if (b.pred_begin() == b.pred_end()) { 2436 if (!AddEHEdges && b.getTerminator() 2437 && isa<CXXTryStmt>(b.getTerminator())) { 2438 // When not adding EH edges from calls, catch clauses 2439 // can otherwise seem dead. Avoid noting them as dead. 2440 count += MarkLive(&b, live); 2441 continue; 2442 } 2443 SourceLocation c = GetUnreachableLoc(b, R1, R2); 2444 if (!c.isValid()) { 2445 // Blocks without a location can't produce a warning, so don't mark 2446 // reachable blocks from here as live. 2447 live.set(b.getBlockID()); 2448 ++count; 2449 continue; 2450 } 2451 lines.push_back(ErrLoc(c, R1, R2)); 2452 // Avoid excessive errors by marking everything reachable from here 2453 count += MarkLive(&b, live); 2454 } 2455 } 2456 } 2457 2458 if (count < cfg->getNumBlockIDs()) { 2459 // And then give warnings for the tops of loops. 2460 for (CFG::iterator I = cfg->begin(), E = cfg->end(); I != E; ++I) { 2461 CFGBlock &b = **I; 2462 if (!live[b.getBlockID()]) 2463 // Avoid excessive errors by marking everything reachable from here 2464 lines.push_back(ErrLoc(MarkLiveTop(&b, live, 2465 Context.getSourceManager()), 2466 SourceRange(), SourceRange())); 2467 } 2468 } 2469 2470 llvm::array_pod_sort(lines.begin(), lines.end(), LineCmp); 2471 for (llvm::SmallVector<ErrLoc, 24>::iterator I = lines.begin(), 2472 E = lines.end(); 2473 I != E; 2474 ++I) 2475 if (I->Loc.isValid()) 2476 Diag(I->Loc, diag::warn_unreachable) << I->R1 << I->R2; 2477} 2478 2479/// CheckFallThrough - Check that we don't fall off the end of a 2480/// Statement that should return a value. 2481/// 2482/// \returns AlwaysFallThrough iff we always fall off the end of the statement, 2483/// MaybeFallThrough iff we might or might not fall off the end, 2484/// NeverFallThroughOrReturn iff we never fall off the end of the statement or 2485/// return. We assume NeverFallThrough iff we never fall off the end of the 2486/// statement but we may return. We assume that functions not marked noreturn 2487/// will return. 2488Sema::ControlFlowKind Sema::CheckFallThrough(AnalysisContext &AC) { 2489 CFG *cfg = AC.getCFG(); 2490 if (cfg == 0) 2491 // FIXME: This should be NeverFallThrough 2492 return NeverFallThroughOrReturn; 2493 2494 // The CFG leaves in dead things, and we don't want the dead code paths to 2495 // confuse us, so we mark all live things first. 2496 std::queue<CFGBlock*> workq; 2497 llvm::BitVector live(cfg->getNumBlockIDs()); 2498 unsigned count = MarkLive(&cfg->getEntry(), live); 2499 2500 bool AddEHEdges = AC.getAddEHEdges(); 2501 if (!AddEHEdges && count != cfg->getNumBlockIDs()) 2502 // When there are things remaining dead, and we didn't add EH edges 2503 // from CallExprs to the catch clauses, we have to go back and 2504 // mark them as live. 2505 for (CFG::iterator I = cfg->begin(), E = cfg->end(); I != E; ++I) { 2506 CFGBlock &b = **I; 2507 if (!live[b.getBlockID()]) { 2508 if (b.pred_begin() == b.pred_end()) { 2509 if (b.getTerminator() && isa<CXXTryStmt>(b.getTerminator())) 2510 // When not adding EH edges from calls, catch clauses 2511 // can otherwise seem dead. Avoid noting them as dead. 2512 count += MarkLive(&b, live); 2513 continue; 2514 } 2515 } 2516 } 2517 2518 // Now we know what is live, we check the live precessors of the exit block 2519 // and look for fall through paths, being careful to ignore normal returns, 2520 // and exceptional paths. 2521 bool HasLiveReturn = false; 2522 bool HasFakeEdge = false; 2523 bool HasPlainEdge = false; 2524 bool HasAbnormalEdge = false; 2525 for (CFGBlock::pred_iterator I=cfg->getExit().pred_begin(), 2526 E = cfg->getExit().pred_end(); 2527 I != E; 2528 ++I) { 2529 CFGBlock& B = **I; 2530 if (!live[B.getBlockID()]) 2531 continue; 2532 if (B.size() == 0) { 2533 if (B.getTerminator() && isa<CXXTryStmt>(B.getTerminator())) { 2534 HasAbnormalEdge = true; 2535 continue; 2536 } 2537 2538 // A labeled empty statement, or the entry block... 