SemaChecking.cpp revision 323ed74658bc8375278eabf074b4777458376540
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/Analyses/PrintfFormatString.h" 17#include "clang/AST/ASTContext.h" 18#include "clang/AST/CharUnits.h" 19#include "clang/AST/DeclObjC.h" 20#include "clang/AST/ExprCXX.h" 21#include "clang/AST/ExprObjC.h" 22#include "clang/AST/DeclObjC.h" 23#include "clang/AST/StmtCXX.h" 24#include "clang/AST/StmtObjC.h" 25#include "clang/Lex/LiteralSupport.h" 26#include "clang/Lex/Preprocessor.h" 27#include "llvm/ADT/BitVector.h" 28#include "llvm/ADT/STLExtras.h" 29#include "clang/Basic/TargetBuiltins.h" 30#include <limits> 31using namespace clang; 32 33/// getLocationOfStringLiteralByte - Return a source location that points to the 34/// specified byte of the specified string literal. 35/// 36/// Strings are amazingly complex. They can be formed from multiple tokens and 37/// can have escape sequences in them in addition to the usual trigraph and 38/// escaped newline business. This routine handles this complexity. 39/// 40SourceLocation Sema::getLocationOfStringLiteralByte(const StringLiteral *SL, 41 unsigned ByteNo) const { 42 assert(!SL->isWide() && "This doesn't work for wide strings yet"); 43 44 // Loop over all of the tokens in this string until we find the one that 45 // contains the byte we're looking for. 46 unsigned TokNo = 0; 47 while (1) { 48 assert(TokNo < SL->getNumConcatenated() && "Invalid byte number!"); 49 SourceLocation StrTokLoc = SL->getStrTokenLoc(TokNo); 50 51 // Get the spelling of the string so that we can get the data that makes up 52 // the string literal, not the identifier for the macro it is potentially 53 // expanded through. 54 SourceLocation StrTokSpellingLoc = SourceMgr.getSpellingLoc(StrTokLoc); 55 56 // Re-lex the token to get its length and original spelling. 57 std::pair<FileID, unsigned> LocInfo = 58 SourceMgr.getDecomposedLoc(StrTokSpellingLoc); 59 bool Invalid = false; 60 llvm::StringRef Buffer = SourceMgr.getBufferData(LocInfo.first, &Invalid); 61 if (Invalid) 62 return StrTokSpellingLoc; 63 64 const char *StrData = Buffer.data()+LocInfo.second; 65 66 // Create a langops struct and enable trigraphs. This is sufficient for 67 // relexing tokens. 68 LangOptions LangOpts; 69 LangOpts.Trigraphs = true; 70 71 // Create a lexer starting at the beginning of this token. 72 Lexer TheLexer(StrTokSpellingLoc, LangOpts, Buffer.begin(), StrData, 73 Buffer.end()); 74 Token TheTok; 75 TheLexer.LexFromRawLexer(TheTok); 76 77 // Use the StringLiteralParser to compute the length of the string in bytes. 78 StringLiteralParser SLP(&TheTok, 1, PP); 79 unsigned TokNumBytes = SLP.GetStringLength(); 80 81 // If the byte is in this token, return the location of the byte. 82 if (ByteNo < TokNumBytes || 83 (ByteNo == TokNumBytes && TokNo == SL->getNumConcatenated())) { 84 unsigned Offset = 85 StringLiteralParser::getOffsetOfStringByte(TheTok, ByteNo, PP); 86 87 // Now that we know the offset of the token in the spelling, use the 88 // preprocessor to get the offset in the original source. 89 return PP.AdvanceToTokenCharacter(StrTokLoc, Offset); 90 } 91 92 // Move to the next string token. 93 ++TokNo; 94 ByteNo -= TokNumBytes; 95 } 96} 97 98/// CheckablePrintfAttr - does a function call have a "printf" attribute 99/// and arguments that merit checking? 100bool Sema::CheckablePrintfAttr(const FormatAttr *Format, CallExpr *TheCall) { 101 if (Format->getType() == "printf") return true; 102 if (Format->getType() == "printf0") { 103 // printf0 allows null "format" string; if so don't check format/args 104 unsigned format_idx = Format->getFormatIdx() - 1; 105 // Does the index refer to the implicit object argument? 106 if (isa<CXXMemberCallExpr>(TheCall)) { 107 if (format_idx == 0) 108 return false; 109 --format_idx; 110 } 111 if (format_idx < TheCall->getNumArgs()) { 112 Expr *Format = TheCall->getArg(format_idx)->IgnoreParenCasts(); 113 if (!Format->isNullPointerConstant(Context, 114 Expr::NPC_ValueDependentIsNull)) 115 return true; 116 } 117 } 118 return false; 119} 120 121Action::OwningExprResult 122Sema::CheckBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) { 123 OwningExprResult TheCallResult(Owned(TheCall)); 124 125 switch (BuiltinID) { 126 case Builtin::BI__builtin___CFStringMakeConstantString: 127 assert(TheCall->getNumArgs() == 1 && 128 "Wrong # arguments to builtin CFStringMakeConstantString"); 129 if (CheckObjCString(TheCall->getArg(0))) 130 return ExprError(); 131 break; 132 case Builtin::BI__builtin_stdarg_start: 133 case Builtin::BI__builtin_va_start: 134 if (SemaBuiltinVAStart(TheCall)) 135 return ExprError(); 136 break; 137 case Builtin::BI__builtin_isgreater: 138 case Builtin::BI__builtin_isgreaterequal: 139 case Builtin::BI__builtin_isless: 140 case Builtin::BI__builtin_islessequal: 141 case Builtin::BI__builtin_islessgreater: 142 case Builtin::BI__builtin_isunordered: 143 if (SemaBuiltinUnorderedCompare(TheCall)) 144 return ExprError(); 145 break; 146 case Builtin::BI__builtin_fpclassify: 147 if (SemaBuiltinFPClassification(TheCall, 6)) 148 return ExprError(); 149 break; 150 case Builtin::BI__builtin_isfinite: 151 case Builtin::BI__builtin_isinf: 152 case Builtin::BI__builtin_isinf_sign: 153 case Builtin::BI__builtin_isnan: 154 case Builtin::BI__builtin_isnormal: 155 if (SemaBuiltinFPClassification(TheCall, 1)) 156 return ExprError(); 157 break; 158 case Builtin::BI__builtin_return_address: 159 case Builtin::BI__builtin_frame_address: { 160 llvm::APSInt Result; 161 if (SemaBuiltinConstantArg(TheCall, 0, Result)) 162 return ExprError(); 163 break; 164 } 165 case Builtin::BI__builtin_eh_return_data_regno: { 166 llvm::APSInt Result; 167 if (SemaBuiltinConstantArg(TheCall, 0, Result)) 168 return ExprError(); 169 break; 170 } 171 case Builtin::BI__builtin_shufflevector: 172 return