SemaChecking.cpp revision 31f8e32788adb299acad455363eb7fd244642c82
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 1036namespace { 1037class CheckPrintfHandler : public FormatStringHandler { 1038 Sema &S; 1039 const StringLiteral *FExpr; 1040 const Expr *OrigFormatExpr; 1041 unsigned NumConversions; 1042 const unsigned NumDataArgs; 1043 const bool IsObjCLiteral; 1044 const char *Beg; // Start of format string. 1045 const bool HasVAListArg; 1046 const CallExpr *TheCall; 1047 unsigned FormatIdx; 1048public: 1049 CheckPrintfHandler(Sema &s, const StringLiteral *fexpr, 1050 const Expr *origFormatExpr, 1051 unsigned numDataArgs, bool isObjCLiteral, 1052 const char *beg, bool hasVAListArg, 1053 const CallExpr *theCall, unsigned formatIdx) 1054 : S(s), FExpr(fexpr), OrigFormatExpr(origFormatExpr), 1055 NumConversions(0), NumDataArgs(numDataArgs), 1056 IsObjCLiteral(isObjCLiteral), Beg(beg), 1057 HasVAListArg(hasVAListArg), 1058 TheCall(theCall), FormatIdx(formatIdx) {} 1059 1060 void DoneProcessing(); 1061 1062 void HandleIncompleteFormatSpecifier(const char *startSpecifier, 1063 unsigned specifierLen); 1064 1065 void HandleInvalidConversionSpecifier(const analyze_printf::FormatSpecifier &FS, 1066 const char *startSpecifier, 1067 unsigned specifierLen); 1068 1069 void HandleNullChar(const char *nullCharacter); 1070 1071 bool HandleFormatSpecifier(const analyze_printf::FormatSpecifier &FS, 1072 const char *startSpecifier, 1073 unsigned specifierLen); 1074private: 1075 SourceRange getFormatStringRange(); 1076 SourceRange getFormatSpecifierRange(const char *startSpecifier, 1077 unsigned specifierLen); 1078 SourceLocation getLocationOfByte(const char *x); 1079 1080 bool HandleAmount(const analyze_printf::OptionalAmount &Amt, 1081 unsigned MissingArgDiag, unsigned BadTypeDiag, 1082 const char *startSpecifier, unsigned specifierLen); 1083 1084 const Expr *getDataArg(unsigned i) const; 1085}; 1086} 1087 1088SourceRange CheckPrintfHandler::getFormatStringRange() { 1089 return OrigFormatExpr->getSourceRange(); 1090} 1091 1092SourceRange CheckPrintfHandler:: 1093getFormatSpecifierRange(const char *startSpecifier, unsigned specifierLen) { 1094 return SourceRange(getLocationOfByte(startSpecifier), 1095 getLocationOfByte(startSpecifier+specifierLen-1)); 1096} 1097 1098SourceLocation CheckPrintfHandler::getLocationOfByte(const char *x) { 1099 return S.getLocationOfStringLiteralByte(FExpr, x - Beg); 1100} 1101 1102void CheckPrintfHandler:: 1103HandleIncompleteFormatSpecifier(const char *startSpecifier, 1104 unsigned specifierLen) { 1105 SourceLocation Loc = getLocationOfByte(startSpecifier); 1106 S.Diag(Loc, diag::warn_printf_incomplete_specifier) 1107 << getFormatSpecifierRange(startSpecifier, specifierLen); 1108} 1109 1110void CheckPrintfHandler:: 1111HandleInvalidConversionSpecifier(const analyze_printf::FormatSpecifier &FS, 1112 const char *startSpecifier, 1113 unsigned specifierLen) { 1114 1115 ++NumConversions; 1116 const analyze_printf::ConversionSpecifier &CS = 1117 FS.getConversionSpecifier(); 1118 SourceLocation Loc = getLocationOfByte(CS.getStart()); 1119 S.Diag(Loc, diag::warn_printf_invalid_conversion) 1120 << llvm::StringRef(CS.getStart(), CS.getLength()) 1121 << getFormatSpecifierRange(startSpecifier, specifierLen); 1122} 1123 1124void CheckPrintfHandler::HandleNullChar(const char *nullCharacter) { 1125 // The presence of a null character is likely an error. 1126 S.Diag(getLocationOfByte(nullCharacter), 1127 diag::warn_printf_format_string_contains_null_char) 1128 << getFormatStringRange(); 1129} 1130 1131const Expr *CheckPrintfHandler::getDataArg(unsigned i) const { 1132 return TheCall->getArg(FormatIdx + i); 1133} 1134 1135bool 1136CheckPrintfHandler::HandleAmount(const analyze_printf::OptionalAmount &Amt, 1137 unsigned MissingArgDiag, 1138 unsigned BadTypeDiag, 1139 const char *startSpecifier, 1140 unsigned specifierLen) { 1141 1142 if (Amt.hasDataArgument()) { 1143 ++NumConversions; 1144 if (!HasVAListArg) { 1145 if (NumConversions > NumDataArgs) { 1146 S.Diag(getLocationOfByte(Amt.getStart()), MissingArgDiag) 1147 << getFormatSpecifierRange(startSpecifier, specifierLen); 1148 // Don't do any more checking. We will just emit 1149 // spurious errors. 1150 return false; 1151 } 1152 1153 // Type check the data argument. It should be an 'int'. 1154 // Although not in conformance with C99, we also allow the argument to be 1155 // an 'unsigned int' as that is a reasonably safe case. GCC also 1156 // doesn't emit a warning for that case. 1157 const Expr *Arg = getDataArg(NumConversions); 1158 QualType T = Arg->getType(); 1159 const BuiltinType *BT = T->getAs<BuiltinType>(); 1160 if (!BT || (BT->getKind() != BuiltinType::Int && 1161 BT->getKind() != BuiltinType::UInt)) { 1162 S.Diag(getLocationOfByte(Amt.getStart()), BadTypeDiag) 1163 << T 1164 << getFormatSpecifierRange(startSpecifier, specifierLen) 1165 << Arg->getSourceRange(); 1166 // Don't do any more checking. We will just emit 1167 // spurious errors. 1168 return false; 1169 } 1170 } 1171 } 1172 return true; 1173} 1174 1175bool 1176CheckPrintfHandler::HandleFormatSpecifier(const analyze_printf::FormatSpecifier &FS, 1177 const char *startSpecifier, 1178 unsigned specifierLen) { 1179 1180 using namespace analyze_printf; 1181 const ConversionSpecifier &CS = FS.getConversionSpecifier(); 1182 1183 // First check if the field width, precision, and conversion specifier 1184 // have matching data arguments. 1185 if (!HandleAmount(FS.getFieldWidth(), 1186 diag::warn_printf_asterisk_width_missing_arg, 1187 diag::warn_printf_asterisk_width_wrong_type, 1188 startSpecifier, specifierLen)) { 1189 return false; 1190 } 1191 1192 if (!HandleAmount(FS.getPrecision(), 1193 diag::warn_printf_asterisk_precision_missing_arg, 1194 diag::warn_printf_asterisk_precision_wrong_type, 1195 startSpecifier, specifierLen)) { 1196 return false; 1197 } 1198 1199 // Check for using an Objective-C specific conversion specifier 1200 // in a non-ObjC literal. 1201 if (!IsObjCLiteral && CS.isObjCArg()) { 1202 HandleInvalidConversionSpecifier(FS, startSpecifier, specifierLen); 1203 1204 // Continue checking the other format specifiers. 1205 return true; 1206 } 1207 1208 if (!CS.consumesDataArgument()) { 1209 // FIXME: Technically specifying a precision or field width here 1210 // makes no sense. Worth issuing a warning at some point. 1211 return true; 1212 } 1213 1214 ++NumConversions; 1215 1216 // Are we using '%n'? Issue a warning about this being 1217 // a possible security issue. 1218 if (CS.getKind() == ConversionSpecifier::OutIntPtrArg) { 1219 S.Diag(getLocationOfByte(CS.getStart()), diag::warn_printf_write_back) 1220 << getFormatSpecifierRange(startSpecifier, specifierLen); 1221 // Continue checking the other format specifiers. 1222 return true; 1223 } 1224 1225 1226 // The remaining checks depend on the data arguments. 