SemaChecking.cpp revision 74d56a168966ff015824279a24aaf566180ed97d
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 if (const Expr *Init = VD->getAnyInitializer()) 850 return SemaCheckStringLiteral(Init, TheCall, 851 HasVAListArg, format_idx, firstDataArg); 852 } 853 854 // For vprintf* functions (i.e., HasVAListArg==true), we add a 855 // special check to see if the format string is a function parameter 856 // of the function calling the printf function. If the function 857 // has an attribute indicating it is a printf-like function, then we 858 // should suppress warnings concerning non-literals being used in a call 859 // to a vprintf function. For example: 860 // 861 // void 862 // logmessage(char const *fmt __attribute__ (format (printf, 1, 2)), ...){ 863 // va_list ap; 864 // va_start(ap, fmt); 865 // vprintf(fmt, ap); // Do NOT emit a warning about "fmt". 866 // ... 867 // 868 // 869 // FIXME: We don't have full attribute support yet, so just check to see 870 // if the argument is a DeclRefExpr that references a parameter. We'll 871 // add proper support for checking the attribute later. 872 if (HasVAListArg) 873 if (isa<ParmVarDecl>(VD)) 874 return true; 875 } 876 877 return false; 878 } 879 880 case Stmt::CallExprClass: { 881 const CallExpr *CE = cast<CallExpr>(E); 882 if (const ImplicitCastExpr *ICE 883 = dyn_cast<ImplicitCastExpr>(CE->getCallee())) { 884 if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(ICE->getSubExpr())) { 885 if (const FunctionDecl *FD = dyn_cast<FunctionDecl>(DRE->getDecl())) { 886 if (const FormatArgAttr *FA = FD->getAttr<FormatArgAttr>()) { 887 unsigned ArgIndex = FA->getFormatIdx(); 888 const Expr *Arg = CE->getArg(ArgIndex - 1); 889 890 return SemaCheckStringLiteral(Arg, TheCall, HasVAListArg, 891 format_idx, firstDataArg); 892 } 893 } 894 } 895 } 896 897 return false; 898 } 899 case Stmt::ObjCStringLiteralClass: 900 case Stmt::StringLiteralClass: { 901 const StringLiteral *StrE = NULL; 902 903 if (const ObjCStringLiteral *ObjCFExpr = dyn_cast<ObjCStringLiteral>(E)) 904 StrE = ObjCFExpr->getString(); 905 else 906 StrE = cast<StringLiteral>(E); 907 908 if (StrE) { 909 CheckPrintfString(StrE, E, TheCall, HasVAListArg, format_idx, 910 firstDataArg); 911 return true; 912 } 913 914 return false; 915 } 916 917 default: 918 return false; 919 } 920} 921 922void 923Sema::CheckNonNullArguments(const NonNullAttr *NonNull, 924 const CallExpr *TheCall) { 925 for (NonNullAttr::iterator i = NonNull->begin(), e = NonNull->end(); 926 i != e; ++i) { 927 const Expr *ArgExpr = TheCall->getArg(*i); 928 if (ArgExpr->isNullPointerConstant(Context, 929 Expr::NPC_ValueDependentIsNotNull)) 930 Diag(TheCall->getCallee()->getLocStart(), diag::warn_null_arg) 931 << ArgExpr->getSourceRange(); 932 } 933} 934 935/// CheckPrintfArguments - Check calls to printf (and similar functions) for 936/// correct use of format strings. 937/// 938/// HasVAListArg - A predicate indicating whether the printf-like 939/// function is passed an explicit va_arg argument (e.g., vprintf) 940/// 941/// format_idx - The index into Args for the format string. 942/// 943/// Improper format strings to functions in the printf family can be 944/// the source of bizarre bugs and very serious security holes. A 945/// good source of information is available in the following paper 946/// (which includes additional references): 947/// 948/// FormatGuard: Automatic Protection From printf Format String 949/// Vulnerabilities, Proceedings of the 10th USENIX Security Symposium, 2001. 950/// 951/// Functionality implemented: 952/// 953/// We can statically check the following properties for string 954/// literal format strings for non v.*printf functions (where the 955/// arguments are passed directly): 956// 957/// (1) Are the number of format conversions equal to the number of 958/// data arguments? 959/// 960/// (2) Does each format conversion correctly match the type of the 961/// corresponding data argument? (TODO) 962/// 963/// Moreover, for all printf functions we can: 964/// 965/// (3) Check for a missing format string (when not caught by type checking). 966/// 967/// (4) Check for no-operation flags; e.g. using "#" with format 968/// conversion 'c' (TODO) 969/// 970/// (5) Check the use of '%n', a major source of security holes. 971/// 972/// (6) Check for malformed format conversions that don't specify anything. 973/// 974/// (7) Check for empty format strings. e.g: printf(""); 975/// 976/// (8) Check that the format string is a wide literal. 977/// 978/// All of these checks can be done by parsing the format string. 979/// 980/// For now, we ONLY do (1), (3), (5), (6), (7), and (8). 981void 982Sema::CheckPrintfArguments(const CallExpr *TheCall, bool HasVAListArg, 983 unsigned format_idx, unsigned firstDataArg) { 984 const Expr *Fn = TheCall->getCallee(); 985 986 // The way the format attribute works in GCC, the implicit this argument 987 // of member functions is counted. However, it doesn't appear in our own 988 // lists, so decrement format_idx in that case. 989 if (isa<CXXMemberCallExpr>(TheCall)) { 990 // Catch a format attribute mistakenly referring to the object argument. 991 if (format_idx == 0) 992 return; 993 --format_idx; 994 if(firstDataArg != 0) 995 --firstDataArg; 996 } 997 998 // CHECK: printf-like function is called with no format string. 999 if (format_idx >= TheCall->getNumArgs()) { 1000 Diag(TheCall->getRParenLoc(), diag::warn_printf_missing_format_string) 1001 << Fn->getSourceRange(); 1002 return; 1003 } 1004 1005 const Expr *OrigFormatExpr = TheCall->getArg(format_idx)->IgnoreParenCasts(); 1006 1007 // CHECK: format string is not a string literal. 1008 // 1009 // Dynamically generated format strings are difficult to 1010 // automatically vet at compile time. Requiring that format strings 1011 // are string literals: (1) permits the checking of format strings by 1012 // the compiler and thereby (2) can practically remove the source of 1013 // many format string exploits. 1014 1015 // Format string can be either ObjC string (e.g. @"%d") or 1016 // C string (e.g. "%d") 1017 // ObjC string uses the same format specifiers as C string, so we can use 1018 // the same format string checking logic for both ObjC and C strings. 1019 if (SemaCheckStringLiteral(OrigFormatExpr, TheCall, HasVAListArg, format_idx, 1020 firstDataArg)) 1021 return; // Literal format string found, check done! 1022 1023 // If there are no arguments specified, warn with -Wformat-security, otherwise 1024 // warn only with -Wformat-nonliteral. 1025 if (TheCall->getNumArgs() == format_idx+1) 1026 Diag(TheCall->getArg(format_idx)->getLocStart(), 1027 diag::warn_printf_nonliteral_noargs) 1028 << OrigFormatExpr->getSourceRange(); 1029 else 1030 Diag(TheCall->getArg(format_idx)->getLocStart(), 1031 diag::warn_printf_nonliteral) 1032 << OrigFormatExpr->getSourceRange(); 1033} 1034 1035namespace { 1036class CheckPrintfHandler : public analyze_printf::FormatStringHandler { 1037 Sema &S; 1038 const StringLiteral *FExpr; 1039 const Expr *OrigFormatExpr; 1040 unsigned NumConversions; 1041 const unsigned NumDataArgs; 1042 const bool IsObjCLiteral; 1043 const char *Beg; // Start of format string. 