SemaBuiltinShuffleVector(TheCall); 173 // TheCall will be freed by the smart pointer here, but that's fine, since 174 // SemaBuiltinShuffleVector guts it, but then doesn't release it. 175 case Builtin::BI__builtin_prefetch: 176 if (SemaBuiltinPrefetch(TheCall)) 177 return ExprError(); 178 break; 179 case Builtin::BI__builtin_object_size: 180 if (SemaBuiltinObjectSize(TheCall)) 181 return ExprError(); 182 break; 183 case Builtin::BI__builtin_longjmp: 184 if (SemaBuiltinLongjmp(TheCall)) 185 return ExprError(); 186 break; 187 case Builtin::BI__sync_fetch_and_add: 188 case Builtin::BI__sync_fetch_and_sub: 189 case Builtin::BI__sync_fetch_and_or: 190 case Builtin::BI__sync_fetch_and_and: 191 case Builtin::BI__sync_fetch_and_xor: 192 case Builtin::BI__sync_add_and_fetch: 193 case Builtin::BI__sync_sub_and_fetch: 194 case Builtin::BI__sync_and_and_fetch: 195 case Builtin::BI__sync_or_and_fetch: 196 case Builtin::BI__sync_xor_and_fetch: 197 case Builtin::BI__sync_val_compare_and_swap: 198 case Builtin::BI__sync_bool_compare_and_swap: 199 case Builtin::BI__sync_lock_test_and_set: 200 case Builtin::BI__sync_lock_release: 201 if (SemaBuiltinAtomicOverloaded(TheCall)) 202 return ExprError(); 203 break; 204 205 // Target specific builtins start here. 206 case X86::BI__builtin_ia32_palignr128: 207 case X86::BI__builtin_ia32_palignr: { 208 llvm::APSInt Result; 209 if (SemaBuiltinConstantArg(TheCall, 2, Result)) 210 return ExprError(); 211 break; 212 } 213 } 214 215 return move(TheCallResult); 216} 217 218/// CheckFunctionCall - Check a direct function call for various correctness 219/// and safety properties not strictly enforced by the C type system. 220bool Sema::CheckFunctionCall(FunctionDecl *FDecl, CallExpr *TheCall) { 221 // Get the IdentifierInfo* for the called function. 222 IdentifierInfo *FnInfo = FDecl->getIdentifier(); 223 224 // None of the checks below are needed for functions that don't have 225 // simple names (e.g., C++ conversion functions). 226 if (!FnInfo) 227 return false; 228 229 // FIXME: This mechanism should be abstracted to be less fragile and 230 // more efficient. For example, just map function ids to custom 231 // handlers. 232 233 // Printf checking. 234 if (const FormatAttr *Format = FDecl->getAttr<FormatAttr>()) { 235 if (CheckablePrintfAttr(Format, TheCall)) { 236 bool HasVAListArg = Format->getFirstArg() == 0; 237 CheckPrintfArguments(TheCall, HasVAListArg, Format->getFormatIdx() - 1, 238 HasVAListArg ? 0 : Format->getFirstArg() - 1); 239 } 240 } 241 242 for (const NonNullAttr *NonNull = FDecl->getAttr<NonNullAttr>(); NonNull; 243 NonNull = NonNull->getNext<NonNullAttr>()) 244 CheckNonNullArguments(NonNull, TheCall); 245 246 return false; 247} 248 249bool Sema::CheckBlockCall(NamedDecl *NDecl, CallExpr *TheCall) { 250 // Printf checking. 251 const FormatAttr *Format = NDecl->getAttr<FormatAttr>(); 252 if (!Format) 253 return false; 254 255 const VarDecl *V = dyn_cast<VarDecl>(NDecl); 256 if (!V) 257 return false; 258 259 QualType Ty = V->getType(); 260 if (!Ty->isBlockPointerType()) 261 return false; 262 263 if (!CheckablePrintfAttr(Format, TheCall)) 264 return false; 265 266 bool HasVAListArg = Format->getFirstArg() == 0; 267 CheckPrintfArguments(TheCall, HasVAListArg, Format->getFormatIdx() - 1, 268 HasVAListArg ? 0 : Format->getFirstArg() - 1); 269 270 return false; 271} 272 273/// SemaBuiltinAtomicOverloaded - We have a call to a function like 274/// __sync_fetch_and_add, which is an overloaded function based on the pointer 275/// type of its first argument. The main ActOnCallExpr routines have already 276/// promoted the types of arguments because all of these calls are prototyped as 277/// void(...). 278/// 279/// This function goes through and does final semantic checking for these 280/// builtins, 281bool Sema::SemaBuiltinAtomicOverloaded(CallExpr *TheCall) { 282 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 283 FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl()); 284 285 // Ensure that we have at least one argument to do type inference from. 286 if (TheCall->getNumArgs() < 1) 287 return Diag(TheCall->getLocEnd(), 288 diag::err_typecheck_call_too_few_args_at_least) 289 << 0 << 1 << TheCall->getNumArgs() 290 << TheCall->getCallee()->getSourceRange(); 291 292 // Inspect the first argument of the atomic builtin. This should always be 293 // a pointer type, whose element is an integral scalar or pointer type. 294 // Because it is a pointer type, we don't have to worry about any implicit 295 // casts here. 296 Expr *FirstArg = TheCall->getArg(0); 297 if (!FirstArg->getType()->isPointerType()) 298 return Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer) 299 << FirstArg->getType() << FirstArg->getSourceRange(); 300 301 QualType ValType = FirstArg->getType()->getAs<PointerType>()->getPointeeType(); 302 if (!ValType->isIntegerType() && !ValType->isPointerType() && 303 !ValType->isBlockPointerType()) 304 return Diag(DRE->getLocStart(), 305 diag::err_atomic_builtin_must_be_pointer_intptr) 306 << FirstArg->getType() << FirstArg->getSourceRange(); 307 308 // We need to figure out which concrete builtin this maps onto. For example, 309 // __sync_fetch_and_add with a 2 byte object turns into 310 // __sync_fetch_and_add_2. 