1227 if (HasVAListArg) 1228 return true; 1229 1230 if (NumConversions > NumDataArgs) { 1231 S.Diag(getLocationOfByte(CS.getStart()), 1232 diag::warn_printf_insufficient_data_args) 1233 << getFormatSpecifierRange(startSpecifier, specifierLen); 1234 // Don't do any more checking. 1235 return false; 1236 } 1237 1238 return true; 1239} 1240 1241void CheckPrintfHandler::DoneProcessing() { 1242 // Does the number of data arguments exceed the number of 1243 // format conversions in the format string? 1244 if (!HasVAListArg && NumConversions < NumDataArgs) 1245 S.Diag(getDataArg(NumConversions+1)->getLocStart(), 1246 diag::warn_printf_too_many_data_args) 1247 << getFormatStringRange(); 1248} 1249 1250void Sema::CheckPrintfString(const StringLiteral *FExpr, 1251 const Expr *OrigFormatExpr, 1252 const CallExpr *TheCall, bool HasVAListArg, 1253 unsigned format_idx, unsigned firstDataArg) { 1254 1255 // CHECK: is the format string a wide literal? 1256 if (FExpr->isWide()) { 1257 Diag(FExpr->getLocStart(), 1258 diag::warn_printf_format_string_is_wide_literal) 1259 << OrigFormatExpr->getSourceRange(); 1260 return; 1261 } 1262 1263 // Str - The format string. NOTE: this is NOT null-terminated! 1264 const char *Str = FExpr->getStrData(); 1265 1266 // CHECK: empty format string? 1267 unsigned StrLen = FExpr->getByteLength(); 1268 1269 if (StrLen == 0) { 1270 Diag(FExpr->getLocStart(), diag::warn_printf_empty_format_string) 1271 << OrigFormatExpr->getSourceRange(); 1272 return; 1273 } 1274 1275 CheckPrintfHandler H(*this, FExpr, OrigFormatExpr, 1276 TheCall->getNumArgs() - firstDataArg, 1277 isa<ObjCStringLiteral>(OrigFormatExpr), Str, 1278 HasVAListArg, TheCall, format_idx); 1279 1280 if (!ParseFormatString(H, Str, Str + StrLen)) 1281 H.DoneProcessing(); 1282} 1283 1284//===--- CHECK: Return Address of Stack Variable --------------------------===// 1285 1286static DeclRefExpr* EvalVal(Expr *E); 1287static DeclRefExpr* EvalAddr(Expr* E); 1288 1289/// CheckReturnStackAddr - Check if a return statement returns the address 1290/// of a stack variable. 1291void 1292Sema::CheckReturnStackAddr(Expr *RetValExp, QualType lhsType, 1293 SourceLocation ReturnLoc) { 1294 1295 // Perform checking for returned stack addresses. 1296 if (lhsType->isPointerType() || lhsType->isBlockPointerType()) { 1297 if (DeclRefExpr *DR = EvalAddr(RetValExp)) 1298 Diag(DR->getLocStart(), diag::warn_ret_stack_addr) 1299 << DR->getDecl()->getDeclName() << RetValExp->getSourceRange(); 1300 1301 // Skip over implicit cast expressions when checking for block expressions. 1302 RetValExp = RetValExp->IgnoreParenCasts(); 1303 1304 if (BlockExpr *C = dyn_cast<BlockExpr>(RetValExp)) 1305 if (C->hasBlockDeclRefExprs()) 1306 Diag(C->getLocStart(), diag::err_ret_local_block) 1307 << C->getSourceRange(); 1308 1309 if (AddrLabelExpr *ALE = dyn_cast<AddrLabelExpr>(RetValExp)) 1310 Diag(ALE->getLocStart(), diag::warn_ret_addr_label) 1311 << ALE->getSourceRange(); 1312 1313 } else if (lhsType->isReferenceType()) { 1314 // Perform checking for stack values returned by reference. 1315 // Check for a reference to the stack 1316 if (DeclRefExpr *DR = EvalVal(RetValExp)) 1317 Diag(DR->getLocStart(), diag::warn_ret_stack_ref) 1318 << DR->getDecl()->getDeclName() << RetValExp->getSourceRange(); 1319 } 1320} 1321 1322/// EvalAddr - EvalAddr and EvalVal are mutually recursive functions that 1323/// check if the expression in a return statement evaluates to an address 1324/// to a location on the stack. The recursion is used to traverse the 1325/// AST of the return expression, with recursion backtracking when we 1326/// encounter a subexpression that (1) clearly does not lead to the address 1327/// of a stack variable or (2) is something we cannot determine leads to 1328/// the address of a stack variable based on such local checking. 1329/// 1330/// EvalAddr processes expressions that are pointers that are used as 1331/// references (and not L-values). EvalVal handles all other values. 1332/// At the base case of the recursion is a check for a DeclRefExpr* in 1333/// the refers to a stack variable. 1334/// 1335/// This implementation handles: 1336/// 1337/// * pointer-to-pointer casts 1338/// * implicit conversions from array references to pointers 1339/// * taking the address of fields 1340/// * arbitrary interplay between "&" and "*" operators 1341/// * pointer arithmetic from an address of a stack variable 1342/// * taking the address of an array element where the array is on the stack 1343static DeclRefExpr* EvalAddr(Expr *E) { 1344 // We should only be called for evaluating pointer expressions. 1345 assert((E->getType()->isAnyPointerType() || 1346 E->getType()->isBlockPointerType() || 1347 E->getType()->isObjCQualifiedIdType()) && 1348 "EvalAddr only works on pointers"); 1349 1350 // Our "symbolic interpreter" is just a dispatch off the currently 1351 // viewed AST node. We then recursively traverse the AST by calling 1352 // EvalAddr and EvalVal appropriately. 1353 switch (E->getStmtClass()) { 1354 case Stmt::ParenExprClass: 1355 // Ignore parentheses. 1356 return EvalAddr(cast<ParenExpr>(E)->getSubExpr()); 1357 1358 case Stmt::UnaryOperatorClass: { 1359 // The only unary operator that make sense to handle here 1360 // is AddrOf. All others don't make sense as pointers. 1361 UnaryOperator *U = cast<UnaryOperator>(E); 1362 1363 if (U->getOpcode() == UnaryOperator::AddrOf) 1364 return EvalVal(U->getSubExpr()); 1365 else 1366 return NULL; 1367 } 1368 1369 case Stmt::BinaryOperatorClass: { 1370 // Handle pointer arithmetic. All other binary operators are not valid 1371 // in this context. 1372 BinaryOperator *B = cast<BinaryOperator>(E); 1373 BinaryOperator::Opcode op = B->getOpcode(); 1374 1375 if (op != BinaryOperator::Add && op != BinaryOperator::Sub) 1376 return NULL; 1377 1378 Expr *Base = B->getLHS(); 1379 1380 // Determine which argument is the real pointer base. It could be 1381 // the RHS argument instead of the LHS. 1382 if (!Base->getType()->isPointerType()) Base = B->getRHS(); 1383 1384 assert (Base->getType()->isPointerType()); 1385 return EvalAddr(Base); 1386 } 1387 1388 // For conditional operators we need to see if either the LHS or RHS are 1389 // valid DeclRefExpr*s. If one of them is valid, we return it. 1390 case Stmt::ConditionalOperatorClass: { 1391 ConditionalOperator *C = cast<ConditionalOperator>(E); 1392 1393 // Handle the GNU extension for missing LHS. 1394 if (Expr *lhsExpr = C->getLHS()) 1395 if (DeclRefExpr* LHS = EvalAddr(lhsExpr)) 1396 return LHS; 1397 1398 return EvalAddr(C->getRHS()); 1399 } 1400 1401 // For casts, we need to handle conversions from arrays to 1402 // pointer values, and pointer-to-pointer conversions. 