1044 const bool HasVAListArg; 1045 const CallExpr *TheCall; 1046 unsigned FormatIdx; 1047public: 1048 CheckPrintfHandler(Sema &s, const StringLiteral *fexpr, 1049 const Expr *origFormatExpr, 1050 unsigned numDataArgs, bool isObjCLiteral, 1051 const char *beg, bool hasVAListArg, 1052 const CallExpr *theCall, unsigned formatIdx) 1053 : S(s), FExpr(fexpr), OrigFormatExpr(origFormatExpr), 1054 NumConversions(0), NumDataArgs(numDataArgs), 1055 IsObjCLiteral(isObjCLiteral), Beg(beg), 1056 HasVAListArg(hasVAListArg), 1057 TheCall(theCall), FormatIdx(formatIdx) {} 1058 1059 void DoneProcessing(); 1060 1061 void HandleIncompleteFormatSpecifier(const char *startSpecifier, 1062 unsigned specifierLen); 1063 1064 void 1065 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 bool MatchType(QualType A, QualType B, bool ignoreSign); 1085 1086 const Expr *getDataArg(unsigned i) const; 1087}; 1088} 1089 1090SourceRange CheckPrintfHandler::getFormatStringRange() { 1091 return OrigFormatExpr->getSourceRange(); 1092} 1093 1094SourceRange CheckPrintfHandler:: 1095getFormatSpecifierRange(const char *startSpecifier, unsigned specifierLen) { 1096 return SourceRange(getLocationOfByte(startSpecifier), 1097 getLocationOfByte(startSpecifier+specifierLen-1)); 1098} 1099 1100SourceLocation CheckPrintfHandler::getLocationOfByte(const char *x) { 1101 return S.getLocationOfStringLiteralByte(FExpr, x - Beg); 1102} 1103 1104void CheckPrintfHandler:: 1105HandleIncompleteFormatSpecifier(const char *startSpecifier, 1106 unsigned specifierLen) { 1107 SourceLocation Loc = getLocationOfByte(startSpecifier); 1108 S.Diag(Loc, diag::warn_printf_incomplete_specifier) 1109 << getFormatSpecifierRange(startSpecifier, specifierLen); 1110} 1111 1112void CheckPrintfHandler:: 1113HandleInvalidConversionSpecifier(const analyze_printf::FormatSpecifier &FS, 1114 const char *startSpecifier, 1115 unsigned specifierLen) { 1116 1117 ++NumConversions; 1118 const analyze_printf::ConversionSpecifier &CS = 1119 FS.getConversionSpecifier(); 1120 SourceLocation Loc = getLocationOfByte(CS.getStart()); 1121 S.Diag(Loc, diag::warn_printf_invalid_conversion) 1122 << llvm::StringRef(CS.getStart(), CS.getLength()) 1123 << getFormatSpecifierRange(startSpecifier, specifierLen); 1124} 1125 1126void CheckPrintfHandler::HandleNullChar(const char *nullCharacter) { 1127 // The presence of a null character is likely an error. 1128 S.Diag(getLocationOfByte(nullCharacter), 1129 diag::warn_printf_format_string_contains_null_char) 1130 << getFormatStringRange(); 1131} 1132 1133const Expr *CheckPrintfHandler::getDataArg(unsigned i) const { 1134 return TheCall->getArg(FormatIdx + i); 1135} 1136 1137bool CheckPrintfHandler::MatchType(QualType A, QualType B, bool ignoreSign) { 1138 A = S.Context.getCanonicalType(A).getUnqualifiedType(); 1139 B = S.Context.getCanonicalType(B).getUnqualifiedType(); 1140 1141 if (A == B) 1142 return true; 1143 1144 if (ignoreSign) { 1145 if (const BuiltinType *BT = B->getAs<BuiltinType>()) { 1146 switch (BT->getKind()) { 1147 default: 1148 return false; 1149 case BuiltinType::Char_S: 1150 case BuiltinType::SChar: 1151 return A == S.Context.UnsignedCharTy; 1152 case BuiltinType::Char_U: 1153 case BuiltinType::UChar: 1154 return A == S.Context.SignedCharTy; 1155 case BuiltinType::Short: 1156 return A == S.Context.UnsignedShortTy; 1157 case BuiltinType::UShort: 1158 return A == S.Context.ShortTy; 1159 case BuiltinType::Int: 1160 return A == S.Context.UnsignedIntTy; 1161 case BuiltinType::UInt: 1162 return A == S.Context.IntTy; 1163 case BuiltinType::Long: 1164 return A == S.Context.UnsignedLongTy; 1165 case BuiltinType::ULong: 1166 return A == S.Context.LongTy; 1167 case BuiltinType::LongLong: 1168 return A == S.Context.UnsignedLongLongTy; 1169 case BuiltinType::ULongLong: 1170 return A == S.Context.LongLongTy; 1171 } 1172 return A == B; 1173 } 1174 } 1175 return false; 1176} 1177 1178bool 1179CheckPrintfHandler::HandleAmount(const analyze_printf::OptionalAmount &Amt, 1180 unsigned MissingArgDiag, 1181 unsigned BadTypeDiag, 1182 const char *startSpecifier, 1183 unsigned specifierLen) { 1184 1185 if (Amt.hasDataArgument()) { 1186 ++NumConversions; 1187 if (!HasVAListArg) { 1188 if (NumConversions > NumDataArgs) { 1189 S.Diag(getLocationOfByte(Amt.getStart()), MissingArgDiag) 1190 << getFormatSpecifierRange(startSpecifier, specifierLen); 1191 // Don't do any more checking. We will just emit 1192 // spurious errors. 1193 return false; 1194 } 1195 1196 // Type check the data argument. It should be an 'int'. 1197 // Although not in conformance with C99, we also allow the argument to be 1198 // an 'unsigned int' as that is a reasonably safe case. GCC also 1199 // doesn't emit a warning for that case. 1200 const Expr *Arg = getDataArg(NumConversions); 1201 QualType T = Arg->getType(); 1202 if (!MatchType(T, S.Context.IntTy, true)) { 1203 S.Diag(getLocationOfByte(Amt.getStart()), BadTypeDiag) 1204 << S.Context.IntTy << T 1205 << getFormatSpecifierRange(startSpecifier, specifierLen) 1206 << Arg->getSourceRange(); 1207 // Don't do any more checking. We will just emit 1208 // spurious errors. 1209 return false; 1210 } 1211 } 1212 } 1213 return true; 1214} 1215 1216bool 1217CheckPrintfHandler::HandleFormatSpecifier(const analyze_printf::FormatSpecifier &FS, 1218 const char *startSpecifier, 1219 unsigned specifierLen) { 1220 1221 using namespace analyze_printf; 1222 const ConversionSpecifier &CS = FS.getConversionSpecifier(); 1223 1224 // First check if the field width, precision, and conversion specifier 1225 // have matching data arguments. 1226 if (!HandleAmount(FS.getFieldWidth(), 1227 diag::warn_printf_asterisk_width_missing_arg, 1228 diag::warn_printf_asterisk_width_wrong_type, 1229 startSpecifier, specifierLen)) { 1230 return false; 1231 } 1232 1233 if (!HandleAmount(FS.getPrecision(), 1234 diag::warn_printf_asterisk_precision_missing_arg, 1235 diag::warn_printf_asterisk_precision_wrong_type, 1236 startSpecifier, specifierLen)) { 1237 return false; 1238 } 1239 1240 // Check for using an Objective-C specific conversion specifier 1241 // in a non-ObjC literal. 1242 if (!IsObjCLiteral && CS.isObjCArg()) { 1243 HandleInvalidConversionSpecifier(FS, startSpecifier, specifierLen); 1244 1245 // Continue checking the other format specifiers. 1246 return true; 1247 } 1248 1249 if (!CS.consumesDataArgument()) { 1250 // FIXME: Technically specifying a precision or field width here 1251 // makes no sense. Worth issuing a warning at some point. 1252 return true; 1253 } 1254 1255 ++NumConversions; 1256 1257 // Are we using '%n'? Issue a warning about this being 1258 // a possible security issue. 1259 if (CS.getKind() == ConversionSpecifier::OutIntPtrArg) { 1260 S.Diag(getLocationOfByte(CS.getStart()), diag::warn_printf_write_back) 1261 << getFormatSpecifierRange(startSpecifier, specifierLen); 1262 // Continue checking the other format specifiers. 1263 return true; 1264 } 1265 1266 1267 // The remaining checks depend on the data arguments. 1268 if (HasVAListArg) 1269 return true; 1270 1271 if (NumConversions > NumDataArgs) { 1272 S.Diag(getLocationOfByte(CS.getStart()), 1273 diag::warn_printf_insufficient_data_args) 1274 << getFormatSpecifierRange(startSpecifier, specifierLen); 1275 // Don't do any more checking. 1276 return false; 1277 } 1278 1279 // Now type check the data expression that matches the 1280 // format specifier. 1281 const Expr *Ex = getDataArg(NumConversions); 1282 const analyze_printf::ArgTypeResult &ATR = FS.getArgType(S.Context); 1283 1284 if (const QualType *T = ATR.getSpecificType()) { 1285 if (!MatchType(*T, Ex->getType(), true)) { 1286 // Check if we didn't match because of an implicit cast from a 'char' 1287 // or 'short' to an 'int'. This is done because printf is a varargs 1288 // function. 1289 if (const ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(Ex)) 1290 if (ICE->getType() == S.Context.IntTy) 1291 if (MatchType(*T, ICE->getSubExpr()->getType(), true)) 1292 return true; 1293 1294 S.Diag(getLocationOfByte(CS.getStart()), 1295 diag::warn_printf_conversion_argument_type_mismatch) 1296 << *T << Ex->getType(); 1297// << getFormatSpecifierRange(startSpecifier, specifierLen) 1298// << Ex->getSourceRange(); 1299 } 1300 return true; 1301 } 1302 1303 return true; 1304} 1305 1306void CheckPrintfHandler::DoneProcessing() { 1307 // Does the number of data arguments exceed the number of 1308 // format conversions in the format string? 1309 if (!HasVAListArg && NumConversions < NumDataArgs) 1310 S.Diag(getDataArg(NumConversions+1)->getLocStart(), 1311 diag::warn_printf_too_many_data_args) 1312 << getFormatStringRange(); 1313} 1314 1315void Sema::CheckPrintfString(const StringLiteral *FExpr, 1316 const Expr *OrigFormatExpr, 1317 const CallExpr *TheCall, bool HasVAListArg, 1318 unsigned format_idx, unsigned firstDataArg) { 1319 1320 // CHECK: is the format string a wide literal? 