311#define BUILTIN_ROW(x) \ 312 { Builtin::BI##x##_1, Builtin::BI##x##_2, Builtin::BI##x##_4, \ 313 Builtin::BI##x##_8, Builtin::BI##x##_16 } 314 315 static const unsigned BuiltinIndices[][5] = { 316 BUILTIN_ROW(__sync_fetch_and_add), 317 BUILTIN_ROW(__sync_fetch_and_sub), 318 BUILTIN_ROW(__sync_fetch_and_or), 319 BUILTIN_ROW(__sync_fetch_and_and), 320 BUILTIN_ROW(__sync_fetch_and_xor), 321 322 BUILTIN_ROW(__sync_add_and_fetch), 323 BUILTIN_ROW(__sync_sub_and_fetch), 324 BUILTIN_ROW(__sync_and_and_fetch), 325 BUILTIN_ROW(__sync_or_and_fetch), 326 BUILTIN_ROW(__sync_xor_and_fetch), 327 328 BUILTIN_ROW(__sync_val_compare_and_swap), 329 BUILTIN_ROW(__sync_bool_compare_and_swap), 330 BUILTIN_ROW(__sync_lock_test_and_set), 331 BUILTIN_ROW(__sync_lock_release) 332 }; 333#undef BUILTIN_ROW 334 335 // Determine the index of the size. 336 unsigned SizeIndex; 337 switch (Context.getTypeSizeInChars(ValType).getQuantity()) { 338 case 1: SizeIndex = 0; break; 339 case 2: SizeIndex = 1; break; 340 case 4: SizeIndex = 2; break; 341 case 8: SizeIndex = 3; break; 342 case 16: SizeIndex = 4; break; 343 default: 344 return Diag(DRE->getLocStart(), diag::err_atomic_builtin_pointer_size) 345 << FirstArg->getType() << FirstArg->getSourceRange(); 346 } 347 348 // Each of these builtins has one pointer argument, followed by some number of 349 // values (0, 1 or 2) followed by a potentially empty varags list of stuff 350 // that we ignore. Find out which row of BuiltinIndices to read from as well 351 // as the number of fixed args. 352 unsigned BuiltinID = FDecl->getBuiltinID(); 353 unsigned BuiltinIndex, NumFixed = 1; 354 switch (BuiltinID) { 355 default: assert(0 && "Unknown overloaded atomic builtin!"); 356 case Builtin::BI__sync_fetch_and_add: BuiltinIndex = 0; break; 357 case Builtin::BI__sync_fetch_and_sub: BuiltinIndex = 1; break; 358 case Builtin::BI__sync_fetch_and_or: BuiltinIndex = 2; break; 359 case Builtin::BI__sync_fetch_and_and: BuiltinIndex = 3; break; 360 case Builtin::BI__sync_fetch_and_xor: BuiltinIndex = 4; break; 361 362 case Builtin::BI__sync_add_and_fetch: BuiltinIndex = 5; break; 363 case Builtin::BI__sync_sub_and_fetch: BuiltinIndex = 6; break; 364 case Builtin::BI__sync_and_and_fetch: BuiltinIndex = 7; break; 365 case Builtin::BI__sync_or_and_fetch: BuiltinIndex = 8; break; 366 case Builtin::BI__sync_xor_and_fetch: BuiltinIndex = 9; break; 367 368 case Builtin::BI__sync_val_compare_and_swap: 369 BuiltinIndex = 10; 370 NumFixed = 2; 371 break; 372 case Builtin::BI__sync_bool_compare_and_swap: 373 BuiltinIndex = 11; 374 NumFixed = 2; 375 break; 376 case Builtin::BI__sync_lock_test_and_set: BuiltinIndex = 12; break; 377 case Builtin::BI__sync_lock_release: 378 BuiltinIndex = 13; 379 NumFixed = 0; 380 break; 381 } 382 383 // Now that we know how many fixed arguments we expect, first check that we 384 // have at least that many. 385 if (TheCall->getNumArgs() < 1+NumFixed) 386 return Diag(TheCall->getLocEnd(), 387 diag::err_typecheck_call_too_few_args_at_least) 388 << 0 << 1+NumFixed << TheCall->getNumArgs() 389 << TheCall->getCallee()->getSourceRange(); 390 391 392 // Get the decl for the concrete builtin from this, we can tell what the 393 // concrete integer type we should convert to is. 394 unsigned NewBuiltinID = BuiltinIndices[BuiltinIndex][SizeIndex]; 395 const char *NewBuiltinName = Context.BuiltinInfo.GetName(NewBuiltinID); 396 IdentifierInfo *NewBuiltinII = PP.getIdentifierInfo(NewBuiltinName); 397 FunctionDecl *NewBuiltinDecl = 398 cast<FunctionDecl>(LazilyCreateBuiltin(NewBuiltinII, NewBuiltinID, 399 TUScope, false, DRE->getLocStart())); 400 const FunctionProtoType *BuiltinFT = 401 NewBuiltinDecl->getType()->getAs<FunctionProtoType>(); 402 ValType = BuiltinFT->getArgType(0)->getAs<PointerType>()->getPointeeType(); 403 404 // If the first type needs to be converted (e.g. void** -> int*), do it now. 405 if (BuiltinFT->getArgType(0) != FirstArg->getType()) { 406 ImpCastExprToType(FirstArg, BuiltinFT->getArgType(0), CastExpr::CK_BitCast); 407 TheCall->setArg(0, FirstArg); 408 } 409 410 // Next, walk the valid ones promoting to the right type. 411 for (unsigned i = 0; i != NumFixed; ++i) { 412 Expr *Arg = TheCall->getArg(i+1); 413 414 // If the argument is an implicit cast, then there was a promotion due to 415 // "...", just remove it now. 416 if (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(Arg)) { 417 Arg = ICE->getSubExpr(); 418 ICE->setSubExpr(0); 419 ICE->Destroy(Context); 420 TheCall->setArg(i+1, Arg); 421 } 422 423 // GCC does an implicit conversion to the pointer or integer ValType. This 424 // can fail in some cases (1i -> int**), check for this error case now. 425 CastExpr::CastKind Kind = CastExpr::CK_Unknown; 426 CXXBaseSpecifierArray BasePath; 427 if (CheckCastTypes(Arg->getSourceRange(), ValType, Arg, Kind, BasePath)) 428 return true; 429 430 // Okay, we have something that *can* be converted to the right type. Check 431 // to see if there is a potentially weird extension going on here. This can 432 // happen when you do an atomic operation on something like an char* and 433 // pass in 42. The 42 gets converted to char. This is even more strange 434 // for things like 45.123 -> char, etc. 435 // FIXME: Do this check. 436 ImpCastExprToType(Arg, ValType, Kind); 437 TheCall->setArg(i+1, Arg); 438 } 439 440 // Switch the DeclRefExpr to refer to the new decl. 441 DRE->setDecl(NewBuiltinDecl); 442 DRE->setType(NewBuiltinDecl->getType()); 443 444 // Set the callee in the CallExpr. 445 // FIXME: This leaks the original parens and implicit casts. 446 Expr *PromotedCall = DRE; 447 UsualUnaryConversions(PromotedCall); 448 TheCall->setCallee(PromotedCall); 449 450 451 // Change the result type of the call to match the result type of the decl. 452 TheCall->setType(NewBuiltinDecl->getResultType()); 453 return false; 454} 455 456 457/// CheckObjCString - Checks that the argument to the builtin 458/// CFString constructor is correct 459/// FIXME: GCC currently emits the following warning: 460/// "warning: input conversion stopped due to an input byte that does not 461/// belong to the input codeset UTF-8" 462/// Note: It might also make sense to do the UTF-16 conversion here (would 463/// simplify the backend). 