1403 case Stmt::ImplicitCastExprClass: 1404 case Stmt::CStyleCastExprClass: 1405 case Stmt::CXXFunctionalCastExprClass: { 1406 Expr* SubExpr = cast<CastExpr>(E)->getSubExpr(); 1407 QualType T = SubExpr->getType(); 1408 1409 if (SubExpr->getType()->isPointerType() || 1410 SubExpr->getType()->isBlockPointerType() || 1411 SubExpr->getType()->isObjCQualifiedIdType()) 1412 return EvalAddr(SubExpr); 1413 else if (T->isArrayType()) 1414 return EvalVal(SubExpr); 1415 else 1416 return 0; 1417 } 1418 1419 // C++ casts. For dynamic casts, static casts, and const casts, we 1420 // are always converting from a pointer-to-pointer, so we just blow 1421 // through the cast. In the case the dynamic cast doesn't fail (and 1422 // return NULL), we take the conservative route and report cases 1423 // where we return the address of a stack variable. For Reinterpre 1424 // FIXME: The comment about is wrong; we're not always converting 1425 // from pointer to pointer. I'm guessing that this code should also 1426 // handle references to objects. 1427 case Stmt::CXXStaticCastExprClass: 1428 case Stmt::CXXDynamicCastExprClass: 1429 case Stmt::CXXConstCastExprClass: 1430 case Stmt::CXXReinterpretCastExprClass: { 1431 Expr *S = cast<CXXNamedCastExpr>(E)->getSubExpr(); 1432 if (S->getType()->isPointerType() || S->getType()->isBlockPointerType()) 1433 return EvalAddr(S); 1434 else 1435 return NULL; 1436 } 1437 1438 // Everything else: we simply don't reason about them. 1439 default: 1440 return NULL; 1441 } 1442} 1443 1444 1445/// EvalVal - This function is complements EvalAddr in the mutual recursion. 1446/// See the comments for EvalAddr for more details. 1447static DeclRefExpr* EvalVal(Expr *E) { 1448 1449 // We should only be called for evaluating non-pointer expressions, or 1450 // expressions with a pointer type that are not used as references but instead 1451 // are l-values (e.g., DeclRefExpr with a pointer type). 1452 1453 // Our "symbolic interpreter" is just a dispatch off the currently 1454 // viewed AST node. We then recursively traverse the AST by calling 1455 // EvalAddr and EvalVal appropriately. 1456 switch (E->getStmtClass()) { 1457 case Stmt::DeclRefExprClass: { 1458 // DeclRefExpr: the base case. When we hit a DeclRefExpr we are looking 1459 // at code that refers to a variable's name. We check if it has local 1460 // storage within the function, and if so, return the expression. 1461 DeclRefExpr *DR = cast<DeclRefExpr>(E); 1462 1463 if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl())) 1464 if (V->hasLocalStorage() && !V->getType()->isReferenceType()) return DR; 1465 1466 return NULL; 1467 } 1468 1469 case Stmt::ParenExprClass: 1470 // Ignore parentheses. 1471 return EvalVal(cast<ParenExpr>(E)->getSubExpr()); 1472 1473 case Stmt::UnaryOperatorClass: { 1474 // The only unary operator that make sense to handle here 1475 // is Deref. All others don't resolve to a "name." This includes 1476 // handling all sorts of rvalues passed to a unary operator. 1477 UnaryOperator *U = cast<UnaryOperator>(E); 1478 1479 if (U->getOpcode() == UnaryOperator::Deref) 1480 return EvalAddr(U->getSubExpr()); 1481 1482 return NULL; 1483 } 1484 1485 case Stmt::ArraySubscriptExprClass: { 1486 // Array subscripts are potential references to data on the stack. We 1487 // retrieve the DeclRefExpr* for the array variable if it indeed 1488 // has local storage. 1489 return EvalAddr(cast<ArraySubscriptExpr>(E)->getBase()); 1490 } 1491 1492 case Stmt::ConditionalOperatorClass: { 1493 // For conditional operators we need to see if either the LHS or RHS are 1494 // non-NULL DeclRefExpr's. If one is non-NULL, we return it. 1495 ConditionalOperator *C = cast<ConditionalOperator>(E); 1496 1497 // Handle the GNU extension for missing LHS. 1498 if (Expr *lhsExpr = C->getLHS()) 1499 if (DeclRefExpr *LHS = EvalVal(lhsExpr)) 1500 return LHS; 1501 1502 return EvalVal(C->getRHS()); 1503 } 1504 1505 // Accesses to members are potential references to data on the stack. 1506 case Stmt::MemberExprClass: { 1507 MemberExpr *M = cast<MemberExpr>(E); 1508 1509 // Check for indirect access. We only want direct field accesses. 1510 if (!M->isArrow()) 1511 return EvalVal(M->getBase()); 1512 else 1513 return NULL; 1514 } 1515 1516 // Everything else: we simply don't reason about them. 1517 default: 1518 return NULL; 1519 } 1520} 1521 1522//===--- CHECK: Floating-Point comparisons (-Wfloat-equal) ---------------===// 1523 1524/// Check for comparisons of floating point operands using != and ==. 1525/// Issue a warning if these are no self-comparisons, as they are not likely 1526/// to do what the programmer intended. 1527void Sema::CheckFloatComparison(SourceLocation loc, Expr* lex, Expr *rex) { 1528 bool EmitWarning = true; 1529 1530 Expr* LeftExprSansParen = lex->IgnoreParens(); 1531 Expr* RightExprSansParen = rex->IgnoreParens(); 1532 1533 // Special case: check for x == x (which is OK). 1534 // Do not emit warnings for such cases. 1535 if (DeclRefExpr* DRL = dyn_cast<DeclRefExpr>(LeftExprSansParen)) 1536 if (DeclRefExpr* DRR = dyn_cast<DeclRefExpr>(RightExprSansParen)) 1537 if (DRL->getDecl() == DRR->getDecl()) 1538 EmitWarning = false; 1539 1540 1541 // Special case: check for comparisons against literals that can be exactly 1542 // represented by APFloat. In such cases, do not emit a warning. This 1543 // is a heuristic: often comparison against such literals are used to 1544 // detect if a value in a variable has not changed. This clearly can 1545 // lead to false negatives. 1546 if (EmitWarning) { 1547 if (FloatingLiteral* FLL = dyn_cast<FloatingLiteral>(LeftExprSansParen)) { 1548 if (FLL->isExact()) 1549 EmitWarning = false; 1550 } else 1551 if (FloatingLiteral* FLR = dyn_cast<FloatingLiteral>(RightExprSansParen)){ 1552 if (FLR->isExact()) 1553 EmitWarning = false; 1554 } 1555 } 1556 1557 // Check for comparisons with builtin types. 1558 if (EmitWarning) 1559 if (CallExpr* CL = dyn_cast<CallExpr>(LeftExprSansParen)) 1560 if (CL->isBuiltinCall(Context)) 1561 EmitWarning = false; 1562 1563 if (EmitWarning) 1564 if (CallExpr* CR = dyn_cast<CallExpr>(RightExprSansParen)) 1565 if (CR->isBuiltinCall(Context)) 1566 EmitWarning = false; 1567 1568 // Emit the diagnostic. 1569 if (EmitWarning) 1570 Diag(loc, diag::warn_floatingpoint_eq) 1571 << lex->getSourceRange() << rex->getSourceRange(); 1572} 1573 1574//===--- CHECK: Integer mixed-sign comparisons (-Wsign-compare) --------===// 1575//===--- CHECK: Lossy implicit conversions (-Wconversion) --------------===// 1576 1577namespace { 1578 1579/// Structure recording the 'active' range of an integer-valued 1580/// expression. 1581struct IntRange { 1582 /// The number of bits active in the int. 1583 unsigned Width; 1584 1585 /// True if the int is known not to have negative values. 1586 bool NonNegative; 1587 1588 IntRange() {} 1589 IntRange(unsigned Width, bool NonNegative) 1590 : Width(Width), NonNegative(NonNegative) 1591 {} 1592 1593 // Returns the range of the bool type. 1594 static IntRange forBoolType() { 1595 return IntRange(1, true); 1596 } 1597 1598 // Returns the range of an integral type. 1599 static IntRange forType(ASTContext &C, QualType T) { 1600 return forCanonicalType(C, T->getCanonicalTypeInternal().getTypePtr()); 1601 } 1602 1603 // Returns the range of an integeral type based on its canonical 1604 // representation. 