1321 if (FExpr->isWide()) { 1322 Diag(FExpr->getLocStart(), 1323 diag::warn_printf_format_string_is_wide_literal) 1324 << OrigFormatExpr->getSourceRange(); 1325 return; 1326 } 1327 1328 // Str - The format string. NOTE: this is NOT null-terminated! 1329 const char *Str = FExpr->getStrData(); 1330 1331 // CHECK: empty format string? 1332 unsigned StrLen = FExpr->getByteLength(); 1333 1334 if (StrLen == 0) { 1335 Diag(FExpr->getLocStart(), diag::warn_printf_empty_format_string) 1336 << OrigFormatExpr->getSourceRange(); 1337 return; 1338 } 1339 1340 CheckPrintfHandler H(*this, FExpr, OrigFormatExpr, 1341 TheCall->getNumArgs() - firstDataArg, 1342 isa<ObjCStringLiteral>(OrigFormatExpr), Str, 1343 HasVAListArg, TheCall, format_idx); 1344 1345 if (!analyze_printf::ParseFormatString(H, Str, Str + StrLen)) 1346 H.DoneProcessing(); 1347} 1348 1349//===--- CHECK: Return Address of Stack Variable --------------------------===// 1350 1351static DeclRefExpr* EvalVal(Expr *E); 1352static DeclRefExpr* EvalAddr(Expr* E); 1353 1354/// CheckReturnStackAddr - Check if a return statement returns the address 1355/// of a stack variable. 1356void 1357Sema::CheckReturnStackAddr(Expr *RetValExp, QualType lhsType, 1358 SourceLocation ReturnLoc) { 1359 1360 // Perform checking for returned stack addresses. 1361 if (lhsType->isPointerType() || lhsType->isBlockPointerType()) { 1362 if (DeclRefExpr *DR = EvalAddr(RetValExp)) 1363 Diag(DR->getLocStart(), diag::warn_ret_stack_addr) 1364 << DR->getDecl()->getDeclName() << RetValExp->getSourceRange(); 1365 1366 // Skip over implicit cast expressions when checking for block expressions. 1367 RetValExp = RetValExp->IgnoreParenCasts(); 1368 1369 if (BlockExpr *C = dyn_cast<BlockExpr>(RetValExp)) 1370 if (C->hasBlockDeclRefExprs()) 1371 Diag(C->getLocStart(), diag::err_ret_local_block) 1372 << C->getSourceRange(); 1373 1374 if (AddrLabelExpr *ALE = dyn_cast<AddrLabelExpr>(RetValExp)) 1375 Diag(ALE->getLocStart(), diag::warn_ret_addr_label) 1376 << ALE->getSourceRange(); 1377 1378 } else if (lhsType->isReferenceType()) { 1379 // Perform checking for stack values returned by reference. 1380 // Check for a reference to the stack 1381 if (DeclRefExpr *DR = EvalVal(RetValExp)) 1382 Diag(DR->getLocStart(), diag::warn_ret_stack_ref) 1383 << DR->getDecl()->getDeclName() << RetValExp->getSourceRange(); 1384 } 1385} 1386 1387/// EvalAddr - EvalAddr and EvalVal are mutually recursive functions that 1388/// check if the expression in a return statement evaluates to an address 1389/// to a location on the stack. The recursion is used to traverse the 1390/// AST of the return expression, with recursion backtracking when we 1391/// encounter a subexpression that (1) clearly does not lead to the address 1392/// of a stack variable or (2) is something we cannot determine leads to 1393/// the address of a stack variable based on such local checking. 1394/// 1395/// EvalAddr processes expressions that are pointers that are used as 1396/// references (and not L-values). EvalVal handles all other values. 1397/// At the base case of the recursion is a check for a DeclRefExpr* in 1398/// the refers to a stack variable. 1399/// 1400/// This implementation handles: 1401/// 1402/// * pointer-to-pointer casts 1403/// * implicit conversions from array references to pointers 1404/// * taking the address of fields 1405/// * arbitrary interplay between "&" and "*" operators 1406/// * pointer arithmetic from an address of a stack variable 1407/// * taking the address of an array element where the array is on the stack 1408static DeclRefExpr* EvalAddr(Expr *E) { 1409 // We should only be called for evaluating pointer expressions. 1410 assert((E->getType()->isAnyPointerType() || 1411 E->getType()->isBlockPointerType() || 1412 E->getType()->isObjCQualifiedIdType()) && 1413 "EvalAddr only works on pointers"); 1414 1415 // Our "symbolic interpreter" is just a dispatch off the currently 1416 // viewed AST node. We then recursively traverse the AST by calling 1417 // EvalAddr and EvalVal appropriately. 1418 switch (E->getStmtClass()) { 1419 case Stmt::ParenExprClass: 1420 // Ignore parentheses. 1421 return EvalAddr(cast<ParenExpr>(E)->getSubExpr()); 1422 1423 case Stmt::UnaryOperatorClass: { 1424 // The only unary operator that make sense to handle here 1425 // is AddrOf. All others don't make sense as pointers. 1426 UnaryOperator *U = cast<UnaryOperator>(E); 1427 1428 if (U->getOpcode() == UnaryOperator::AddrOf) 1429 return EvalVal(U->getSubExpr()); 1430 else 1431 return NULL; 1432 } 1433 1434 case Stmt::BinaryOperatorClass: { 1435 // Handle pointer arithmetic. All other binary operators are not valid 1436 // in this context. 1437 BinaryOperator *B = cast<BinaryOperator>(E); 1438 BinaryOperator::Opcode op = B->getOpcode(); 1439 1440 if (op != BinaryOperator::Add && op != BinaryOperator::Sub) 1441 return NULL; 1442 1443 Expr *Base = B->getLHS(); 1444 1445 // Determine which argument is the real pointer base. It could be 1446 // the RHS argument instead of the LHS. 1447 if (!Base->getType()->isPointerType()) Base = B->getRHS(); 1448 1449 assert (Base->getType()->isPointerType()); 1450 return EvalAddr(Base); 1451 } 1452 1453 // For conditional operators we need to see if either the LHS or RHS are 1454 // valid DeclRefExpr*s. If one of them is valid, we return it. 1455 case Stmt::ConditionalOperatorClass: { 1456 ConditionalOperator *C = cast<ConditionalOperator>(E); 1457 1458 // Handle the GNU extension for missing LHS. 1459 if (Expr *lhsExpr = C->getLHS()) 1460 if (DeclRefExpr* LHS = EvalAddr(lhsExpr)) 1461 return LHS; 1462 1463 return EvalAddr(C->getRHS()); 1464 } 1465 1466 // For casts, we need to handle conversions from arrays to 1467 // pointer values, and pointer-to-pointer conversions. 1468 case Stmt::ImplicitCastExprClass: 1469 case Stmt::CStyleCastExprClass: 1470 case Stmt::CXXFunctionalCastExprClass: { 1471 Expr* SubExpr = cast<CastExpr>(E)->getSubExpr(); 1472 QualType T = SubExpr->getType(); 1473 1474 if (SubExpr->getType()->isPointerType() || 1475 SubExpr->getType()->isBlockPointerType() || 1476 SubExpr->getType()->isObjCQualifiedIdType()) 1477 return EvalAddr(SubExpr); 1478 else if (T->isArrayType()) 1479 return EvalVal(SubExpr); 1480 else 1481 return 0; 1482 } 1483 1484 // C++ casts. For dynamic casts, static casts, and const casts, we 1485 // are always converting from a pointer-to-pointer, so we just blow 1486 // through the cast. In the case the dynamic cast doesn't fail (and 1487 // return NULL), we take the conservative route and report cases 1488 // where we return the address of a stack variable. For Reinterpre 1489 // FIXME: The comment about is wrong; we're not always converting 1490 // from pointer to pointer. I'm guessing that this code should also 1491 // handle references to objects. 1492 case Stmt::CXXStaticCastExprClass: 1493 case Stmt::CXXDynamicCastExprClass: 1494 case Stmt::CXXConstCastExprClass: 1495 case Stmt::CXXReinterpretCastExprClass: { 1496 Expr *S = cast<CXXNamedCastExpr>(E)->getSubExpr(); 1497 if (S->getType()->isPointerType() || S->getType()->isBlockPointerType()) 1498 return EvalAddr(S); 1499 else 1500 return NULL; 1501 } 1502 1503 // Everything else: we simply don't reason about them. 1504 default: 1505 return NULL; 1506 } 1507} 1508 1509 1510/// EvalVal - This function is complements EvalAddr in the mutual recursion. 1511/// See the comments for EvalAddr for more details. 1512static DeclRefExpr* EvalVal(Expr *E) { 1513 1514 // We should only be called for evaluating non-pointer expressions, or 1515 // expressions with a pointer type that are not used as references but instead 1516 // are l-values (e.g., DeclRefExpr with a pointer type). 1517 1518 // Our "symbolic interpreter" is just a dispatch off the currently 1519 // viewed AST node. We then recursively traverse the AST by calling 1520 // EvalAddr and EvalVal appropriately. 