464bool Sema::CheckObjCString(Expr *Arg) { 465 Arg = Arg->IgnoreParenCasts(); 466 StringLiteral *Literal = dyn_cast<StringLiteral>(Arg); 467 468 if (!Literal || Literal->isWide()) { 469 Diag(Arg->getLocStart(), diag::err_cfstring_literal_not_string_constant) 470 << Arg->getSourceRange(); 471 return true; 472 } 473 474 const char *Data = Literal->getStrData(); 475 unsigned Length = Literal->getByteLength(); 476 477 for (unsigned i = 0; i < Length; ++i) { 478 if (!Data[i]) { 479 Diag(getLocationOfStringLiteralByte(Literal, i), 480 diag::warn_cfstring_literal_contains_nul_character) 481 << Arg->getSourceRange(); 482 break; 483 } 484 } 485 486 return false; 487} 488 489/// SemaBuiltinVAStart - Check the arguments to __builtin_va_start for validity. 490/// Emit an error and return true on failure, return false on success. 491bool Sema::SemaBuiltinVAStart(CallExpr *TheCall) { 492 Expr *Fn = TheCall->getCallee(); 493 if (TheCall->getNumArgs() > 2) { 494 Diag(TheCall->getArg(2)->getLocStart(), 495 diag::err_typecheck_call_too_many_args) 496 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 497 << Fn->getSourceRange() 498 << SourceRange(TheCall->getArg(2)->getLocStart(), 499 (*(TheCall->arg_end()-1))->getLocEnd()); 500 return true; 501 } 502 503 if (TheCall->getNumArgs() < 2) { 504 return Diag(TheCall->getLocEnd(), 505 diag::err_typecheck_call_too_few_args_at_least) 506 << 0 /*function call*/ << 2 << TheCall->getNumArgs(); 507 } 508 509 // Determine whether the current function is variadic or not. 510 BlockScopeInfo *CurBlock = getCurBlock(); 511 bool isVariadic; 512 if (CurBlock) 513 isVariadic = CurBlock->isVariadic; 514 else if (FunctionDecl *FD = getCurFunctionDecl()) 515 isVariadic = FD->isVariadic(); 516 else 517 isVariadic = getCurMethodDecl()->isVariadic(); 518 519 if (!isVariadic) { 520 Diag(Fn->getLocStart(), diag::err_va_start_used_in_non_variadic_function); 521 return true; 522 } 523 524 // Verify that the second argument to the builtin is the last argument of the 525 // current function or method. 526 bool SecondArgIsLastNamedArgument = false; 527 const Expr *Arg = TheCall->getArg(1)->IgnoreParenCasts(); 528 529 if (const DeclRefExpr *DR = dyn_cast<DeclRefExpr>(Arg)) { 530 if (const ParmVarDecl *PV = dyn_cast<ParmVarDecl>(DR->getDecl())) { 531 // FIXME: This isn't correct for methods (results in bogus warning). 532 // Get the last formal in the current function. 533 const ParmVarDecl *LastArg; 534 if (CurBlock) 535 LastArg = *(CurBlock->TheDecl->param_end()-1); 536 else if (FunctionDecl *FD = getCurFunctionDecl()) 537 LastArg = *(FD->param_end()-1); 538 else 539 LastArg = *(getCurMethodDecl()->param_end()-1); 540 SecondArgIsLastNamedArgument = PV == LastArg; 541 } 542 } 543 544 if (!SecondArgIsLastNamedArgument) 545 Diag(TheCall->getArg(1)->getLocStart(), 546 diag::warn_second_parameter_of_va_start_not_last_named_argument); 547 return false; 548} 549 550/// SemaBuiltinUnorderedCompare - Handle functions like __builtin_isgreater and 551/// friends. This is declared to take (...), so we have to check everything. 552bool Sema::SemaBuiltinUnorderedCompare(CallExpr *TheCall) { 553 if (TheCall->getNumArgs() < 2) 554 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 555 << 0 << 2 << TheCall->getNumArgs()/*function call*/; 556 if (TheCall->getNumArgs() > 2) 557 return Diag(TheCall->getArg(2)->getLocStart(), 558 diag::err_typecheck_call_too_many_args) 559 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 560 << SourceRange(TheCall->getArg(2)->getLocStart(), 561 (*(TheCall->arg_end()-1))->getLocEnd()); 562 563 Expr *OrigArg0 = TheCall->getArg(0); 564 Expr *OrigArg1 = TheCall->getArg(1); 565 566 // Do standard promotions between the two arguments, returning their common 567 // type. 568 QualType Res = UsualArithmeticConversions(OrigArg0, OrigArg1, false); 569 570 // Make sure any conversions are pushed back into the call; this is 571 // type safe since unordered compare builtins are declared as "_Bool 572 // foo(...)". 573 TheCall->setArg(0, OrigArg0); 574 TheCall->setArg(1, OrigArg1); 575 576 if (OrigArg0->isTypeDependent() || OrigArg1->isTypeDependent()) 577 return false; 578 579 // If the common type isn't a real floating type, then the arguments were 580 // invalid for this operation. 581 if (!Res->isRealFloatingType()) 582 return Diag(OrigArg0->getLocStart(), 583 diag::err_typecheck_call_invalid_ordered_compare) 584 << OrigArg0->getType() << OrigArg1->getType() 585 << SourceRange(OrigArg0->getLocStart(), OrigArg1->getLocEnd()); 586 587 return false; 588} 589 590/// SemaBuiltinSemaBuiltinFPClassification - Handle functions like 591/// __builtin_isnan and friends. This is declared to take (...), so we have 592/// to check everything. We expect the last argument to be a floating point 593/// value. 594bool Sema::SemaBuiltinFPClassification(CallExpr *TheCall, unsigned NumArgs) { 595 if (TheCall->getNumArgs() < NumArgs) 596 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 597 << 0 << NumArgs << TheCall->getNumArgs()/*function call*/; 598 if (TheCall->getNumArgs() > NumArgs) 599 return Diag(TheCall->getArg(NumArgs)->getLocStart(), 600 diag::err_typecheck_call_too_many_args) 601 << 0 /*function call*/ << NumArgs << TheCall->getNumArgs() 602 << SourceRange(TheCall->getArg(NumArgs)->getLocStart(), 603 (*(TheCall->arg_end()-1))->getLocEnd()); 604 605 Expr *OrigArg = TheCall->getArg(NumArgs-1); 606 607 if (OrigArg->isTypeDependent()) 608 return false; 609 610 // This operation requires a non-_Complex floating-point number. 611 if (!OrigArg->getType()->isRealFloatingType()) 612 return Diag(OrigArg->getLocStart(), 613 diag::err_typecheck_call_invalid_unary_fp) 614 << OrigArg->getType() << OrigArg->getSourceRange(); 615 616 // If this is an implicit conversion from float -> double, remove it. 617 if (ImplicitCastExpr *Cast = dyn_cast<ImplicitCastExpr>(OrigArg)) { 618 Expr *CastArg = Cast->getSubExpr(); 619 if (CastArg->getType()->isSpecificBuiltinType(BuiltinType::Float)) { 620 assert(Cast->getType()->isSpecificBuiltinType(BuiltinType::Double) && 621 "promotion from float to double is the only expected cast here"); 622 Cast->setSubExpr(0); 623 Cast->Destroy(Context); 624 TheCall->setArg(NumArgs-1, CastArg); 625 OrigArg = CastArg; 626 } 627 } 628 629 return false; 630} 631 632/// SemaBuiltinShuffleVector - Handle __builtin_shufflevector. 