2539 HasPlainEdge = true; 2540 continue; 2541 } 2542 Stmt *S = B[B.size()-1]; 2543 if (isa<ReturnStmt>(S)) { 2544 HasLiveReturn = true; 2545 continue; 2546 } 2547 if (isa<ObjCAtThrowStmt>(S)) { 2548 HasFakeEdge = true; 2549 continue; 2550 } 2551 if (isa<CXXThrowExpr>(S)) { 2552 HasFakeEdge = true; 2553 continue; 2554 } 2555 if (const AsmStmt *AS = dyn_cast<AsmStmt>(S)) { 2556 if (AS->isMSAsm()) { 2557 HasFakeEdge = true; 2558 HasLiveReturn = true; 2559 continue; 2560 } 2561 } 2562 if (isa<CXXTryStmt>(S)) { 2563 HasAbnormalEdge = true; 2564 continue; 2565 } 2566 2567 bool NoReturnEdge = false; 2568 if (CallExpr *C = dyn_cast<CallExpr>(S)) { 2569 if (B.succ_begin()[0] != &cfg->getExit()) { 2570 HasAbnormalEdge = true; 2571 continue; 2572 } 2573 Expr *CEE = C->getCallee()->IgnoreParenCasts(); 2574 if (CEE->getType().getNoReturnAttr()) { 2575 NoReturnEdge = true; 2576 HasFakeEdge = true; 2577 } else if (DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(CEE)) { 2578 ValueDecl *VD = DRE->getDecl(); 2579 if (VD->hasAttr<NoReturnAttr>()) { 2580 NoReturnEdge = true; 2581 HasFakeEdge = true; 2582 } 2583 } 2584 } 2585 // FIXME: Add noreturn message sends. 2586 if (NoReturnEdge == false) 2587 HasPlainEdge = true; 2588 } 2589 if (!HasPlainEdge) { 2590 if (HasLiveReturn) 2591 return NeverFallThrough; 2592 return NeverFallThroughOrReturn; 2593 } 2594 if (HasAbnormalEdge || HasFakeEdge || HasLiveReturn) 2595 return MaybeFallThrough; 2596 // This says AlwaysFallThrough for calls to functions that are not marked 2597 // noreturn, that don't return. If people would like this warning to be more 2598 // accurate, such functions should be marked as noreturn. 2599 return AlwaysFallThrough; 2600} 2601 2602/// CheckFallThroughForFunctionDef - Check that we don't fall off the end of a 2603/// function that should return a value. Check that we don't fall off the end 2604/// of a noreturn function. We assume that functions and blocks not marked 2605/// noreturn will return. 2606void Sema::CheckFallThroughForFunctionDef(Decl *D, Stmt *Body, 2607 AnalysisContext &AC) { 2608 // FIXME: Would be nice if we had a better way to control cascading errors, 2609 // but for now, avoid them. The problem is that when Parse sees: 2610 // int foo() { return a; } 2611 // The return is eaten and the Sema code sees just: 2612 // int foo() { } 2613 // which this code would then warn about. 2614 if (getDiagnostics().hasErrorOccurred()) 2615 return; 2616 2617 bool ReturnsVoid = false; 2618 bool HasNoReturn = false; 2619 if (FunctionDecl *FD = dyn_cast<FunctionDecl>(D)) { 2620 // If the result type of the function is a dependent type, we don't know 2621 // whether it will be void or not, so don't 2622 if (FD->getResultType()->isDependentType()) 2623 return; 2624 if (FD->getResultType()->isVoidType()) 2625 ReturnsVoid = true; 2626 if (FD->hasAttr<NoReturnAttr>()) 2627 HasNoReturn = true; 2628 } else if (ObjCMethodDecl *MD = dyn_cast<ObjCMethodDecl>(D)) { 2629 if (MD->getResultType()->isVoidType()) 2630 ReturnsVoid = true; 2631 if (MD->hasAttr<NoReturnAttr>()) 2632 HasNoReturn = true; 2633 } 2634 2635 // Short circuit for compilation speed. 2636 if ((Diags.getDiagnosticLevel(diag::warn_maybe_falloff_nonvoid_function) 2637 == Diagnostic::Ignored || ReturnsVoid) 2638 && (Diags.getDiagnosticLevel(diag::warn_noreturn_function_has_return_expr) 2639 == Diagnostic::Ignored || !HasNoReturn) 2640 && (Diags.getDiagnosticLevel(diag::warn_suggest_noreturn_block) 2641 == Diagnostic::Ignored || !ReturnsVoid)) 2642 return; 2643 // FIXME: Function try block 2644 if (CompoundStmt *Compound = dyn_cast<CompoundStmt>(Body)) { 2645 switch (CheckFallThrough(AC)) { 2646 case MaybeFallThrough: 2647 if (HasNoReturn) 2648 Diag(Compound->getRBracLoc(), diag::warn_falloff_noreturn_function); 2649 else if (!ReturnsVoid) 2650 Diag(Compound->getRBracLoc(),diag::warn_maybe_falloff_nonvoid_function); 