1521 switch (E->getStmtClass()) { 1522 case Stmt::DeclRefExprClass: { 1523 // DeclRefExpr: the base case. When we hit a DeclRefExpr we are looking 1524 // at code that refers to a variable's name. We check if it has local 1525 // storage within the function, and if so, return the expression. 1526 DeclRefExpr *DR = cast<DeclRefExpr>(E); 1527 1528 if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl())) 1529 if (V->hasLocalStorage() && !V->getType()->isReferenceType()) return DR; 1530 1531 return NULL; 1532 } 1533 1534 case Stmt::ParenExprClass: 1535 // Ignore parentheses. 1536 return EvalVal(cast<ParenExpr>(E)->getSubExpr()); 1537 1538 case Stmt::UnaryOperatorClass: { 1539 // The only unary operator that make sense to handle here 1540 // is Deref. All others don't resolve to a "name." This includes 1541 // handling all sorts of rvalues passed to a unary operator. 1542 UnaryOperator *U = cast<UnaryOperator>(E); 1543 1544 if (U->getOpcode() == UnaryOperator::Deref) 1545 return EvalAddr(U->getSubExpr()); 1546 1547 return NULL; 1548 } 1549 1550 case Stmt::ArraySubscriptExprClass: { 1551 // Array subscripts are potential references to data on the stack. We 1552 // retrieve the DeclRefExpr* for the array variable if it indeed 1553 // has local storage. 1554 return EvalAddr(cast<ArraySubscriptExpr>(E)->getBase()); 1555 } 1556 1557 case Stmt::ConditionalOperatorClass: { 1558 // For conditional operators we need to see if either the LHS or RHS are 1559 // non-NULL DeclRefExpr's. If one is non-NULL, we return it. 1560 ConditionalOperator *C = cast<ConditionalOperator>(E); 1561 1562 // Handle the GNU extension for missing LHS. 1563 if (Expr *lhsExpr = C->getLHS()) 1564 if (DeclRefExpr *LHS = EvalVal(lhsExpr)) 1565 return LHS; 1566 1567 return EvalVal(C->getRHS()); 1568 } 1569 1570 // Accesses to members are potential references to data on the stack. 1571 case Stmt::MemberExprClass: { 1572 MemberExpr *M = cast<MemberExpr>(E); 1573 1574 // Check for indirect access. We only want direct field accesses. 1575 if (!M->isArrow()) 1576 return EvalVal(M->getBase()); 1577 else 1578 return NULL; 1579 } 1580 1581 // Everything else: we simply don't reason about them. 1582 default: 1583 return NULL; 1584 } 1585} 1586 1587//===--- CHECK: Floating-Point comparisons (-Wfloat-equal) ---------------===// 1588 1589/// Check for comparisons of floating point operands using != and ==. 1590/// Issue a warning if these are no self-comparisons, as they are not likely 1591/// to do what the programmer intended. 1592void Sema::CheckFloatComparison(SourceLocation loc, Expr* lex, Expr *rex) { 1593 bool EmitWarning = true; 1594 1595 Expr* LeftExprSansParen = lex->IgnoreParens(); 1596 Expr* RightExprSansParen = rex->IgnoreParens(); 1597 1598 // Special case: check for x == x (which is OK). 1599 // Do not emit warnings for such cases. 1600 if (DeclRefExpr* DRL = dyn_cast<DeclRefExpr>(LeftExprSansParen)) 1601 if (DeclRefExpr* DRR = dyn_cast<DeclRefExpr>(RightExprSansParen)) 1602 if (DRL->getDecl() == DRR->getDecl()) 1603 EmitWarning = false; 1604 1605 1606 // Special case: check for comparisons against literals that can be exactly 1607 // represented by APFloat. In such cases, do not emit a warning. This 1608 // is a heuristic: often comparison against such literals are used to 1609 // detect if a value in a variable has not changed. This clearly can 1610 // lead to false negatives. 1611 if (EmitWarning) { 1612 if (FloatingLiteral* FLL = dyn_cast<FloatingLiteral>(LeftExprSansParen)) { 1613 if (FLL->isExact()) 1614 EmitWarning = false; 1615 } else 1616 if (FloatingLiteral* FLR = dyn_cast<FloatingLiteral>(RightExprSansParen)){ 1617 if (FLR->isExact()) 1618 EmitWarning = false; 1619 } 1620 } 1621 1622 // Check for comparisons with builtin types. 1623 if (EmitWarning) 1624 if (CallExpr* CL = dyn_cast<CallExpr>(LeftExprSansParen)) 1625 if (CL->isBuiltinCall(Context)) 1626 EmitWarning = false; 1627 1628 if (EmitWarning) 1629 if (CallExpr* CR = dyn_cast<CallExpr>(RightExprSansParen)) 1630 if (CR->isBuiltinCall(Context)) 1631 EmitWarning = false; 1632 1633 // Emit the diagnostic. 1634 if (EmitWarning) 1635 Diag(loc, diag::warn_floatingpoint_eq) 1636 << lex->getSourceRange() << rex->getSourceRange(); 1637} 1638 1639//===--- CHECK: Integer mixed-sign comparisons (-Wsign-compare) --------===// 1640//===--- CHECK: Lossy implicit conversions (-Wconversion) --------------===// 1641 1642namespace { 1643 1644/// Structure recording the 'active' range of an integer-valued 1645/// expression. 1646struct IntRange { 1647 /// The number of bits active in the int. 1648 unsigned Width; 1649 1650 /// True if the int is known not to have negative values. 1651 bool NonNegative; 1652 1653 IntRange() {} 1654 IntRange(unsigned Width, bool NonNegative) 1655 : Width(Width), NonNegative(NonNegative) 1656 {} 1657 1658 // Returns the range of the bool type. 1659 static IntRange forBoolType() { 1660 return IntRange(1, true); 1661 } 1662 1663 // Returns the range of an integral type. 1664 static IntRange forType(ASTContext &C, QualType T) { 1665 return forCanonicalType(C, T->getCanonicalTypeInternal().getTypePtr()); 1666 } 1667 1668 // Returns the range of an integeral type based on its canonical 1669 // representation. 1670 static IntRange forCanonicalType(ASTContext &C, const Type *T) { 1671 assert(T->isCanonicalUnqualified()); 1672 1673 if (const VectorType *VT = dyn_cast<VectorType>(T)) 1674 T = VT->getElementType().getTypePtr(); 1675 if (const ComplexType *CT = dyn_cast<ComplexType>(T)) 1676 T = CT->getElementType().getTypePtr(); 1677 if (const EnumType *ET = dyn_cast<EnumType>(T)) 1678 T = ET->getDecl()->getIntegerType().getTypePtr(); 1679 1680 const BuiltinType *BT = cast<BuiltinType>(T); 1681 assert(BT->isInteger()); 1682 1683 return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger()); 1684 } 1685 1686 // Returns the supremum of two ranges: i.e. their conservative merge. 1687 static IntRange join(const IntRange &L, const IntRange &R) { 1688 return IntRange(std::max(L.Width, R.Width), 1689 L.NonNegative && R.NonNegative); 1690 } 1691 1692 // Returns the infinum of two ranges: i.e. their aggressive merge. 1693 static IntRange meet(const IntRange &L, const IntRange &R) { 1694 return IntRange(std::min(L.Width, R.Width), 1695 L.NonNegative || R.NonNegative); 1696 } 1697}; 1698 1699IntRange GetValueRange(ASTContext &C, llvm::APSInt &value, unsigned MaxWidth) { 1700 if (value.isSigned() && value.isNegative()) 1701 return IntRange(value.getMinSignedBits(), false); 1702 1703 if (value.getBitWidth() > MaxWidth) 1704 value.trunc(MaxWidth); 1705 1706 // isNonNegative() just checks the sign bit without considering 1707 // signedness. 1708 return IntRange(value.getActiveBits(), true); 1709} 1710 1711IntRange GetValueRange(ASTContext &C, APValue &result, QualType Ty, 1712 unsigned MaxWidth) { 1713 if (result.isInt()) 1714 return GetValueRange(C, result.getInt(), MaxWidth); 1715 1716 if (result.isVector()) { 1717 IntRange R = GetValueRange(C, result.getVectorElt(0), Ty, MaxWidth); 1718 for (unsigned i = 1, e = result.getVectorLength(); i != e; ++i) { 1719 IntRange El = GetValueRange(C, result.getVectorElt(i), Ty, MaxWidth); 1720 R = IntRange::join(R, El); 1721 } 1722 return R; 1723 } 1724 1725 if (result.isComplexInt()) { 1726 IntRange R = GetValueRange(C, result.getComplexIntReal(), MaxWidth); 1727 IntRange I = GetValueRange(C, result.getComplexIntImag(), MaxWidth); 1728 return IntRange::join(R, I); 1729 } 1730 1731 // This can happen with lossless casts to intptr_t of "based" lvalues. 1732 // Assume it might use arbitrary bits. 1733 // FIXME: The only reason we need to pass the type in here is to get 1734 // the sign right on this one case. It would be nice if APValue 1735 // preserved this. 1736 assert(result.isLValue()); 1737 return IntRange(MaxWidth, Ty->isUnsignedIntegerType()); 1738} 1739 1740/// Pseudo-evaluate the given integer expression, estimating the 1741/// range of values it might take. 1742/// 1743/// \param MaxWidth - the width to which the value will be truncated 1744IntRange GetExprRange(ASTContext &C, Expr *E, unsigned MaxWidth) { 1745 E = E->IgnoreParens(); 1746 1747 // Try a full evaluation first. 