633// This is declared to take (...), so we have to check everything. 634Action::OwningExprResult Sema::SemaBuiltinShuffleVector(CallExpr *TheCall) { 635 if (TheCall->getNumArgs() < 3) 636 return ExprError(Diag(TheCall->getLocEnd(), 637 diag::err_typecheck_call_too_few_args_at_least) 638 << 0 /*function call*/ << 3 << TheCall->getNumArgs() 639 << TheCall->getSourceRange()); 640 641 unsigned numElements = std::numeric_limits<unsigned>::max(); 642 if (!TheCall->getArg(0)->isTypeDependent() && 643 !TheCall->getArg(1)->isTypeDependent()) { 644 QualType FAType = TheCall->getArg(0)->getType(); 645 QualType SAType = TheCall->getArg(1)->getType(); 646 647 if (!FAType->isVectorType() || !SAType->isVectorType()) { 648 Diag(TheCall->getLocStart(), diag::err_shufflevector_non_vector) 649 << SourceRange(TheCall->getArg(0)->getLocStart(), 650 TheCall->getArg(1)->getLocEnd()); 651 return ExprError(); 652 } 653 654 if (!Context.hasSameUnqualifiedType(FAType, SAType)) { 655 Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector) 656 << SourceRange(TheCall->getArg(0)->getLocStart(), 657 TheCall->getArg(1)->getLocEnd()); 658 return ExprError(); 659 } 660 661 numElements = FAType->getAs<VectorType>()->getNumElements(); 662 if (TheCall->getNumArgs() != numElements+2) { 663 if (TheCall->getNumArgs() < numElements+2) 664 return ExprError(Diag(TheCall->getLocEnd(), 665 diag::err_typecheck_call_too_few_args) 666 << 0 /*function call*/ 667 << numElements+2 << TheCall->getNumArgs() 668 << TheCall->getSourceRange()); 669 return ExprError(Diag(TheCall->getLocEnd(), 670 diag::err_typecheck_call_too_many_args) 671 << 0 /*function call*/ 672 << numElements+2 << TheCall->getNumArgs() 673 << TheCall->getSourceRange()); 674 } 675 } 676 677 for (unsigned i = 2; i < TheCall->getNumArgs(); i++) { 678 if (TheCall->getArg(i)->isTypeDependent() || 679 TheCall->getArg(i)->isValueDependent()) 680 continue; 681 682 llvm::APSInt Result; 683 if (SemaBuiltinConstantArg(TheCall, i, Result)) 684 return ExprError(); 685 686 if (Result.getActiveBits() > 64 || Result.getZExtValue() >= numElements*2) 687 return ExprError(Diag(TheCall->getLocStart(), 688 diag::err_shufflevector_argument_too_large) 689 << TheCall->getArg(i)->getSourceRange()); 690 } 691 692 llvm::SmallVector<Expr*, 32> exprs; 693 694 for (unsigned i = 0, e = TheCall->getNumArgs(); i != e; i++) { 695 exprs.push_back(TheCall->getArg(i)); 696 TheCall->setArg(i, 0); 697 } 698 699 return Owned(new (Context) ShuffleVectorExpr(Context, exprs.begin(), 700 exprs.size(), exprs[0]->getType(), 701 TheCall->getCallee()->getLocStart(), 702 TheCall->getRParenLoc())); 703} 704 705/// SemaBuiltinPrefetch - Handle __builtin_prefetch. 706// This is declared to take (const void*, ...) and can take two 707// optional constant int args. 708bool Sema::SemaBuiltinPrefetch(CallExpr *TheCall) { 709 unsigned NumArgs = TheCall->getNumArgs(); 710 711 if (NumArgs > 3) 712 return Diag(TheCall->getLocEnd(), 713 diag::err_typecheck_call_too_many_args_at_most) 714 << 0 /*function call*/ << 3 << NumArgs 715 << TheCall->getSourceRange(); 716 717 // Argument 0 is checked for us and the remaining arguments must be 718 // constant integers. 719 for (unsigned i = 1; i != NumArgs; ++i) { 720 Expr *Arg = TheCall->getArg(i); 721 722 llvm::APSInt Result; 723 if (SemaBuiltinConstantArg(TheCall, i, Result)) 724 return true; 725 726 // FIXME: gcc issues a warning and rewrites these to 0. These 727 // seems especially odd for the third argument since the default 728 // is 3. 729 if (i == 1) { 730 if (Result.getLimitedValue() > 1) 731 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 732 << "0" << "1" << Arg->getSourceRange(); 733 } else { 734 if (Result.getLimitedValue() > 3) 735 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 736 << "0" << "3" << Arg->getSourceRange(); 737 } 738 } 739 740 return false; 741} 742 743/// SemaBuiltinConstantArg - Handle a check if argument ArgNum of CallExpr 744/// TheCall is a constant expression. 745bool Sema::SemaBuiltinConstantArg(CallExpr *TheCall, int ArgNum, 746 llvm::APSInt &Result) { 747 Expr *Arg = TheCall->getArg(ArgNum); 748 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 749 FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl()); 750 751 if (Arg->isTypeDependent() || Arg->isValueDependent()) return false; 752 753 if (!Arg->isIntegerConstantExpr(Result, Context)) 754 return Diag(TheCall->getLocStart(), diag::err_constant_integer_arg_type) 755 << FDecl->getDeclName() << Arg->getSourceRange(); 756 757 return false; 758} 759 760/// SemaBuiltinObjectSize - Handle __builtin_object_size(void *ptr, 761/// int type). This simply type checks that type is one of the defined 762/// constants (0-3). 763// For compatability check 0-3, llvm only handles 0 and 2. 764bool Sema::SemaBuiltinObjectSize(CallExpr *TheCall) { 765 llvm::APSInt Result; 766 767 // Check constant-ness first. 768 if (SemaBuiltinConstantArg(TheCall, 1, Result)) 769 return true; 770 771 Expr *Arg = TheCall->getArg(1); 772 if (Result.getSExtValue() < 0 || Result.getSExtValue() > 3) { 773 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 774 << "0" << "3" << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 775 } 776 777 return false; 778} 779 780/// SemaBuiltinLongjmp - Handle __builtin_longjmp(void *env[5], int val). 781/// This checks that val is a constant 1. 782bool Sema::SemaBuiltinLongjmp(CallExpr *TheCall) { 783 Expr *Arg = TheCall->getArg(1); 784 llvm::APSInt Result; 785 786 // TODO: This is less than ideal. Overload this to take a value. 