2651 break; 2652 case AlwaysFallThrough: 2653 if (HasNoReturn) 2654 Diag(Compound->getRBracLoc(), diag::warn_falloff_noreturn_function); 2655 else if (!ReturnsVoid) 2656 Diag(Compound->getRBracLoc(), diag::warn_falloff_nonvoid_function); 2657 break; 2658 case NeverFallThroughOrReturn: 2659 if (ReturnsVoid && !HasNoReturn) 2660 Diag(Compound->getLBracLoc(), diag::warn_suggest_noreturn_function); 2661 break; 2662 case NeverFallThrough: 2663 break; 2664 } 2665 } 2666} 2667 2668/// CheckFallThroughForBlock - Check that we don't fall off the end of a block 2669/// that should return a value. Check that we don't fall off the end of a 2670/// noreturn block. We assume that functions and blocks not marked noreturn 2671/// will return. 2672void Sema::CheckFallThroughForBlock(QualType BlockTy, Stmt *Body, 2673 AnalysisContext &AC) { 2674 // FIXME: Would be nice if we had a better way to control cascading errors, 2675 // but for now, avoid them. The problem is that when Parse sees: 2676 // int foo() { return a; } 2677 // The return is eaten and the Sema code sees just: 2678 // int foo() { } 2679 // which this code would then warn about. 2680 if (getDiagnostics().hasErrorOccurred()) 2681 return; 2682 bool ReturnsVoid = false; 2683 bool HasNoReturn = false; 2684 if (const FunctionType *FT =BlockTy->getPointeeType()->getAs<FunctionType>()){ 2685 if (FT->getResultType()->isVoidType()) 2686 ReturnsVoid = true; 2687 if (FT->getNoReturnAttr()) 2688 HasNoReturn = true; 2689 } 2690 2691 // Short circuit for compilation speed. 2692 if (ReturnsVoid 2693 && !HasNoReturn 2694 && (Diags.getDiagnosticLevel(diag::warn_suggest_noreturn_block) 2695 == Diagnostic::Ignored || !ReturnsVoid)) 2696 return; 2697 // FIXME: Funtion try block 2698 if (CompoundStmt *Compound = dyn_cast<CompoundStmt>(Body)) { 2699 switch (CheckFallThrough(AC)) { 2700 case MaybeFallThrough: 2701 if (HasNoReturn) 2702 Diag(Compound->getRBracLoc(), diag::err_noreturn_block_has_return_expr); 2703 else if (!ReturnsVoid) 2704 Diag(Compound->getRBracLoc(), diag::err_maybe_falloff_nonvoid_block); 2705 break; 2706 case AlwaysFallThrough: 2707 if (HasNoReturn) 2708 Diag(Compound->getRBracLoc(), diag::err_noreturn_block_has_return_expr); 2709 else if (!ReturnsVoid) 2710 Diag(Compound->getRBracLoc(), diag::err_falloff_nonvoid_block); 2711 break; 2712 case NeverFallThroughOrReturn: 2713 if (ReturnsVoid) 2714 Diag(Compound->getLBracLoc(), diag::warn_suggest_noreturn_block); 2715 break; 2716 case NeverFallThrough: 2717 break; 2718 } 2719 } 2720} 2721 2722/// CheckParmsForFunctionDef - Check that the parameters of the given 2723/// function are appropriate for the definition of a function. This 2724/// takes care of any checks that cannot be performed on the 2725/// declaration itself, e.g., that the types of each of the function 2726/// parameters are complete. 2727bool Sema::CheckParmsForFunctionDef(FunctionDecl *FD) { 2728 bool HasInvalidParm = false; 2729 for (unsigned p = 0, NumParams = FD->getNumParams(); p < NumParams; ++p) { 2730 ParmVarDecl *Param = FD->getParamDecl(p); 2731 2732 // C99 6.7.5.3p4: the parameters in a parameter type list in a 2733 // function declarator that is part of a function definition of 2734 // that function shall not have incomplete type. 2735 // 2736 // This is also C++ [dcl.fct]p6. 2737 if (!Param->isInvalidDecl() && 2738 RequireCompleteType(Param->getLocation(), Param->getType(), 2739 diag::err_typecheck_decl_incomplete_type)) { 2740 Param->setInvalidDecl(); 2741 HasInvalidParm = true; 2742 } 2743 2744 // C99 6.9.1p5: If the declarator includes a parameter type list, the 2745 // declaration of each parameter shall include an identifier. 2746 if (Param->getIdentifier() == 0 && 2747 !Param->isImplicit() && 2748 !getLangOptions().CPlusPlus) 2749 Diag(Param->getLocation(), diag::err_parameter_name_omitted); 2750 } 2751 2752 return HasInvalidParm; 2753} 2754