1605 static IntRange forCanonicalType(ASTContext &C, const Type *T) { 1606 assert(T->isCanonicalUnqualified()); 1607 1608 if (const VectorType *VT = dyn_cast<VectorType>(T)) 1609 T = VT->getElementType().getTypePtr(); 1610 if (const ComplexType *CT = dyn_cast<ComplexType>(T)) 1611 T = CT->getElementType().getTypePtr(); 1612 if (const EnumType *ET = dyn_cast<EnumType>(T)) 1613 T = ET->getDecl()->getIntegerType().getTypePtr(); 1614 1615 const BuiltinType *BT = cast<BuiltinType>(T); 1616 assert(BT->isInteger()); 1617 1618 return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger()); 1619 } 1620 1621 // Returns the supremum of two ranges: i.e. their conservative merge. 1622 static IntRange join(const IntRange &L, const IntRange &R) { 1623 return IntRange(std::max(L.Width, R.Width), 1624 L.NonNegative && R.NonNegative); 1625 } 1626 1627 // Returns the infinum of two ranges: i.e. their aggressive merge. 1628 static IntRange meet(const IntRange &L, const IntRange &R) { 1629 return IntRange(std::min(L.Width, R.Width), 1630 L.NonNegative || R.NonNegative); 1631 } 1632}; 1633 1634IntRange GetValueRange(ASTContext &C, llvm::APSInt &value, unsigned MaxWidth) { 1635 if (value.isSigned() && value.isNegative()) 1636 return IntRange(value.getMinSignedBits(), false); 1637 1638 if (value.getBitWidth() > MaxWidth) 1639 value.trunc(MaxWidth); 1640 1641 // isNonNegative() just checks the sign bit without considering 1642 // signedness. 1643 return IntRange(value.getActiveBits(), true); 1644} 1645 1646IntRange GetValueRange(ASTContext &C, APValue &result, QualType Ty, 1647 unsigned MaxWidth) { 1648 if (result.isInt()) 1649 return GetValueRange(C, result.getInt(), MaxWidth); 1650 1651 if (result.isVector()) { 1652 IntRange R = GetValueRange(C, result.getVectorElt(0), Ty, MaxWidth); 1653 for (unsigned i = 1, e = result.getVectorLength(); i != e; ++i) { 1654 IntRange El = GetValueRange(C, result.getVectorElt(i), Ty, MaxWidth); 1655 R = IntRange::join(R, El); 1656 } 1657 return R; 1658 } 1659 1660 if (result.isComplexInt()) { 1661 IntRange R = GetValueRange(C, result.getComplexIntReal(), MaxWidth); 1662 IntRange I = GetValueRange(C, result.getComplexIntImag(), MaxWidth); 1663 return IntRange::join(R, I); 1664 } 1665 1666 // This can happen with lossless casts to intptr_t of "based" lvalues. 1667 // Assume it might use arbitrary bits. 1668 // FIXME: The only reason we need to pass the type in here is to get 1669 // the sign right on this one case. It would be nice if APValue 1670 // preserved this. 1671 assert(result.isLValue()); 1672 return IntRange(MaxWidth, Ty->isUnsignedIntegerType()); 1673} 1674 1675/// Pseudo-evaluate the given integer expression, estimating the 1676/// range of values it might take. 1677/// 1678/// \param MaxWidth - the width to which the value will be truncated 1679IntRange GetExprRange(ASTContext &C, Expr *E, unsigned MaxWidth) { 1680 E = E->IgnoreParens(); 1681 1682 // Try a full evaluation first. 1683 Expr::EvalResult result; 1684 if (E->Evaluate(result, C)) 1685 return GetValueRange(C, result.Val, E->getType(), MaxWidth); 1686 1687 // I think we only want to look through implicit casts here; if the 1688 // user has an explicit widening cast, we should treat the value as 1689 // being of the new, wider type. 1690 if (ImplicitCastExpr *CE = dyn_cast<ImplicitCastExpr>(E)) { 1691 if (CE->getCastKind() == CastExpr::CK_NoOp) 1692 return GetExprRange(C, CE->getSubExpr(), MaxWidth); 1693 1694 IntRange OutputTypeRange = IntRange::forType(C, CE->getType()); 1695 1696 bool isIntegerCast = (CE->getCastKind() == CastExpr::CK_IntegralCast); 1697 if (!isIntegerCast && CE->getCastKind() == CastExpr::CK_Unknown) 1698 isIntegerCast = CE->getSubExpr()->getType()->isIntegerType(); 1699 1700 // Assume that non-integer casts can span the full range of the type. 1701 if (!isIntegerCast) 1702 return OutputTypeRange; 1703 1704 IntRange SubRange 1705 = GetExprRange(C, CE->getSubExpr(), 1706 std::min(MaxWidth, OutputTypeRange.Width)); 1707 1708 // Bail out if the subexpr's range is as wide as the cast type. 1709 if (SubRange.Width >= OutputTypeRange.Width) 1710 return OutputTypeRange; 1711 1712 // Otherwise, we take the smaller width, and we're non-negative if 1713 // either the output type or the subexpr is. 1714 return IntRange(SubRange.Width, 1715 SubRange.NonNegative || OutputTypeRange.NonNegative); 1716 } 1717 1718 if (ConditionalOperator *CO = dyn_cast<ConditionalOperator>(E)) { 1719 // If we can fold the condition, just take that operand. 1720 bool CondResult; 1721 if (CO->getCond()->EvaluateAsBooleanCondition(CondResult, C)) 1722 return GetExprRange(C, CondResult ? CO->getTrueExpr() 1723 : CO->getFalseExpr(), 1724 MaxWidth); 1725 1726 // Otherwise, conservatively merge. 1727 IntRange L = GetExprRange(C, CO->getTrueExpr(), MaxWidth); 1728 IntRange R = GetExprRange(C, CO->getFalseExpr(), MaxWidth); 1729 return IntRange::join(L, R); 1730 } 1731 1732 if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) { 1733 switch (BO->getOpcode()) { 1734 1735 // Boolean-valued operations are single-bit and positive. 1736 case BinaryOperator::LAnd: 1737 case BinaryOperator::LOr: 1738 case BinaryOperator::LT: 1739 case BinaryOperator::GT: 1740 case BinaryOperator::LE: 1741 case BinaryOperator::GE: 1742 case BinaryOperator::EQ: 1743 case BinaryOperator::NE: 1744 return IntRange::forBoolType(); 1745 1746 // Operations with opaque sources are black-listed. 1747 case BinaryOperator::PtrMemD: 1748 case BinaryOperator::PtrMemI: 1749 return IntRange::forType(C, E->getType()); 1750 1751 // Bitwise-and uses the *infinum* of the two source ranges. 1752 case BinaryOperator::And: 1753 return IntRange::meet(GetExprRange(C, BO->getLHS(), MaxWidth), 1754 GetExprRange(C, BO->getRHS(), MaxWidth)); 1755 1756 // Left shift gets black-listed based on a judgement call. 1757 case BinaryOperator::Shl: 1758 return IntRange::forType(C, E->getType()); 1759 1760 // Right shift by a constant can narrow its left argument. 1761 case BinaryOperator::Shr: { 1762 IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth); 1763 1764 // If the shift amount is a positive constant, drop the width by 1765 // that much. 1766 llvm::APSInt shift; 1767 if (BO->getRHS()->isIntegerConstantExpr(shift, C) && 1768 shift.isNonNegative()) { 1769 unsigned zext = shift.getZExtValue(); 1770 if (zext >= L.Width) 1771 L.Width = (L.NonNegative ? 0 : 1); 1772 else 1773 L.Width -= zext; 1774 } 1775 1776 return L; 1777 } 1778 1779 // Comma acts as its right operand. 1780 case BinaryOperator::Comma: 1781 return GetExprRange(C, BO->getRHS(), MaxWidth); 1782 1783 // Black-list pointer subtractions. 