787 if (SemaBuiltinConstantArg(TheCall, 1, Result)) 788 return true; 789 790 if (Result != 1) 791 return Diag(TheCall->getLocStart(), diag::err_builtin_longjmp_invalid_val) 792 << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 793 794 return false; 795} 796 797// Handle i > 1 ? "x" : "y", recursivelly 798bool Sema::SemaCheckStringLiteral(const Expr *E, const CallExpr *TheCall, 799 bool HasVAListArg, 800 unsigned format_idx, unsigned firstDataArg) { 801 if (E->isTypeDependent() || E->isValueDependent()) 802 return false; 803 804 switch (E->getStmtClass()) { 805 case Stmt::ConditionalOperatorClass: { 806 const ConditionalOperator *C = cast<ConditionalOperator>(E); 807 return SemaCheckStringLiteral(C->getTrueExpr(), TheCall, 808 HasVAListArg, format_idx, firstDataArg) 809 && SemaCheckStringLiteral(C->getRHS(), TheCall, 810 HasVAListArg, format_idx, firstDataArg); 811 } 812 813 case Stmt::ImplicitCastExprClass: { 814 const ImplicitCastExpr *Expr = cast<ImplicitCastExpr>(E); 815 return SemaCheckStringLiteral(Expr->getSubExpr(), TheCall, HasVAListArg, 816 format_idx, firstDataArg); 817 } 818 819 case Stmt::ParenExprClass: { 820 const ParenExpr *Expr = cast<ParenExpr>(E); 821 return SemaCheckStringLiteral(Expr->getSubExpr(), TheCall, HasVAListArg, 822 format_idx, firstDataArg); 823 } 824 825 case Stmt::DeclRefExprClass: { 826 const DeclRefExpr *DR = cast<DeclRefExpr>(E); 827 828 // As an exception, do not flag errors for variables binding to 829 // const string literals. 830 if (const VarDecl *VD = dyn_cast<VarDecl>(DR->getDecl())) { 831 bool isConstant = false; 832 QualType T = DR->getType(); 833 834 if (const ArrayType *AT = Context.getAsArrayType(T)) { 835 isConstant = AT->getElementType().isConstant(Context); 836 } else if (const PointerType *PT = T->getAs<PointerType>()) { 837 isConstant = T.isConstant(Context) && 838 PT->getPointeeType().isConstant(Context); 839 } 840 841 if (isConstant) { 842 if (const Expr *Init = VD->getAnyInitializer()) 843 return SemaCheckStringLiteral(Init, TheCall, 844 HasVAListArg, format_idx, firstDataArg); 845 } 846 847 // For vprintf* functions (i.e., HasVAListArg==true), we add a 848 // special check to see if the format string is a function parameter 849 // of the function calling the printf function. If the function 850 // has an attribute indicating it is a printf-like function, then we 851 // should suppress warnings concerning non-literals being used in a call 852 // to a vprintf function. For example: 853 // 854 // void 855 // logmessage(char const *fmt __attribute__ (format (printf, 1, 2)), ...){ 856 // va_list ap; 857 // va_start(ap, fmt); 858 // vprintf(fmt, ap); // Do NOT emit a warning about "fmt". 859 // ... 860 // 861 // 862 // FIXME: We don't have full attribute support yet, so just check to see 863 // if the argument is a DeclRefExpr that references a parameter. We'll 864 // add proper support for checking the attribute later. 865 if (HasVAListArg) 866 if (isa<ParmVarDecl>(VD)) 867 return true; 868 } 869 870 return false; 871 } 872 873 case Stmt::CallExprClass: { 874 const CallExpr *CE = cast<CallExpr>(E); 875 if (const ImplicitCastExpr *ICE 876 = dyn_cast<ImplicitCastExpr>(CE->getCallee())) { 877 if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(ICE->getSubExpr())) { 878 if (const FunctionDecl *FD = dyn_cast<FunctionDecl>(DRE->getDecl())) { 879 if (const FormatArgAttr *FA = FD->getAttr<FormatArgAttr>()) { 880 unsigned ArgIndex = FA->getFormatIdx(); 881 const Expr *Arg = CE->getArg(ArgIndex - 1); 882 883 return SemaCheckStringLiteral(Arg, TheCall, HasVAListArg, 884 format_idx, firstDataArg); 885 } 886 } 887 } 888 } 889 890 return false; 891 } 892 case Stmt::ObjCStringLiteralClass: 893 case Stmt::StringLiteralClass: { 894 const StringLiteral *StrE = NULL; 895 896 if (const ObjCStringLiteral *ObjCFExpr = dyn_cast<ObjCStringLiteral>(E)) 897 StrE = ObjCFExpr->getString(); 898 else 899 StrE = cast<StringLiteral>(E); 900 901 if (StrE) { 902 CheckPrintfString(StrE, E, TheCall, HasVAListArg, format_idx, 903 firstDataArg); 904 return true; 905 } 906 907 return false; 908 } 909 910 default: 911 return false; 912 } 913} 914 915void 916Sema::CheckNonNullArguments(const NonNullAttr *NonNull, 917 const CallExpr *TheCall) { 918 for (NonNullAttr::iterator i = NonNull->begin(), e = NonNull->end(); 919 i != e; ++i) { 920 const Expr *ArgExpr = TheCall->getArg(*i); 921 if (ArgExpr->isNullPointerConstant(Context, 922 Expr::NPC_ValueDependentIsNotNull)) 923 Diag(TheCall->getCallee()->getLocStart(), diag::warn_null_arg) 924 << ArgExpr->getSourceRange(); 925 } 926} 927 928/// CheckPrintfArguments - Check calls to printf (and similar functions) for 929/// correct use of format strings. 930/// 931/// HasVAListArg - A predicate indicating whether the printf-like 932/// function is passed an explicit va_arg argument (e.g., vprintf) 933/// 934/// format_idx - The index into Args for the format string. 935/// 936/// Improper format strings to functions in the printf family can be 937/// the source of bizarre bugs and very serious security holes. A 938/// good source of information is available in the following paper 939/// (which includes additional references): 940/// 941/// FormatGuard: Automatic Protection From printf Format String 942/// Vulnerabilities, Proceedings of the 10th USENIX Security Symposium, 2001. 943/// 944/// TODO: 945/// Functionality implemented: 946/// 947/// We can statically check the following properties for string 948/// literal format strings for non v.*printf functions (where the 949/// arguments are passed directly): 950// 951/// (1) Are the number of format conversions equal to the number of 952/// data arguments? 953/// 954/// (2) Does each format conversion correctly match the type of the 955/// corresponding data argument? 