1748 Expr::EvalResult result; 1749 if (E->Evaluate(result, C)) 1750 return GetValueRange(C, result.Val, E->getType(), MaxWidth); 1751 1752 // I think we only want to look through implicit casts here; if the 1753 // user has an explicit widening cast, we should treat the value as 1754 // being of the new, wider type. 1755 if (ImplicitCastExpr *CE = dyn_cast<ImplicitCastExpr>(E)) { 1756 if (CE->getCastKind() == CastExpr::CK_NoOp) 1757 return GetExprRange(C, CE->getSubExpr(), MaxWidth); 1758 1759 IntRange OutputTypeRange = IntRange::forType(C, CE->getType()); 1760 1761 bool isIntegerCast = (CE->getCastKind() == CastExpr::CK_IntegralCast); 1762 if (!isIntegerCast && CE->getCastKind() == CastExpr::CK_Unknown) 1763 isIntegerCast = CE->getSubExpr()->getType()->isIntegerType(); 1764 1765 // Assume that non-integer casts can span the full range of the type. 1766 if (!isIntegerCast) 1767 return OutputTypeRange; 1768 1769 IntRange SubRange 1770 = GetExprRange(C, CE->getSubExpr(), 1771 std::min(MaxWidth, OutputTypeRange.Width)); 1772 1773 // Bail out if the subexpr's range is as wide as the cast type. 1774 if (SubRange.Width >= OutputTypeRange.Width) 1775 return OutputTypeRange; 1776 1777 // Otherwise, we take the smaller width, and we're non-negative if 1778 // either the output type or the subexpr is. 1779 return IntRange(SubRange.Width, 1780 SubRange.NonNegative || OutputTypeRange.NonNegative); 1781 } 1782 1783 if (ConditionalOperator *CO = dyn_cast<ConditionalOperator>(E)) { 1784 // If we can fold the condition, just take that operand. 1785 bool CondResult; 1786 if (CO->getCond()->EvaluateAsBooleanCondition(CondResult, C)) 1787 return GetExprRange(C, CondResult ? CO->getTrueExpr() 1788 : CO->getFalseExpr(), 1789 MaxWidth); 1790 1791 // Otherwise, conservatively merge. 1792 IntRange L = GetExprRange(C, CO->getTrueExpr(), MaxWidth); 1793 IntRange R = GetExprRange(C, CO->getFalseExpr(), MaxWidth); 1794 return IntRange::join(L, R); 1795 } 1796 1797 if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) { 1798 switch (BO->getOpcode()) { 1799 1800 // Boolean-valued operations are single-bit and positive. 1801 case BinaryOperator::LAnd: 1802 case BinaryOperator::LOr: 1803 case BinaryOperator::LT: 1804 case BinaryOperator::GT: 1805 case BinaryOperator::LE: 1806 case BinaryOperator::GE: 1807 case BinaryOperator::EQ: 1808 case BinaryOperator::NE: 1809 return IntRange::forBoolType(); 1810 1811 // Operations with opaque sources are black-listed. 1812 case BinaryOperator::PtrMemD: 1813 case BinaryOperator::PtrMemI: 1814 return IntRange::forType(C, E->getType()); 1815 1816 // Bitwise-and uses the *infinum* of the two source ranges. 1817 case BinaryOperator::And: 1818 return IntRange::meet(GetExprRange(C, BO->getLHS(), MaxWidth), 1819 GetExprRange(C, BO->getRHS(), MaxWidth)); 1820 1821 // Left shift gets black-listed based on a judgement call. 1822 case BinaryOperator::Shl: 1823 return IntRange::forType(C, E->getType()); 1824 1825 // Right shift by a constant can narrow its left argument. 1826 case BinaryOperator::Shr: { 1827 IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth); 1828 1829 // If the shift amount is a positive constant, drop the width by 1830 // that much. 1831 llvm::APSInt shift; 1832 if (BO->getRHS()->isIntegerConstantExpr(shift, C) && 1833 shift.isNonNegative()) { 1834 unsigned zext = shift.getZExtValue(); 1835 if (zext >= L.Width) 1836 L.Width = (L.NonNegative ? 0 : 1); 1837 else 1838 L.Width -= zext; 1839 } 1840 1841 return L; 1842 } 1843 1844 // Comma acts as its right operand. 1845 case BinaryOperator::Comma: 1846 return GetExprRange(C, BO->getRHS(), MaxWidth); 1847 1848 // Black-list pointer subtractions. 1849 case BinaryOperator::Sub: 1850 if (BO->getLHS()->getType()->isPointerType()) 1851 return IntRange::forType(C, E->getType()); 1852 // fallthrough 1853 1854 default: 1855 break; 1856 } 1857 1858 // Treat every other operator as if it were closed on the 1859 // narrowest type that encompasses both operands. 1860 IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth); 1861 IntRange R = GetExprRange(C, BO->getRHS(), MaxWidth); 1862 return IntRange::join(L, R); 1863 } 1864 1865 if (UnaryOperator *UO = dyn_cast<UnaryOperator>(E)) { 1866 switch (UO->getOpcode()) { 1867 // Boolean-valued operations are white-listed. 1868 case UnaryOperator::LNot: 1869 return IntRange::forBoolType(); 1870 1871 // Operations with opaque sources are black-listed. 1872 case UnaryOperator::Deref: 1873 case UnaryOperator::AddrOf: // should be impossible 1874 case UnaryOperator::OffsetOf: 1875 return IntRange::forType(C, E->getType()); 1876 1877 default: 1878 return GetExprRange(C, UO->getSubExpr(), MaxWidth); 1879 } 1880 } 1881 1882 FieldDecl *BitField = E->getBitField(); 1883 if (BitField) { 1884 llvm::APSInt BitWidthAP = BitField->getBitWidth()->EvaluateAsInt(C); 1885 unsigned BitWidth = BitWidthAP.getZExtValue(); 1886 1887 return IntRange(BitWidth, BitField->getType()->isUnsignedIntegerType()); 1888 } 1889 1890 return IntRange::forType(C, E->getType()); 1891} 1892 1893/// Checks whether the given value, which currently has the given 1894/// source semantics, has the same value when coerced through the 1895/// target semantics. 1896bool IsSameFloatAfterCast(const llvm::APFloat &value, 1897 const llvm::fltSemantics &Src, 1898 const llvm::fltSemantics &Tgt) { 1899 llvm::APFloat truncated = value; 1900 1901 bool ignored; 1902 truncated.convert(Src, llvm::APFloat::rmNearestTiesToEven, &ignored); 1903 truncated.convert(Tgt, llvm::APFloat::rmNearestTiesToEven, &ignored); 1904 1905 return truncated.bitwiseIsEqual(value); 1906} 1907 1908/// Checks whether the given value, which currently has the given 1909/// source semantics, has the same value when coerced through the 1910/// target semantics. 1911/// 1912/// The value might be a vector of floats (or a complex number). 1913bool IsSameFloatAfterCast(const APValue &value, 1914 const llvm::fltSemantics &Src, 1915 const llvm::fltSemantics &Tgt) { 1916 if (value.isFloat()) 1917 return IsSameFloatAfterCast(value.getFloat(), Src, Tgt); 1918 1919 if (value.isVector()) { 1920 for (unsigned i = 0, e = value.getVectorLength(); i != e; ++i) 1921 if (!IsSameFloatAfterCast(value.getVectorElt(i), Src, Tgt)) 1922 return false; 1923 return true; 1924 } 1925 1926 assert(value.isComplexFloat()); 1927 return (IsSameFloatAfterCast(value.getComplexFloatReal(), Src, Tgt) && 1928 IsSameFloatAfterCast(value.getComplexFloatImag(), Src, Tgt)); 1929} 1930 1931} // end anonymous namespace 1932 1933/// \brief Implements -Wsign-compare. 1934/// 1935/// \param lex the left-hand expression 1936/// \param rex the right-hand expression 1937/// \param OpLoc the location of the joining operator 1938/// \param Equality whether this is an "equality-like" join, which 1939/// suppresses the warning in some cases 1940void Sema::CheckSignCompare(Expr *lex, Expr *rex, SourceLocation OpLoc, 1941 const PartialDiagnostic &PD, bool Equality) { 1942 // Don't warn if we're in an unevaluated context. 1943 if (ExprEvalContexts.back().Context == Unevaluated) 1944 return; 1945 1946 // If either expression is value-dependent, don't warn. We'll get another 1947 // chance at instantiation time. 1948 if (lex->isValueDependent() || rex->isValueDependent()) 1949 return; 1950 1951 QualType lt = lex->getType(), rt = rex->getType(); 1952 1953 // Only warn if both operands are integral. 1954 if (!lt->isIntegerType() || !rt->isIntegerType()) 1955 return; 1956 1957 // In C, the width of a bitfield determines its type, and the 1958 // declared type only contributes the signedness. This duplicates 1959 // the work that will later be done by UsualUnaryConversions. 1960 // Eventually, this check will be reorganized in a way that avoids 1961 // this duplication. 1962 if (!getLangOptions().CPlusPlus) { 1963 QualType tmp; 1964 tmp = Context.isPromotableBitField(lex); 1965 if (!tmp.isNull()) lt = tmp; 1966 tmp = Context.isPromotableBitField(rex); 1967 if (!tmp.isNull()) rt = tmp; 1968 } 1969 1970 // The rule is that the signed operand becomes unsigned, so isolate the 1971 // signed operand. 