1784 case BinaryOperator::Sub: 1785 if (BO->getLHS()->getType()->isPointerType()) 1786 return IntRange::forType(C, E->getType()); 1787 // fallthrough 1788 1789 default: 1790 break; 1791 } 1792 1793 // Treat every other operator as if it were closed on the 1794 // narrowest type that encompasses both operands. 1795 IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth); 1796 IntRange R = GetExprRange(C, BO->getRHS(), MaxWidth); 1797 return IntRange::join(L, R); 1798 } 1799 1800 if (UnaryOperator *UO = dyn_cast<UnaryOperator>(E)) { 1801 switch (UO->getOpcode()) { 1802 // Boolean-valued operations are white-listed. 1803 case UnaryOperator::LNot: 1804 return IntRange::forBoolType(); 1805 1806 // Operations with opaque sources are black-listed. 1807 case UnaryOperator::Deref: 1808 case UnaryOperator::AddrOf: // should be impossible 1809 case UnaryOperator::OffsetOf: 1810 return IntRange::forType(C, E->getType()); 1811 1812 default: 1813 return GetExprRange(C, UO->getSubExpr(), MaxWidth); 1814 } 1815 } 1816 1817 FieldDecl *BitField = E->getBitField(); 1818 if (BitField) { 1819 llvm::APSInt BitWidthAP = BitField->getBitWidth()->EvaluateAsInt(C); 1820 unsigned BitWidth = BitWidthAP.getZExtValue(); 1821 1822 return IntRange(BitWidth, BitField->getType()->isUnsignedIntegerType()); 1823 } 1824 1825 return IntRange::forType(C, E->getType()); 1826} 1827 1828/// Checks whether the given value, which currently has the given 1829/// source semantics, has the same value when coerced through the 1830/// target semantics. 1831bool IsSameFloatAfterCast(const llvm::APFloat &value, 1832 const llvm::fltSemantics &Src, 1833 const llvm::fltSemantics &Tgt) { 1834 llvm::APFloat truncated = value; 1835 1836 bool ignored; 1837 truncated.convert(Src, llvm::APFloat::rmNearestTiesToEven, &ignored); 1838 truncated.convert(Tgt, llvm::APFloat::rmNearestTiesToEven, &ignored); 1839 1840 return truncated.bitwiseIsEqual(value); 1841} 1842 1843/// Checks whether the given value, which currently has the given 1844/// source semantics, has the same value when coerced through the 1845/// target semantics. 1846/// 1847/// The value might be a vector of floats (or a complex number). 1848bool IsSameFloatAfterCast(const APValue &value, 1849 const llvm::fltSemantics &Src, 1850 const llvm::fltSemantics &Tgt) { 1851 if (value.isFloat()) 1852 return IsSameFloatAfterCast(value.getFloat(), Src, Tgt); 1853 1854 if (value.isVector()) { 1855 for (unsigned i = 0, e = value.getVectorLength(); i != e; ++i) 1856 if (!IsSameFloatAfterCast(value.getVectorElt(i), Src, Tgt)) 1857 return false; 1858 return true; 1859 } 1860 1861 assert(value.isComplexFloat()); 1862 return (IsSameFloatAfterCast(value.getComplexFloatReal(), Src, Tgt) && 1863 IsSameFloatAfterCast(value.getComplexFloatImag(), Src, Tgt)); 1864} 1865 1866} // end anonymous namespace 1867 1868/// \brief Implements -Wsign-compare. 1869/// 1870/// \param lex the left-hand expression 1871/// \param rex the right-hand expression 1872/// \param OpLoc the location of the joining operator 1873/// \param Equality whether this is an "equality-like" join, which 1874/// suppresses the warning in some cases 1875void Sema::CheckSignCompare(Expr *lex, Expr *rex, SourceLocation OpLoc, 1876 const PartialDiagnostic &PD, bool Equality) { 1877 // Don't warn if we're in an unevaluated context. 1878 if (ExprEvalContexts.back().Context == Unevaluated) 1879 return; 1880 1881 // If either expression is value-dependent, don't warn. We'll get another 1882 // chance at instantiation time. 1883 if (lex->isValueDependent() || rex->isValueDependent()) 1884 return; 1885 1886 QualType lt = lex->getType(), rt = rex->getType(); 1887 1888 // Only warn if both operands are integral. 1889 if (!lt->isIntegerType() || !rt->isIntegerType()) 1890 return; 1891 1892 // In C, the width of a bitfield determines its type, and the 1893 // declared type only contributes the signedness. This duplicates 1894 // the work that will later be done by UsualUnaryConversions. 1895 // Eventually, this check will be reorganized in a way that avoids 1896 // this duplication. 1897 if (!getLangOptions().CPlusPlus) { 1898 QualType tmp; 1899 tmp = Context.isPromotableBitField(lex); 1900 if (!tmp.isNull()) lt = tmp; 1901 tmp = Context.isPromotableBitField(rex); 1902 if (!tmp.isNull()) rt = tmp; 1903 } 1904 1905 // The rule is that the signed operand becomes unsigned, so isolate the 1906 // signed operand. 1907 Expr *signedOperand = lex, *unsignedOperand = rex; 1908 QualType signedType = lt, unsignedType = rt; 1909 if (lt->isSignedIntegerType()) { 1910 if (rt->isSignedIntegerType()) return; 1911 } else { 1912 if (!rt->isSignedIntegerType()) return; 1913 std::swap(signedOperand, unsignedOperand); 1914 std::swap(signedType, unsignedType); 1915 } 1916 1917 unsigned unsignedWidth = Context.getIntWidth(unsignedType); 1918 unsigned signedWidth = Context.getIntWidth(signedType); 1919 1920 // If the unsigned type is strictly smaller than the signed type, 1921 // then (1) the result type will be signed and (2) the unsigned 1922 // value will fit fully within the signed type, and thus the result 1923 // of the comparison will be exact. 1924 if (signedWidth > unsignedWidth) 1925 return; 1926 1927 // Otherwise, calculate the effective ranges. 1928 IntRange signedRange = GetExprRange(Context, signedOperand, signedWidth); 1929 IntRange unsignedRange = GetExprRange(Context, unsignedOperand, unsignedWidth); 1930 1931 // We should never be unable to prove that the unsigned operand is 1932 // non-negative. 1933 assert(unsignedRange.NonNegative && "unsigned range includes negative?"); 1934 1935 // If the signed operand is non-negative, then the signed->unsigned 1936 // conversion won't change it. 1937 if (signedRange.NonNegative) 1938 return; 1939 1940 // For (in)equality comparisons, if the unsigned operand is a 1941 // constant which cannot collide with a overflowed signed operand, 1942 // then reinterpreting the signed operand as unsigned will not 1943 // change the result of the comparison. 1944 if (Equality && unsignedRange.Width < unsignedWidth) 1945 return; 1946 1947 Diag(OpLoc, PD) 1948 << lt << rt << lex->getSourceRange() << rex->getSourceRange(); 1949} 1950 1951/// Diagnose an implicit cast; purely a helper for CheckImplicitConversion. 1952static void DiagnoseImpCast(Sema &S, Expr *E, QualType T, unsigned diag) { 1953 S.Diag(E->getExprLoc(), diag) << E->getType() << T << E->getSourceRange(); 1954} 1955 1956/// Implements -Wconversion. 1957void Sema::CheckImplicitConversion(Expr *E, QualType T) { 1958 // Don't diagnose in unevaluated contexts. 1959 if (ExprEvalContexts.back().Context == Sema::Unevaluated) 1960 return; 1961 1962 // Don't diagnose for value-dependent expressions. 