956/// 957/// Moreover, for all printf functions we can: 958/// 959/// (3) Check for a missing format string (when not caught by type checking). 960/// 961/// (4) Check for no-operation flags; e.g. using "#" with format 962/// conversion 'c' (TODO) 963/// 964/// (5) Check the use of '%n', a major source of security holes. 965/// 966/// (6) Check for malformed format conversions that don't specify anything. 967/// 968/// (7) Check for empty format strings. e.g: printf(""); 969/// 970/// (8) Check that the format string is a wide literal. 971/// 972/// All of these checks can be done by parsing the format string. 973/// 974void 975Sema::CheckPrintfArguments(const CallExpr *TheCall, bool HasVAListArg, 976 unsigned format_idx, unsigned firstDataArg) { 977 const Expr *Fn = TheCall->getCallee(); 978 979 // The way the format attribute works in GCC, the implicit this argument 980 // of member functions is counted. However, it doesn't appear in our own 981 // lists, so decrement format_idx in that case. 982 if (isa<CXXMemberCallExpr>(TheCall)) { 983 // Catch a format attribute mistakenly referring to the object argument. 984 if (format_idx == 0) 985 return; 986 --format_idx; 987 if(firstDataArg != 0) 988 --firstDataArg; 989 } 990 991 // CHECK: printf-like function is called with no format string. 992 if (format_idx >= TheCall->getNumArgs()) { 993 Diag(TheCall->getRParenLoc(), diag::warn_printf_missing_format_string) 994 << Fn->getSourceRange(); 995 return; 996 } 997 998 const Expr *OrigFormatExpr = TheCall->getArg(format_idx)->IgnoreParenCasts(); 999 1000 // CHECK: format string is not a string literal. 1001 // 1002 // Dynamically generated format strings are difficult to 1003 // automatically vet at compile time. Requiring that format strings 1004 // are string literals: (1) permits the checking of format strings by 1005 // the compiler and thereby (2) can practically remove the source of 1006 // many format string exploits. 1007 1008 // Format string can be either ObjC string (e.g. @"%d") or 1009 // C string (e.g. "%d") 1010 // ObjC string uses the same format specifiers as C string, so we can use 1011 // the same format string checking logic for both ObjC and C strings. 1012 if (SemaCheckStringLiteral(OrigFormatExpr, TheCall, HasVAListArg, format_idx, 1013 firstDataArg)) 1014 return; // Literal format string found, check done! 1015 1016 // If there are no arguments specified, warn with -Wformat-security, otherwise 1017 // warn only with -Wformat-nonliteral. 1018 if (TheCall->getNumArgs() == format_idx+1) 1019 Diag(TheCall->getArg(format_idx)->getLocStart(), 1020 diag::warn_printf_nonliteral_noargs) 1021 << OrigFormatExpr->getSourceRange(); 1022 else 1023 Diag(TheCall->getArg(format_idx)->getLocStart(), 1024 diag::warn_printf_nonliteral) 1025 << OrigFormatExpr->getSourceRange(); 1026} 1027 1028namespace { 1029class CheckPrintfHandler : public analyze_printf::FormatStringHandler { 1030 Sema &S; 1031 const StringLiteral *FExpr; 1032 const Expr *OrigFormatExpr; 1033 const unsigned FirstDataArg; 1034 const unsigned NumDataArgs; 1035 const bool IsObjCLiteral; 1036 const char *Beg; // Start of format string. 1037 const bool HasVAListArg; 1038 const CallExpr *TheCall; 1039 unsigned FormatIdx; 1040 llvm::BitVector CoveredArgs; 1041 bool usesPositionalArgs; 1042 bool atFirstArg; 1043public: 1044 CheckPrintfHandler(Sema &s, const StringLiteral *fexpr, 1045 const Expr *origFormatExpr, unsigned firstDataArg, 1046 unsigned numDataArgs, bool isObjCLiteral, 1047 const char *beg, bool hasVAListArg, 1048 const CallExpr *theCall, unsigned formatIdx) 1049 : S(s), FExpr(fexpr), OrigFormatExpr(origFormatExpr), 1050 FirstDataArg(firstDataArg), 1051 NumDataArgs(numDataArgs), 1052 IsObjCLiteral(isObjCLiteral), Beg(beg), 1053 HasVAListArg(hasVAListArg), 1054 TheCall(theCall), FormatIdx(formatIdx), 1055 usesPositionalArgs(false), atFirstArg(true) { 1056 CoveredArgs.resize(numDataArgs); 1057 CoveredArgs.reset(); 1058 } 1059 1060 void DoneProcessing(); 1061 1062 void HandleIncompleteFormatSpecifier(const char *startSpecifier, 1063 unsigned specifierLen); 1064 1065 bool 1066 HandleInvalidConversionSpecifier(const analyze_printf::FormatSpecifier &FS, 1067 const char *startSpecifier, 1068 unsigned specifierLen); 1069 1070 virtual void HandleInvalidPosition(const char *startSpecifier, 1071 unsigned specifierLen, 1072 analyze_printf::PositionContext p); 1073 1074 virtual void HandleZeroPosition(const char *startPos, unsigned posLen); 1075 1076 void HandleNullChar(const char *nullCharacter); 1077 1078 bool HandleFormatSpecifier(const analyze_printf::FormatSpecifier &FS, 1079 const char *startSpecifier, 1080 unsigned specifierLen); 1081private: 1082 SourceRange getFormatStringRange(); 1083 SourceRange getFormatSpecifierRange(const char *startSpecifier, 1084 unsigned specifierLen); 1085 SourceLocation getLocationOfByte(const char *x); 1086 1087 bool HandleAmount(const analyze_printf::OptionalAmount &Amt, unsigned k, 1088 const char *startSpecifier, unsigned specifierLen); 1089 void HandleFlags(const analyze_printf::FormatSpecifier &FS, 1090 llvm::StringRef flag, llvm::StringRef cspec, 1091 const char *startSpecifier, unsigned specifierLen); 1092 1093 const Expr *getDataArg(unsigned i) const; 1094}; 1095} 1096 1097SourceRange CheckPrintfHandler::getFormatStringRange() { 1098 return OrigFormatExpr->getSourceRange(); 1099} 1100 1101SourceRange CheckPrintfHandler:: 1102getFormatSpecifierRange(const char *startSpecifier, unsigned specifierLen) { 1103 return SourceRange(getLocationOfByte(startSpecifier), 1104 getLocationOfByte(startSpecifier+specifierLen-1)); 1105} 1106 1107SourceLocation CheckPrintfHandler::getLocationOfByte(const char *x) { 1108 return S.getLocationOfStringLiteralByte(FExpr, x - Beg); 1109} 1110 1111void CheckPrintfHandler:: 1112HandleIncompleteFormatSpecifier(const