1972 Expr *signedOperand = lex, *unsignedOperand = rex; 1973 QualType signedType = lt, unsignedType = rt; 1974 if (lt->isSignedIntegerType()) { 1975 if (rt->isSignedIntegerType()) return; 1976 } else { 1977 if (!rt->isSignedIntegerType()) return; 1978 std::swap(signedOperand, unsignedOperand); 1979 std::swap(signedType, unsignedType); 1980 } 1981 1982 unsigned unsignedWidth = Context.getIntWidth(unsignedType); 1983 unsigned signedWidth = Context.getIntWidth(signedType); 1984 1985 // If the unsigned type is strictly smaller than the signed type, 1986 // then (1) the result type will be signed and (2) the unsigned 1987 // value will fit fully within the signed type, and thus the result 1988 // of the comparison will be exact. 1989 if (signedWidth > unsignedWidth) 1990 return; 1991 1992 // Otherwise, calculate the effective ranges. 1993 IntRange signedRange = GetExprRange(Context, signedOperand, signedWidth); 1994 IntRange unsignedRange = GetExprRange(Context, unsignedOperand, unsignedWidth); 1995 1996 // We should never be unable to prove that the unsigned operand is 1997 // non-negative. 1998 assert(unsignedRange.NonNegative && "unsigned range includes negative?"); 1999 2000 // If the signed operand is non-negative, then the signed->unsigned 2001 // conversion won't change it. 2002 if (signedRange.NonNegative) 2003 return; 2004 2005 // For (in)equality comparisons, if the unsigned operand is a 2006 // constant which cannot collide with a overflowed signed operand, 2007 // then reinterpreting the signed operand as unsigned will not 2008 // change the result of the comparison. 2009 if (Equality && unsignedRange.Width < unsignedWidth) 2010 return; 2011 2012 Diag(OpLoc, PD) 2013 << lt << rt << lex->getSourceRange() << rex->getSourceRange(); 2014} 2015 2016/// Diagnose an implicit cast; purely a helper for CheckImplicitConversion. 2017static void DiagnoseImpCast(Sema &S, Expr *E, QualType T, unsigned diag) { 2018 S.Diag(E->getExprLoc(), diag) << E->getType() << T << E->getSourceRange(); 2019} 2020 2021/// Implements -Wconversion. 2022void Sema::CheckImplicitConversion(Expr *E, QualType T) { 2023 // Don't diagnose in unevaluated contexts. 2024 if (ExprEvalContexts.back().Context == Sema::Unevaluated) 2025 return; 2026 2027 // Don't diagnose for value-dependent expressions. 2028 if (E->isValueDependent()) 2029 return; 2030 2031 const Type *Source = Context.getCanonicalType(E->getType()).getTypePtr(); 2032 const Type *Target = Context.getCanonicalType(T).getTypePtr(); 2033 2034 // Never diagnose implicit casts to bool. 2035 if (Target->isSpecificBuiltinType(BuiltinType::Bool)) 2036 return; 2037 2038 // Strip vector types. 2039 if (isa<VectorType>(Source)) { 2040 if (!isa<VectorType>(Target)) 2041 return DiagnoseImpCast(*this, E, T, diag::warn_impcast_vector_scalar); 2042 2043 Source = cast<VectorType>(Source)->getElementType().getTypePtr(); 2044 Target = cast<VectorType>(Target)->getElementType().getTypePtr(); 2045 } 2046 2047 // Strip complex types. 2048 if (isa<ComplexType>(Source)) { 2049 if (!isa<ComplexType>(Target)) 2050 return DiagnoseImpCast(*this, E, T, diag::warn_impcast_complex_scalar); 2051 2052 Source = cast<ComplexType>(Source)->getElementType().getTypePtr(); 2053 Target = cast<ComplexType>(Target)->getElementType().getTypePtr(); 2054 } 2055 2056 const BuiltinType *SourceBT = dyn_cast<BuiltinType>(Source); 2057 const BuiltinType *TargetBT = dyn_cast<BuiltinType>(Target); 2058 2059 // If the source is floating point... 2060 if (SourceBT && SourceBT->isFloatingPoint()) { 2061 // ...and the target is floating point... 2062 if (TargetBT && TargetBT->isFloatingPoint()) { 2063 // ...then warn if we're dropping FP rank. 2064 2065 // Builtin FP kinds are ordered by increasing FP rank. 2066 if (SourceBT->getKind() > TargetBT->getKind()) { 2067 // Don't warn about float constants that are precisely 2068 // representable in the target type. 2069 Expr::EvalResult result; 2070 if (E->Evaluate(result, Context)) { 2071 // Value might be a float, a float vector, or a float complex. 2072 if (IsSameFloatAfterCast(result.Val, 2073 Context.getFloatTypeSemantics(QualType(TargetBT, 0)), 2074 Context.getFloatTypeSemantics(QualType(SourceBT, 0)))) 2075 return; 2076 } 2077 2078 DiagnoseImpCast(*this, E, T, diag::warn_impcast_float_precision); 2079 } 2080 return; 2081 } 2082 2083 // If the target is integral, always warn. 2084 if ((TargetBT && TargetBT->isInteger())) 2085 // TODO: don't warn for integer values? 2086 return DiagnoseImpCast(*this, E, T, diag::warn_impcast_float_integer); 2087 2088 return; 2089 } 2090 2091 if (!Source->isIntegerType() || !Target->isIntegerType()) 2092 return; 2093 2094 IntRange SourceRange = GetExprRange(Context, E, Context.getIntWidth(E->getType())); 2095 IntRange TargetRange = IntRange::forCanonicalType(Context, Target); 2096 2097 // FIXME: also signed<->unsigned? 2098 2099 if (SourceRange.Width > TargetRange.Width) { 2100 // People want to build with -Wshorten-64-to-32 and not -Wconversion 2101 // and by god we'll let them. 2102 if (SourceRange.Width == 64 && TargetRange.Width == 32) 2103 return DiagnoseImpCast(*this, E, T, diag::warn_impcast_integer_64_32); 2104 return DiagnoseImpCast(*this, E, T, diag::warn_impcast_integer_precision); 2105 } 2106 2107 return; 2108} 2109 2110// MarkLive - Mark all the blocks reachable from e as live. Returns the total 2111// number of blocks just marked live. 2112static unsigned MarkLive(CFGBlock *e, llvm::BitVector &live) { 2113 unsigned count = 0; 2114 std::queue<CFGBlock*> workq; 2115 // Prep work queue 2116 live.set(e->getBlockID()); 2117 ++count; 2118 workq.push(e); 2119 // Solve 2120 while (!workq.empty()) { 2121 CFGBlock *item = workq.front(); 2122 workq.pop(); 2123 for (CFGBlock::succ_iterator I=item->succ_begin(), 2124 E=item->succ_end(); 2125 I != E; 2126 ++I) { 2127 if ((*I) && !live[(*I)->getBlockID()]) { 2128 live.set((*I)->getBlockID()); 2129 ++count; 2130 workq.push(*I); 2131 } 2132 } 2133 } 2134 return count; 2135} 2136 2137static SourceLocation GetUnreachableLoc(CFGBlock &b, SourceRange &R1, 2138 SourceRange &R2) { 2139 Stmt *S; 2140 unsigned sn = 0; 2141 R1 = R2 = SourceRange(); 2142 2143 top: 2144 if (sn < b.size()) 2145 S = b[sn].getStmt(); 2146 else if (b.getTerminator()) 2147 S = b.getTerminator(); 2148 else 2149 return SourceLocation(); 2150 2151 switch (S->getStmtClass()) { 2152 case Expr::BinaryOperatorClass: { 2153 BinaryOperator *BO = cast<BinaryOperator>(S); 2154 if (BO->getOpcode() == BinaryOperator::Comma) { 2155 if (sn+1 < b.size()) 2156 return b[sn+1].getStmt()->getLocStart(); 2157 CFGBlock *n = &b; 2158 while (1) { 2159 if (n->getTerminator()) 2160 return n->getTerminator()->getLocStart(); 2161 if (n->succ_size() != 1) 2162 return SourceLocation(); 2163 n = n[0].succ_begin()[0]; 2164 if (n->pred_size() != 1) 2165 return SourceLocation(); 2166 if (!n->empty()) 2167 return n[0][0].getStmt()->getLocStart(); 2168 } 2169 } 2170 R1 = BO->getLHS()->getSourceRange(); 2171 R2 = BO->getRHS()->getSourceRange(); 2172 return BO->getOperatorLoc(); 2173 } 2174 case Expr::UnaryOperatorClass: { 2175 const UnaryOperator *UO = cast<UnaryOperator>(S); 2176 R1 = UO->getSubExpr()->getSourceRange(); 2177 return UO->getOperatorLoc(); 2178 } 2179 case Expr::CompoundAssignOperatorClass: { 2180 const CompoundAssignOperator *CAO = cast<CompoundAssignOperator>(S); 2181 R1 = CAO->getLHS()->getSourceRange(); 2182 R2 = CAO->getRHS()->getSourceRange(); 2183 return CAO->getOperatorLoc(); 2184 } 2185 case Expr::ConditionalOperatorClass: { 2186 const ConditionalOperator *CO = cast<ConditionalOperator>(S); 2187 return CO->getQuestionLoc(); 2188 } 2189 case Expr::MemberExprClass: { 2190 const MemberExpr *ME = cast<MemberExpr>(S); 2191 R1 = ME->getSourceRange(); 2192 return ME->getMemberLoc(); 2193 } 2194 case Expr::ArraySubscriptExprClass: { 2195 const ArraySubscriptExpr *ASE = cast<ArraySubscriptExpr>(S); 2196 R1 = ASE->getLHS()->getSourceRange(); 2197 R2 = ASE->getRHS()->getSourceRange(); 2198 return ASE->getRBracketLoc(); 2199 } 2200 case Expr::CStyleCastExprClass: { 2201 const CStyleCastExpr *CSC = cast<CStyleCastExpr>(S); 2202 R1 = CSC->getSubExpr()->getSourceRange(); 2203 return CSC->getLParenLoc(); 2204 } 2205 case Expr::CXXFunctionalCastExprClass: { 2206 const CXXFunctionalCastExpr *CE = cast <CXXFunctionalCastExpr>(S); 2207 R1 = CE->getSubExpr()->getSourceRange(); 2208 return CE->getTypeBeginLoc(); 2209 } 2210 case Expr::ImplicitCastExprClass: 2211 ++sn; 2212 goto top; 2213 case Stmt::CXXTryStmtClass: { 2214 return cast<CXXTryStmt>(S)->getHandler(0)->getCatchLoc(); 2215 } 2216 default: ; 2217 } 2218 R1 = S->getSourceRange(); 2219 return S->getLocStart(); 2220} 2221 2222static SourceLocation MarkLiveTop(CFGBlock *e, llvm::BitVector &live, 2223 SourceManager &SM) { 2224 std::queue<CFGBlock*> workq; 2225 // Prep work queue 2226 workq.push(e); 2227 SourceRange R1, R2; 2228 SourceLocation top = GetUnreachableLoc(*e, R1, R2); 2229 bool FromMainFile = false; 2230 bool FromSystemHeader = false; 2231 bool TopValid = false; 2232 if (top.isValid()) { 2233 FromMainFile = SM.isFromMainFile(top); 2234 FromSystemHeader = SM.isInSystemHeader(top); 2235 TopValid = true; 2236 } 2237 // Solve 2238 while (!workq.empty()) { 2239 CFGBlock *item = workq.front(); 2240 workq.pop(); 2241 SourceLocation c = GetUnreachableLoc(*item, R1, R2); 2242 if (c.isValid() 2243 && (!TopValid 2244 || (SM.isFromMainFile(c) && !FromMainFile) 2245 || (FromSystemHeader && !SM.isInSystemHeader(c)) 2246 || SM.isBeforeInTranslationUnit(c, top))) { 2247 top = c; 2248 FromMainFile = SM.isFromMainFile(top); 2249 FromSystemHeader = SM.isInSystemHeader(top); 2250 } 2251 live.set(item->getBlockID()); 2252 for (CFGBlock::succ_iterator I=item->succ_begin(), 2253 E=item->succ_end(); 2254 I != E; 2255 ++I) { 2256 if ((*I) && !live[(*I)->getBlockID()]) { 2257 live.set((*I)->getBlockID()); 2258 workq.push(*I); 2259 } 2260 } 2261 } 2262 return top; 2263} 2264 2265static int LineCmp(const void *p1, const void *p2) { 2266 SourceLocation *Line1 = (SourceLocation *)p1; 2267 SourceLocation *Line2 = (SourceLocation *)p2; 2268 return !(*Line1 < *Line2); 2269} 2270 2271namespace { 2272 struct ErrLoc { 2273 SourceLocation Loc; 2274 SourceRange R1; 2275 SourceRange R2; 2276 ErrLoc(SourceLocation l, SourceRange r1, SourceRange r2) 2277 : Loc(l), R1(r1), R2(r2) { } 2278 }; 2279} 2280 2281/// CheckUnreachable - Check for unreachable code. 2282void Sema::CheckUnreachable(AnalysisContext &AC) { 2283 unsigned count; 2284 // We avoid checking when there are errors, as the CFG won't faithfully match 2285 // the user's code. 2286 if (getDiagnostics().hasErrorOccurred()) 2287 return; 2288 if (Diags.getDiagnosticLevel(diag::warn_unreachable) == Diagnostic::Ignored) 2289 return; 2290 2291 CFG *cfg = AC.getCFG(); 2292 if (cfg == 0) 2293 return; 2294 2295 llvm::BitVector live(cfg->getNumBlockIDs()); 2296 // Mark all live things first. 2297 count = MarkLive(&cfg->getEntry(), live); 2298 2299 if (count == cfg->getNumBlockIDs()) 2300 // If there are no dead blocks, we're done. 2301 return; 2302 2303 SourceRange R1, R2; 2304 2305 llvm::SmallVector<ErrLoc, 24> lines; 2306 bool AddEHEdges = AC.getAddEHEdges(); 2307 // First, give warnings for blocks with no predecessors, as they 2308 // can't be part of a loop. 2309 for (CFG::iterator I = cfg->begin(), E = cfg->end(); I != E; ++I) { 2310 CFGBlock &b = **I; 2311 if (!live[b.getBlockID()]) { 2312 if (b.pred_begin() == b.pred_end()) { 2313 if (!AddEHEdges && b.getTerminator() 2314 && isa<CXXTryStmt>(b.getTerminator())) { 2315 // When not adding EH edges from calls, catch clauses 2316 // can otherwise seem dead. Avoid noting them as dead. 2317 count += MarkLive(&b, live); 2318 continue; 2319 } 2320 SourceLocation c = GetUnreachableLoc(b, R1, R2); 2321 if (!c.isValid()) { 2322 // Blocks without a location can't produce a warning, so don't mark 2323 // reachable blocks from here as live. 2324 live.set(b.getBlockID()); 2325 ++count; 2326 continue; 2327 } 2328 lines.push_back(ErrLoc(c, R1, R2)); 2329 // Avoid excessive errors by marking everything reachable from here 2330 count += MarkLive(&b, live); 2331 } 2332 } 2333 } 2334 2335 if (count < cfg->getNumBlockIDs()) { 2336 // And then give warnings for the tops of loops. 2337 for (CFG::iterator I = cfg->begin(), E = cfg->end(); I != E; ++I) { 2338 CFGBlock &b = **I; 2339 if (!live[b.getBlockID()]) 2340 // Avoid excessive errors by marking everything reachable from here 2341 lines.push_back(ErrLoc(MarkLiveTop(&b, live, 2342 Context.getSourceManager()), 2343 SourceRange(), SourceRange())); 2344 } 2345 } 2346 2347 llvm::array_pod_sort(lines.begin(), lines.end(), LineCmp); 2348 for (llvm::SmallVector<ErrLoc, 24>::iterator I = lines.begin(), 2349 E = lines.end(); 2350 I != E; 2351 ++I) 2352 if (I->Loc.isValid()) 2353 Diag(I->Loc, diag::warn_unreachable) << I->R1 << I->R2; 2354} 2355 2356/// CheckFallThrough - Check that we don't fall off the end of a 2357/// Statement that should return a value. 2358/// 2359/// \returns AlwaysFallThrough iff we always fall off the end of the statement, 2360/// MaybeFallThrough iff we might or might not fall off the end, 2361/// NeverFallThroughOrReturn iff we never fall off the end of the statement or 2362/// return. We assume NeverFallThrough iff we never fall off the end of the 2363/// statement but we may return. We assume that functions not marked noreturn 2364/// will return. 2365Sema::ControlFlowKind Sema::CheckFallThrough(AnalysisContext &AC) { 2366 CFG *cfg = AC.getCFG(); 2367 if (cfg == 0) 2368 // FIXME: This should be NeverFallThrough 2369 return NeverFallThroughOrReturn; 2370 2371 // The CFG leaves in dead things, and we don't want the dead code paths to 2372 // confuse us, so we mark all live things first. 2373 std::queue<CFGBlock*> workq; 2374 llvm::BitVector live(cfg->getNumBlockIDs()); 2375 unsigned count = MarkLive(&cfg->getEntry(), live); 2376 2377 bool AddEHEdges = AC.getAddEHEdges(); 2378 if (!AddEHEdges && count != cfg->getNumBlockIDs()) 2379 // When there are things remaining dead, and we didn't add EH edges 2380 // from CallExprs to the catch clauses, we have to go back and 2381 // mark them as live. 2382 for (CFG::iterator I = cfg->begin(), E = cfg->end(); I != E; ++I) { 2383 CFGBlock &b = **I; 2384 if (!live[b.getBlockID()]) { 2385 if (b.pred_begin() == b.pred_end()) { 2386 if (b.getTerminator() && isa<CXXTryStmt>(b.getTerminator())) 2387 // When not adding EH edges from calls, catch clauses 2388 // can otherwise seem dead. Avoid noting them as dead. 2389 count += MarkLive(&b, live); 2390 continue; 2391 } 2392 } 2393 } 2394 2395 // Now we know what is live, we check the live precessors of the exit block 2396 // and look for fall through paths, being careful to ignore normal returns, 2397 // and exceptional paths. 2398 bool HasLiveReturn = false; 2399 bool HasFakeEdge = false; 2400 bool HasPlainEdge = false; 2401 bool HasAbnormalEdge = false; 2402 for (CFGBlock::pred_iterator I=cfg->getExit().pred_begin(), 2403 E = cfg->getExit().pred_end(); 2404 I != E; 2405 ++I) { 2406 CFGBlock& B = **I; 2407 if (!live[B.getBlockID()]) 2408 continue; 2409 if (B.size() == 0) { 2410 if (B.getTerminator() && isa<CXXTryStmt>(B.getTerminator())) { 2411 HasAbnormalEdge = true; 2412 continue; 2413 } 2414 2415 // A labeled empty statement, or the entry block... 