1963 if (E->isValueDependent()) 1964 return; 1965 1966 const Type *Source = Context.getCanonicalType(E->getType()).getTypePtr(); 1967 const Type *Target = Context.getCanonicalType(T).getTypePtr(); 1968 1969 // Never diagnose implicit casts to bool. 1970 if (Target->isSpecificBuiltinType(BuiltinType::Bool)) 1971 return; 1972 1973 // Strip vector types. 1974 if (isa<VectorType>(Source)) { 1975 if (!isa<VectorType>(Target)) 1976 return DiagnoseImpCast(*this, E, T, diag::warn_impcast_vector_scalar); 1977 1978 Source = cast<VectorType>(Source)->getElementType().getTypePtr(); 1979 Target = cast<VectorType>(Target)->getElementType().getTypePtr(); 1980 } 1981 1982 // Strip complex types. 1983 if (isa<ComplexType>(Source)) { 1984 if (!isa<ComplexType>(Target)) 1985 return DiagnoseImpCast(*this, E, T, diag::warn_impcast_complex_scalar); 1986 1987 Source = cast<ComplexType>(Source)->getElementType().getTypePtr(); 1988 Target = cast<ComplexType>(Target)->getElementType().getTypePtr(); 1989 } 1990 1991 const BuiltinType *SourceBT = dyn_cast<BuiltinType>(Source); 1992 const BuiltinType *TargetBT = dyn_cast<BuiltinType>(Target); 1993 1994 // If the source is floating point... 1995 if (SourceBT && SourceBT->isFloatingPoint()) { 1996 // ...and the target is floating point... 1997 if (TargetBT && TargetBT->isFloatingPoint()) { 1998 // ...then warn if we're dropping FP rank. 1999 2000 // Builtin FP kinds are ordered by increasing FP rank. 2001 if (SourceBT->getKind() > TargetBT->getKind()) { 2002 // Don't warn about float constants that are precisely 2003 // representable in the target type. 2004 Expr::EvalResult result; 2005 if (E->Evaluate(result, Context)) { 2006 // Value might be a float, a float vector, or a float complex. 2007 if (IsSameFloatAfterCast(result.Val, 2008 Context.getFloatTypeSemantics(QualType(TargetBT, 0)), 2009 Context.getFloatTypeSemantics(QualType(SourceBT, 0)))) 2010 return; 2011 } 2012 2013 DiagnoseImpCast(*this, E, T, diag::warn_impcast_float_precision); 2014 } 2015 return; 2016 } 2017 2018 // If the target is integral, always warn. 2019 if ((TargetBT && TargetBT->isInteger())) 2020 // TODO: don't warn for integer values? 2021 return DiagnoseImpCast(*this, E, T, diag::warn_impcast_float_integer); 2022 2023 return; 2024 } 2025 2026 if (!Source->isIntegerType() || !Target->isIntegerType()) 2027 return; 2028 2029 IntRange SourceRange = GetExprRange(Context, E, Context.getIntWidth(E->getType())); 2030 IntRange TargetRange = IntRange::forCanonicalType(Context, Target); 2031 2032 // FIXME: also signed<->unsigned? 2033 2034 if (SourceRange.Width > TargetRange.Width) { 2035 // People want to build with -Wshorten-64-to-32 and not -Wconversion 2036 // and by god we'll let them. 2037 if (SourceRange.Width == 64 && TargetRange.Width == 32) 2038 return DiagnoseImpCast(*this, E, T, diag::warn_impcast_integer_64_32); 2039 return DiagnoseImpCast(*this, E, T, diag::warn_impcast_integer_precision); 2040 } 2041 2042 return; 2043} 2044 2045// MarkLive - Mark all the blocks reachable from e as live. Returns the total 2046// number of blocks just marked live. 2047static unsigned MarkLive(CFGBlock *e, llvm::BitVector &live) { 2048 unsigned count = 0; 2049 std::queue<CFGBlock*> workq; 2050 // Prep work queue 2051 live.set(e->getBlockID()); 2052 ++count; 2053 workq.push(e); 2054 // Solve 2055 while (!workq.empty()) { 2056 CFGBlock *item = workq.front(); 2057 workq.pop(); 2058 for (CFGBlock::succ_iterator I=item->succ_begin(), 2059 E=item->succ_end(); 2060 I != E; 2061 ++I) { 2062 if ((*I) && !live[(*I)->getBlockID()]) { 2063 live.set((*I)->getBlockID()); 2064 ++count; 2065 workq.push(*I); 2066 } 2067 } 2068 } 2069 return count; 2070} 2071 2072static SourceLocation GetUnreachableLoc(CFGBlock &b, SourceRange &R1, 2073 SourceRange &R2) { 2074 Stmt *S; 2075 unsigned sn = 0; 2076 R1 = R2 = SourceRange(); 2077 2078 top: 2079 if (sn < b.size()) 2080 S = b[sn].getStmt(); 2081 else if (b.getTerminator()) 2082 S = b.getTerminator(); 2083 else 2084 return SourceLocation(); 2085 2086 switch (S->getStmtClass()) { 2087 case Expr::BinaryOperatorClass: { 2088 BinaryOperator *BO = cast<BinaryOperator>(S); 2089 if (BO->getOpcode() == BinaryOperator::Comma) { 2090 if (sn+1 < b.size()) 2091 return b[sn+1].getStmt()->getLocStart(); 2092 CFGBlock *n = &b; 2093 while (1) { 2094 if (n->getTerminator()) 2095 return n->getTerminator()->getLocStart(); 2096 if (n->succ_size() != 1) 2097 return SourceLocation(); 2098 n = n[0].succ_begin()[0]; 2099 if (n->pred_size() != 1) 2100 return SourceLocation(); 2101 if (!n->empty()) 2102 return n[0][0].getStmt()->getLocStart(); 2103 } 2104 } 2105 R1 = BO->getLHS()->getSourceRange(); 2106 R2 = BO->getRHS()->getSourceRange(); 2107 return BO->getOperatorLoc(); 2108 } 2109 case Expr::UnaryOperatorClass: { 2110 const UnaryOperator *UO = cast<UnaryOperator>(S); 2111 R1 = UO->getSubExpr()->getSourceRange(); 2112 return UO->getOperatorLoc(); 2113 } 2114 case Expr::CompoundAssignOperatorClass: { 2115 const CompoundAssignOperator *CAO = cast<CompoundAssignOperator>(S); 2116 R1 = CAO->getLHS()->getSourceRange(); 2117 R2 = CAO->getRHS()->getSourceRange(); 2118 return CAO->getOperatorLoc(); 2119 } 2120 case Expr::ConditionalOperatorClass: { 2121 const ConditionalOperator *CO = cast<ConditionalOperator>(S); 2122 return CO->getQuestionLoc(); 2123 } 2124 case Expr::MemberExprClass: { 2125 const MemberExpr *ME = cast<MemberExpr>(S); 2126 R1 = ME->getSourceRange(); 2127 return ME->getMemberLoc(); 2128 } 2129 case Expr::ArraySubscriptExprClass: { 2130 const ArraySubscriptExpr *ASE = cast<ArraySubscriptExpr>(S); 2131 R1 = ASE->getLHS()->getSourceRange(); 2132 R2 = ASE->getRHS()->getSourceRange(); 2133 return ASE->getRBracketLoc(); 2134 } 2135 case Expr::CStyleCastExprClass: { 2136 const CStyleCastExpr *CSC = cast<CStyleCastExpr>(S); 2137 R1 = CSC->getSubExpr()->getSourceRange(); 2138 return CSC->getLParenLoc(); 2139 } 2140 case Expr::CXXFunctionalCastExprClass: { 2141 const CXXFunctionalCastExpr *CE = cast <CXXFunctionalCastExpr>(S); 2142 R1 = CE->getSubExpr()->getSourceRange(); 2143 return CE->getTypeBeginLoc(); 2144 } 2145 case Expr::ImplicitCastExprClass: 2146 ++sn; 2147 goto top; 2148 case Stmt::CXXTryStmtClass: { 2149 return cast<CXXTryStmt>(S)->getHandler(0)->getCatchLoc(); 2150 } 2151 default: ; 2152 } 2153 R1 = S->getSourceRange(); 2154 return S->getLocStart(); 2155} 2156 2157static SourceLocation MarkLiveTop(CFGBlock *e, llvm::BitVector &live, 2158 SourceManager &SM) { 2159 std::queue<CFGBlock*> workq; 2160 // Prep work queue 2161 workq.push(e); 2162 SourceRange R1, R2; 2163 SourceLocation top = GetUnreachableLoc(*e, R1, R2); 2164 bool FromMainFile = false; 2165 bool FromSystemHeader = false; 2166 bool TopValid = false; 2167 if (top.isValid()) { 2168 FromMainFile = SM.isFromMainFile(top); 2169 FromSystemHeader = SM.isInSystemHeader(top); 2170 TopValid = true; 2171 } 2172 // Solve 2173 while (!workq.empty()) { 2174 CFGBlock *item = workq.front(); 2175 workq.pop(); 2176 SourceLocation c = GetUnreachableLoc(*item, R1, R2); 2177 if (c.isValid() 2178 && (!TopValid 2179 || (SM.isFromMainFile(c) && !FromMainFile) 2180 || (FromSystemHeader && !SM.isInSystemHeader(c)) 2181 || SM.isBeforeInTranslationUnit(c, top))) { 2182 top = c; 2183 FromMainFile = SM.isFromMainFile(top); 2184 FromSystemHeader = SM.isInSystemHeader(top); 2185 } 2186 live.set(item->getBlockID()); 2187 for (CFGBlock::succ_iterator I=item->succ_begin(), 2188 E=item->succ_end(); 2189 I != E; 2190 ++I) { 2191 if ((*I) && !live[(*I)->getBlockID()]) { 2192 live.set((*I)->getBlockID()); 2193 workq.push(*I); 2194 } 2195 } 2196 } 2197 return top; 2198} 2199 2200static int LineCmp(const void *p1, const void *p2) { 2201 SourceLocation *Line1 = (SourceLocation *)p1; 2202 SourceLocation *Line2 = (SourceLocation *)p2; 2203 return !(*Line1 < *Line2); 2204} 2205 2206namespace { 2207 struct ErrLoc { 2208 SourceLocation Loc; 2209 SourceRange R1; 2210 SourceRange R2; 2211 ErrLoc(SourceLocation l, SourceRange r1, SourceRange r2) 2212 : Loc(l), R1(r1), R2(r2) { } 2213 }; 2214} 2215 2216/// CheckUnreachable - Check for unreachable code. 2217void Sema::CheckUnreachable(AnalysisContext &AC) { 2218 unsigned count; 2219 // We avoid checking when there are errors, as the CFG won't faithfully match 2220 // the user's code. 2221 if (getDiagnostics().hasErrorOccurred()) 2222 return; 2223 if (Diags.getDiagnosticLevel(diag::warn_unreachable) == Diagnostic::Ignored) 2224 return; 2225 2226 CFG *cfg = AC.getCFG(); 2227 if (cfg == 0) 2228 return; 2229 2230 llvm::BitVector live(cfg->getNumBlockIDs()); 2231 // Mark all live things first. 2232 count = MarkLive(&cfg->getEntry(), live); 2233 2234 if (count == cfg->getNumBlockIDs()) 2235 // If there are no dead blocks, we're done. 2236 return; 2237 2238 SourceRange R1, R2; 2239 2240 llvm::SmallVector<ErrLoc, 24> lines; 2241 bool AddEHEdges = AC.getAddEHEdges(); 2242 // First, give warnings for blocks with no predecessors, as they 2243 // can't be part of a loop. 2244 for (CFG::iterator I = cfg->begin(), E = cfg->end(); I != E; ++I) { 2245 CFGBlock &b = **I; 2246 if (!live[b.getBlockID()]) { 2247 if (b.pred_begin() == b.pred_end()) { 2248 if (!AddEHEdges && b.getTerminator() 2249 && isa<CXXTryStmt>(b.getTerminator())) { 2250 // When not adding EH edges from calls, catch clauses 2251 // can otherwise seem dead. Avoid noting them as dead. 2252 count += MarkLive(&b, live); 2253 continue; 2254 } 2255 SourceLocation c = GetUnreachableLoc(b, R1, R2); 2256 if (!c.isValid()) { 2257 // Blocks without a location can't produce a warning, so don't mark 2258 // reachable blocks from here as live. 2259 live.set(b.getBlockID()); 2260 ++count; 2261 continue; 2262 } 2263 lines.push_back(ErrLoc(c, R1, R2)); 2264 // Avoid excessive errors by marking everything reachable from here 2265 count += MarkLive(&b, live); 2266 } 2267 } 2268 } 2269 2270 if (count < cfg->getNumBlockIDs()) { 2271 // And then give warnings for the tops of loops. 2272 for (CFG::iterator I = cfg->begin(), E = cfg->end(); I != E; ++I) { 2273 CFGBlock &b = **I; 2274 if (!live[b.getBlockID()]) 2275 // Avoid excessive errors by marking everything reachable from here 2276 lines.push_back(ErrLoc(MarkLiveTop(&b, live, 2277 Context.getSourceManager()), 2278 SourceRange(), SourceRange())); 2279 } 2280 } 2281 2282 llvm::array_pod_sort(lines.begin(), lines.end(), LineCmp); 2283 for (llvm::SmallVector<ErrLoc, 24>::iterator I = lines.begin(), 2284 E = lines.end(); 2285 I != E; 2286 ++I) 2287 if (I->Loc.isValid()) 2288 Diag(I->Loc, diag::warn_unreachable) << I->R1 << I->R2; 2289} 2290 2291/// CheckFallThrough - Check that we don't fall off the end of a 2292/// Statement that should return a value. 2293/// 2294/// \returns AlwaysFallThrough iff we always fall off the end of the statement, 2295/// MaybeFallThrough iff we might or might not fall off the end, 2296/// NeverFallThroughOrReturn iff we never fall off the end of the statement or 2297/// return. We assume NeverFallThrough iff we never fall off the end of the 2298/// statement but we may return. We assume that functions not marked noreturn 2299/// will return. 2300Sema::ControlFlowKind Sema::CheckFallThrough(AnalysisContext &AC) { 2301 CFG *cfg = AC.getCFG(); 2302 if (cfg == 0) 2303 // FIXME: This should be NeverFallThrough 2304 return NeverFallThroughOrReturn; 2305 2306 // The CFG leaves in dead things, and we don't want the dead code paths to 2307 // confuse us, so we mark all live things first. 2308 std::queue<CFGBlock*> workq; 2309 llvm::BitVector live(cfg->getNumBlockIDs()); 2310 unsigned count = MarkLive(&cfg->getEntry(), live); 2311 2312 bool AddEHEdges = AC.getAddEHEdges(); 2313 if (!AddEHEdges && count != cfg->getNumBlockIDs()) 2314 // When there are things remaining dead, and we didn't add EH edges 2315 // from CallExprs to the catch clauses, we have to go back and 2316 // mark them as live. 2317 for (CFG::iterator I = cfg->begin(), E = cfg->end(); I != E; ++I) { 2318 CFGBlock &b = **I; 2319 if (!live[b.getBlockID()]) { 2320 if (b.pred_begin() == b.pred_end()) { 2321 if (b.getTerminator() && isa<CXXTryStmt>(b.getTerminator())) 2322 // When not adding EH edges from calls, catch clauses 2323 // can otherwise seem dead. Avoid noting them as dead. 2324 count += MarkLive(&b, live); 2325 continue; 2326 } 2327 } 2328 } 2329 2330 // Now we know what is live, we check the live precessors of the exit block 2331 // and look for fall through paths, being careful to ignore normal returns, 2332 // and exceptional paths. 2333 bool HasLiveReturn = false; 2334 bool HasFakeEdge = false; 2335 bool HasPlainEdge = false; 2336 bool HasAbnormalEdge = false; 2337 for (CFGBlock::pred_iterator I=cfg->getExit().pred_begin(), 2338 E = cfg->getExit().pred_end(); 2339 I != E; 2340 ++I) { 2341 CFGBlock& B = **I; 2342 if (!live[B.getBlockID()]) 2343 continue; 2344 if (B.size() == 0) { 2345 if (B.getTerminator() && isa<CXXTryStmt>(B.getTerminator())) { 2346 HasAbnormalEdge = true; 2347 continue; 2348 } 2349 2350 // A labeled empty statement, or the entry block... 2351 HasPlainEdge = true; 2352 continue; 2353 } 2354 Stmt *S = B[B.size()-1]; 2355 if (isa<ReturnStmt>(S)) { 2356 HasLiveReturn = true; 2357 continue; 2358 } 2359 if (isa<ObjCAtThrowStmt>(S)) { 2360 HasFakeEdge = true; 2361 continue; 2362 } 2363 if (isa<CXXThrowExpr>(S)) { 2364 HasFakeEdge = true; 2365 continue; 2366 } 2367 if (const AsmStmt *AS = dyn_cast<AsmStmt>(S)) { 2368 if (AS->isMSAsm()) { 2369 HasFakeEdge = true; 2370 HasLiveReturn = true; 2371 continue; 2372 } 2373 } 2374 if (isa<CXXTryStmt>(S)) { 2375 HasAbnormalEdge = true; 2376 continue; 2377 } 2378 2379 bool NoReturnEdge = false; 2380 if (CallExpr *C = dyn_cast<CallExpr>(S)) { 2381 if (B.succ_begin()[0] != &cfg->getExit()) { 2382 HasAbnormalEdge = true; 2383 continue; 2384 } 2385 Expr *CEE = C->getCallee()->IgnoreParenCasts(); 2386 if (CEE->getType().getNoReturnAttr()) { 2387 NoReturnEdge = true; 2388 HasFakeEdge = true; 2389 } else if (DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(CEE)) { 2390 ValueDecl *VD = DRE->getDecl(); 2391 if (VD->hasAttr<NoReturnAttr>()) { 2392 NoReturnEdge = true; 2393 HasFakeEdge = true; 2394 } 2395 } 2396 } 2397 // FIXME: Add noreturn message sends. 