2416 HasPlainEdge = true; 2417 continue; 2418 } 2419 Stmt *S = B[B.size()-1]; 2420 if (isa<ReturnStmt>(S)) { 2421 HasLiveReturn = true; 2422 continue; 2423 } 2424 if (isa<ObjCAtThrowStmt>(S)) { 2425 HasFakeEdge = true; 2426 continue; 2427 } 2428 if (isa<CXXThrowExpr>(S)) { 2429 HasFakeEdge = true; 2430 continue; 2431 } 2432 if (const AsmStmt *AS = dyn_cast<AsmStmt>(S)) { 2433 if (AS->isMSAsm()) { 2434 HasFakeEdge = true; 2435 HasLiveReturn = true; 2436 continue; 2437 } 2438 } 2439 if (isa<CXXTryStmt>(S)) { 2440 HasAbnormalEdge = true; 2441 continue; 2442 } 2443 2444 bool NoReturnEdge = false; 2445 if (CallExpr *C = dyn_cast<CallExpr>(S)) { 2446 if (B.succ_begin()[0] != &cfg->getExit()) { 2447 HasAbnormalEdge = true; 2448 continue; 2449 } 2450 Expr *CEE = C->getCallee()->IgnoreParenCasts(); 2451 if (CEE->getType().getNoReturnAttr()) { 2452 NoReturnEdge = true; 2453 HasFakeEdge = true; 2454 } else if (DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(CEE)) { 2455 ValueDecl *VD = DRE->getDecl(); 2456 if (VD->hasAttr<NoReturnAttr>()) { 2457 NoReturnEdge = true; 2458 HasFakeEdge = true; 2459 } 2460 } 2461 } 2462 // FIXME: Add noreturn message sends. 2463 if (NoReturnEdge == false) 2464 HasPlainEdge = true; 2465 } 2466 if (!HasPlainEdge) { 2467 if (HasLiveReturn) 2468 return NeverFallThrough; 2469 return NeverFallThroughOrReturn; 2470 } 2471 if (HasAbnormalEdge || HasFakeEdge || HasLiveReturn) 2472 return MaybeFallThrough; 2473 // This says AlwaysFallThrough for calls to functions that are not marked 2474 // noreturn, that don't return. If people would like this warning to be more 2475 // accurate, such functions should be marked as noreturn. 2476 return AlwaysFallThrough; 2477} 2478 2479/// CheckFallThroughForFunctionDef - Check that we don't fall off the end of a 2480/// function that should return a value. Check that we don't fall off the end 2481/// of a noreturn function. We assume that functions and blocks not marked 2482/// noreturn will return. 2483void Sema::CheckFallThroughForFunctionDef(Decl *D, Stmt *Body, 2484 AnalysisContext &AC) { 2485 // FIXME: Would be nice if we had a better way to control cascading errors, 2486 // but for now, avoid them. The problem is that when Parse sees: 2487 // int foo() { return a; } 2488 // The return is eaten and the Sema code sees just: 2489 // int foo() { } 2490 // which this code would then warn about. 2491 if (getDiagnostics().hasErrorOccurred()) 2492 return; 2493 2494 bool ReturnsVoid = false; 2495 bool HasNoReturn = false; 2496 if (FunctionDecl *FD = dyn_cast<FunctionDecl>(D)) { 2497 // If the result type of the function is a dependent type, we don't know 2498 // whether it will be void or not, so don't 2499 if (FD->getResultType()->isDependentType()) 2500 return; 2501 if (FD->getResultType()->isVoidType()) 2502 ReturnsVoid = true; 2503 if (FD->hasAttr<NoReturnAttr>()) 2504 HasNoReturn = true; 2505 } else if (ObjCMethodDecl *MD = dyn_cast<ObjCMethodDecl>(D)) { 2506 if (MD->getResultType()->isVoidType()) 2507 ReturnsVoid = true; 2508 if (MD->hasAttr<NoReturnAttr>()) 2509 HasNoReturn = true; 2510 } 2511 2512 // Short circuit for compilation speed. 2513 if ((Diags.getDiagnosticLevel(diag::warn_maybe_falloff_nonvoid_function) 2514 == Diagnostic::Ignored || ReturnsVoid) 2515 && (Diags.getDiagnosticLevel(diag::warn_noreturn_function_has_return_expr) 2516 == Diagnostic::Ignored || !HasNoReturn) 2517 && (Diags.getDiagnosticLevel(diag::warn_suggest_noreturn_block) 2518 == Diagnostic::Ignored || !ReturnsVoid)) 2519 return; 2520 // FIXME: Function try block 2521 if (CompoundStmt *Compound = dyn_cast<CompoundStmt>(Body)) { 2522 switch (CheckFallThrough(AC)) { 2523 case MaybeFallThrough: 2524 if (HasNoReturn) 2525 Diag(Compound->getRBracLoc(), diag::warn_falloff_noreturn_function); 2526 else if (!ReturnsVoid) 2527 Diag(Compound->getRBracLoc(),diag::warn_maybe_falloff_nonvoid_function); 2528 break; 2529 case AlwaysFallThrough: 2530 if (HasNoReturn) 2531 Diag(Compound->getRBracLoc(), diag::warn_falloff_noreturn_function); 2532 else if (!ReturnsVoid) 2533 Diag(Compound->getRBracLoc(), diag::warn_falloff_nonvoid_function); 2534 break; 2535 case NeverFallThroughOrReturn: 2536 if (ReturnsVoid && !HasNoReturn) 2537 Diag(Compound->getLBracLoc(), diag::warn_suggest_noreturn_function); 2538 break; 2539 case NeverFallThrough: 2540 break; 2541 } 2542 } 2543} 2544 2545/// CheckFallThroughForBlock - Check that we don't fall off the end of a block 2546/// that should return a value. Check that we don't fall off the end of a 2547/// noreturn block. We assume that functions and blocks not marked noreturn 2548/// will return. 2549void Sema::CheckFallThroughForBlock(QualType BlockTy, Stmt *Body, 2550 AnalysisContext &AC) { 2551 // FIXME: Would be nice if we had a better way to control cascading errors, 2552 // but for now, avoid them. The problem is that when Parse sees: 2553 // int foo() { return a; } 2554 // The return is eaten and the Sema code sees just: 2555 // int foo() { } 2556 // which this code would then warn about. 2557 if (getDiagnostics().hasErrorOccurred()) 2558 return; 2559 bool ReturnsVoid = false; 2560 bool HasNoReturn = false; 2561 if (const FunctionType *FT =BlockTy->getPointeeType()->getAs<FunctionType>()){ 2562 if (FT->getResultType()->isVoidType()) 2563 ReturnsVoid = true; 2564 if (FT->getNoReturnAttr()) 2565 HasNoReturn = true; 2566 } 2567 2568 // Short circuit for compilation speed. 2569 if (ReturnsVoid 2570 && !HasNoReturn 2571 && (Diags.getDiagnosticLevel(diag::warn_suggest_noreturn_block) 2572 == Diagnostic::Ignored || !ReturnsVoid)) 2573 return; 2574 // FIXME: Funtion try block 2575 if (CompoundStmt *Compound = dyn_cast<CompoundStmt>(Body)) { 2576 switch (CheckFallThrough(AC)) { 2577 case MaybeFallThrough: 2578 if (HasNoReturn) 2579 Diag(Compound->getRBracLoc(), diag::err_noreturn_block_has_return_expr); 2580 else if (!ReturnsVoid) 2581 Diag(Compound->getRBracLoc(), diag::err_maybe_falloff_nonvoid_block); 2582 break; 2583 case AlwaysFallThrough: 2584 if (HasNoReturn) 2585 Diag(Compound->getRBracLoc(), diag::err_noreturn_block_has_return_expr); 2586 else if (!ReturnsVoid) 2587 Diag(Compound->getRBracLoc(), diag::err_falloff_nonvoid_block); 2588 break; 2589 case NeverFallThroughOrReturn: 2590 if (ReturnsVoid) 2591 Diag(Compound->getLBracLoc(), diag::warn_suggest_noreturn_block); 2592 break; 2593 case NeverFallThrough: 2594 break; 2595 } 2596 } 2597} 2598 2599/// CheckParmsForFunctionDef - Check that the parameters of the given 2600/// function are appropriate for the definition of a function. This 2601/// takes care of any checks that cannot be performed on the 2602/// declaration itself, e.g., that the types of each of the function 2603/// parameters are complete. 2604bool Sema::CheckParmsForFunctionDef(FunctionDecl *FD) { 2605 bool HasInvalidParm = false; 2606 for (unsigned p = 0, NumParams = FD->getNumParams(); p < NumParams; ++p) { 2607 ParmVarDecl *Param = FD->getParamDecl(p); 2608 2609 // C99 6.7.5.3p4: the parameters in a parameter type list in a 2610 // function declarator that is part of a function definition of 2611 // that function shall not have incomplete type. 2612 // 2613 // This is also C++ [dcl.fct]p6. 2614 if (!Param->isInvalidDecl() && 2615 RequireCompleteType(Param->getLocation(), Param->getType(), 2616 diag::err_typecheck_decl_incomplete_type)) { 2617 Param->setInvalidDecl(); 2618 HasInvalidParm = true; 2619 } 2620 2621 // C99 6.9.1p5: If the declarator includes a parameter type list, the 2622 // declaration of each parameter shall include an identifier. 2623 if (Param->getIdentifier() == 0 && 2624 !Param->isImplicit() && 2625 !getLangOptions().CPlusPlus) 2626 Diag(Param->getLocation(), diag::err_parameter_name_omitted); 2627 2628 // C99 6.7.5.3p12: 2629 // If the function declarator is not part of a definition of that 2630 // function, parameters may have incomplete type and may use the [*] 2631 // notation in their sequences of declarator specifiers to specify 2632 // variable length array types. 2633 QualType PType = Param->getOriginalType(); 2634 if (const ArrayType *AT = Context.getAsArrayType(PType)) { 2635 if (AT->getSizeModifier() == ArrayType::Star) { 2636 // FIXME: This diagnosic should point the the '[*]' if source-location 2637 // information is added for it. 2638 Diag(Param->getLocation(), diag::err_array_star_in_function_definition); 2639 } 2640 } 2641 2642 if (getLangOptions().CPlusPlus) 2643 if (const RecordType *RT = Param->getType()->getAs<RecordType>()) 2644 FinalizeVarWithDestructor(Param, RT); 2645 } 2646 2647 return HasInvalidParm; 2648} 2649