2398 if (NoReturnEdge == false) 2399 HasPlainEdge = true; 2400 } 2401 if (!HasPlainEdge) { 2402 if (HasLiveReturn) 2403 return NeverFallThrough; 2404 return NeverFallThroughOrReturn; 2405 } 2406 if (HasAbnormalEdge || HasFakeEdge || HasLiveReturn) 2407 return MaybeFallThrough; 2408 // This says AlwaysFallThrough for calls to functions that are not marked 2409 // noreturn, that don't return. If people would like this warning to be more 2410 // accurate, such functions should be marked as noreturn. 2411 return AlwaysFallThrough; 2412} 2413 2414/// CheckFallThroughForFunctionDef - Check that we don't fall off the end of a 2415/// function that should return a value. Check that we don't fall off the end 2416/// of a noreturn function. We assume that functions and blocks not marked 2417/// noreturn will return. 2418void Sema::CheckFallThroughForFunctionDef(Decl *D, Stmt *Body, 2419 AnalysisContext &AC) { 2420 // FIXME: Would be nice if we had a better way to control cascading errors, 2421 // but for now, avoid them. The problem is that when Parse sees: 2422 // int foo() { return a; } 2423 // The return is eaten and the Sema code sees just: 2424 // int foo() { } 2425 // which this code would then warn about. 2426 if (getDiagnostics().hasErrorOccurred()) 2427 return; 2428 2429 bool ReturnsVoid = false; 2430 bool HasNoReturn = false; 2431 if (FunctionDecl *FD = dyn_cast<FunctionDecl>(D)) { 2432 // If the result type of the function is a dependent type, we don't know 2433 // whether it will be void or not, so don't 2434 if (FD->getResultType()->isDependentType()) 2435 return; 2436 if (FD->getResultType()->isVoidType()) 2437 ReturnsVoid = true; 2438 if (FD->hasAttr<NoReturnAttr>()) 2439 HasNoReturn = true; 2440 } else if (ObjCMethodDecl *MD = dyn_cast<ObjCMethodDecl>(D)) { 2441 if (MD->getResultType()->isVoidType()) 2442 ReturnsVoid = true; 2443 if (MD->hasAttr<NoReturnAttr>()) 2444 HasNoReturn = true; 2445 } 2446 2447 // Short circuit for compilation speed. 2448 if ((Diags.getDiagnosticLevel(diag::warn_maybe_falloff_nonvoid_function) 2449 == Diagnostic::Ignored || ReturnsVoid) 2450 && (Diags.getDiagnosticLevel(diag::warn_noreturn_function_has_return_expr) 2451 == Diagnostic::Ignored || !HasNoReturn) 2452 && (Diags.getDiagnosticLevel(diag::warn_suggest_noreturn_block) 2453 == Diagnostic::Ignored || !ReturnsVoid)) 2454 return; 2455 // FIXME: Function try block 2456 if (CompoundStmt *Compound = dyn_cast<CompoundStmt>(Body)) { 2457 switch (CheckFallThrough(AC)) { 2458 case MaybeFallThrough: 2459 if (HasNoReturn) 2460 Diag(Compound->getRBracLoc(), diag::warn_falloff_noreturn_function); 2461 else if (!ReturnsVoid) 2462 Diag(Compound->getRBracLoc(),diag::warn_maybe_falloff_nonvoid_function); 2463 break; 2464 case AlwaysFallThrough: 2465 if (HasNoReturn) 2466 Diag(Compound->getRBracLoc(), diag::warn_falloff_noreturn_function); 2467 else if (!ReturnsVoid) 2468 Diag(Compound->getRBracLoc(), diag::warn_falloff_nonvoid_function); 2469 break; 2470 case NeverFallThroughOrReturn: 2471 if (ReturnsVoid && !HasNoReturn) 2472 Diag(Compound->getLBracLoc(), diag::warn_suggest_noreturn_function); 2473 break; 2474 case NeverFallThrough: 2475 break; 2476 } 2477 } 2478} 2479 2480/// CheckFallThroughForBlock - Check that we don't fall off the end of a block 2481/// that should return a value. Check that we don't fall off the end of a 2482/// noreturn block. We assume that functions and blocks not marked noreturn 2483/// will return. 2484void Sema::CheckFallThroughForBlock(QualType BlockTy, Stmt *Body, 2485 AnalysisContext &AC) { 2486 // FIXME: Would be nice if we had a better way to control cascading errors, 2487 // but for now, avoid them. The problem is that when Parse sees: 2488 // int foo() { return a; } 2489 // The return is eaten and the Sema code sees just: 2490 // int foo() { } 2491 // which this code would then warn about. 2492 if (getDiagnostics().hasErrorOccurred()) 2493 return; 2494 bool ReturnsVoid = false; 2495 bool HasNoReturn = false; 2496 if (const FunctionType *FT =BlockTy->getPointeeType()->getAs<FunctionType>()){ 2497 if (FT->getResultType()->isVoidType()) 2498 ReturnsVoid = true; 2499 if (FT->getNoReturnAttr()) 2500 HasNoReturn = true; 2501 } 2502 2503 // Short circuit for compilation speed. 2504 if (ReturnsVoid 2505 && !HasNoReturn 2506 && (Diags.getDiagnosticLevel(diag::warn_suggest_noreturn_block) 2507 == Diagnostic::Ignored || !ReturnsVoid)) 2508 return; 2509 // FIXME: Funtion try block 2510 if (CompoundStmt *Compound = dyn_cast<CompoundStmt>(Body)) { 2511 switch (CheckFallThrough(AC)) { 2512 case MaybeFallThrough: 2513 if (HasNoReturn) 2514 Diag(Compound->getRBracLoc(), diag::err_noreturn_block_has_return_expr); 2515 else if (!ReturnsVoid) 2516 Diag(Compound->getRBracLoc(), diag::err_maybe_falloff_nonvoid_block); 2517 break; 2518 case AlwaysFallThrough: 2519 if (HasNoReturn) 2520 Diag(Compound->getRBracLoc(), diag::err_noreturn_block_has_return_expr); 2521 else if (!ReturnsVoid) 2522 Diag(Compound->getRBracLoc(), diag::err_falloff_nonvoid_block); 2523 break; 2524 case NeverFallThroughOrReturn: 2525 if (ReturnsVoid) 2526 Diag(Compound->getLBracLoc(), diag::warn_suggest_noreturn_block); 2527 break; 2528 case NeverFallThrough: 2529 break; 2530 } 2531 } 2532} 2533 2534/// CheckParmsForFunctionDef - Check that the parameters of the given 2535/// function are appropriate for the definition of a function. This 2536/// takes care of any checks that cannot be performed on the 2537/// declaration itself, e.g., that the types of each of the function 2538/// parameters are complete. 2539bool Sema::CheckParmsForFunctionDef(FunctionDecl *FD) { 2540 bool HasInvalidParm = false; 2541 for (unsigned p = 0, NumParams = FD->getNumParams(); p < NumParams; ++p) { 2542 ParmVarDecl *Param = FD->getParamDecl(p); 2543 2544 // C99 6.7.5.3p4: the parameters in a parameter type list in a 2545 // function declarator that is part of a function definition of 2546 // that function shall not have incomplete type. 2547 // 2548 // This is also C++ [dcl.fct]p6. 2549 if (!Param->isInvalidDecl() && 2550 RequireCompleteType(Param->getLocation(), Param->getType(), 2551 diag::err_typecheck_decl_incomplete_type)) { 2552 Param->setInvalidDecl(); 2553 HasInvalidParm = true; 2554 } 2555 2556 // C99 6.9.1p5: If the declarator includes a parameter type list, the 2557 // declaration of each parameter shall include an identifier. 2558 if (Param->getIdentifier() == 0 && 2559 !Param->isImplicit() && 2560 !getLangOptions().CPlusPlus) 2561 Diag(Param->getLocation(), diag::err_parameter_name_omitted); 2562 } 2563 2564 return HasInvalidParm; 2565} 2566