char *startSpecifier, 1113 unsigned specifierLen) { 1114 SourceLocation Loc = getLocationOfByte(startSpecifier); 1115 S.Diag(Loc, diag::warn_printf_incomplete_specifier) 1116 << getFormatSpecifierRange(startSpecifier, specifierLen); 1117} 1118 1119void 1120CheckPrintfHandler::HandleInvalidPosition(const char *startPos, unsigned posLen, 1121 analyze_printf::PositionContext p) { 1122 SourceLocation Loc = getLocationOfByte(startPos); 1123 S.Diag(Loc, diag::warn_printf_invalid_positional_specifier) 1124 << (unsigned) p << getFormatSpecifierRange(startPos, posLen); 1125} 1126 1127void CheckPrintfHandler::HandleZeroPosition(const char *startPos, 1128 unsigned posLen) { 1129 SourceLocation Loc = getLocationOfByte(startPos); 1130 S.Diag(Loc, diag::warn_printf_zero_positional_specifier) 1131 << getFormatSpecifierRange(startPos, posLen); 1132} 1133 1134bool CheckPrintfHandler:: 1135HandleInvalidConversionSpecifier(const analyze_printf::FormatSpecifier &FS, 1136 const char *startSpecifier, 1137 unsigned specifierLen) { 1138 1139 unsigned argIndex = FS.getArgIndex(); 1140 bool keepGoing = true; 1141 if (argIndex < NumDataArgs) { 1142 // Consider the argument coverered, even though the specifier doesn't 1143 // make sense. 1144 CoveredArgs.set(argIndex); 1145 } 1146 else { 1147 // If argIndex exceeds the number of data arguments we 1148 // don't issue a warning because that is just a cascade of warnings (and 1149 // they may have intended '%%' anyway). We don't want to continue processing 1150 // the format string after this point, however, as we will like just get 1151 // gibberish when trying to match arguments. 1152 keepGoing = false; 1153 } 1154 1155 const analyze_printf::ConversionSpecifier &CS = 1156 FS.getConversionSpecifier(); 1157 SourceLocation Loc = getLocationOfByte(CS.getStart()); 1158 S.Diag(Loc, diag::warn_printf_invalid_conversion) 1159 << llvm::StringRef(CS.getStart(), CS.getLength()) 1160 << getFormatSpecifierRange(startSpecifier, specifierLen); 1161 1162 return keepGoing; 1163} 1164 1165void CheckPrintfHandler::HandleNullChar(const char *nullCharacter) { 1166 // The presence of a null character is likely an error. 1167 S.Diag(getLocationOfByte(nullCharacter), 1168 diag::warn_printf_format_string_contains_null_char) 1169 << getFormatStringRange(); 1170} 1171 1172const Expr *CheckPrintfHandler::getDataArg(unsigned i) const { 1173 return TheCall->getArg(FirstDataArg + i); 1174} 1175 1176void CheckPrintfHandler::HandleFlags(const analyze_printf::FormatSpecifier &FS, 1177 llvm::StringRef flag, 1178 llvm::StringRef cspec, 1179 const char *startSpecifier, 1180 unsigned specifierLen) { 1181 const analyze_printf::ConversionSpecifier &CS = FS.getConversionSpecifier(); 1182 S.Diag(getLocationOfByte(CS.getStart()), diag::warn_printf_nonsensical_flag) 1183 << flag << cspec << getFormatSpecifierRange(startSpecifier, specifierLen); 1184} 1185 1186bool 1187CheckPrintfHandler::HandleAmount(const analyze_printf::OptionalAmount &Amt, 1188 unsigned k, const char *startSpecifier, 1189 unsigned specifierLen) { 1190 1191 if (Amt.hasDataArgument()) { 1192 if (!HasVAListArg) { 1193 unsigned argIndex = Amt.getArgIndex(); 1194 if (argIndex >= NumDataArgs) { 1195 S.Diag(getLocationOfByte(Amt.getStart()), 1196 diag::warn_printf_asterisk_missing_arg) 1197 << k << getFormatSpecifierRange(startSpecifier, specifierLen); 1198 // Don't do any more checking. We will just emit 1199 // spurious errors. 1200 return false; 1201 } 1202 1203 // Type check the data argument. It should be an 'int'. 1204 // Although not in conformance with C99, we also allow the argument to be 1205 // an 'unsigned int' as that is a reasonably safe case. GCC also 1206 // doesn't emit a warning for that case. 1207 CoveredArgs.set(argIndex); 1208 const Expr *Arg = getDataArg(argIndex); 1209 QualType T = Arg->getType(); 1210 1211 const analyze_printf::ArgTypeResult &ATR = Amt.getArgType(S.Context); 1212 assert(ATR.isValid()); 1213 1214 if (!ATR.matchesType(S.Context, T)) { 1215 S.Diag(getLocationOfByte(Amt.getStart()), 1216 diag::warn_printf_asterisk_wrong_type) 1217 << k 1218 << ATR.getRepresentativeType(S.Context) << T 1219 << getFormatSpecifierRange(startSpecifier, specifierLen) 1220 << Arg->getSourceRange(); 1221 // Don't do any more checking. We will just emit 1222 // spurious errors. 1223 return false; 1224 } 1225 } 1226 } 1227 return true; 1228} 1229 1230bool 1231CheckPrintfHandler::HandleFormatSpecifier(const analyze_printf::FormatSpecifier 1232 &FS, 1233 const char *startSpecifier, 1234 unsigned specifierLen) { 1235 1236 using namespace analyze_printf; 1237 const ConversionSpecifier &CS = FS.getConversionSpecifier(); 1238 1239 if (atFirstArg) { 1240 atFirstArg = false; 1241 usesPositionalArgs = FS.usesPositionalArg(); 1242 } 1243 else if (usesPositionalArgs != FS.usesPositionalArg()) { 1244 // Cannot mix-and-match positional and non-positional arguments. 1245 S.Diag(getLocationOfByte(CS.getStart()), 1246 diag::warn_printf_mix_positional_nonpositional_args) 1247 << getFormatSpecifierRange(startSpecifier, specifierLen); 1248 return false; 1249 } 1250 1251 // First check if the field width, precision, and conversion specifier 1252 // have matching data arguments. 1253 if (!HandleAmount(FS.getFieldWidth(), /* field width */ 0, 1254 startSpecifier, specifierLen)) { 1255 return false; 1256 } 1257 1258 if (!HandleAmount(FS.getPrecision(), /* precision */ 1, 1259 startSpecifier, specifierLen)) { 1260 return false; 1261 } 1262 1263 if (!CS.consumesDataArgument()) { 1264 // FIXME: Technically specifying a precision or field width here 1265 // makes no sense. Worth issuing a warning at some point. 1266 return true; 1267 } 1268 1269 // Consume the argument. 1270 unsigned argIndex = FS.getArgIndex(); 1271 if (argIndex < NumDataArgs) { 1272 // The check to see if the argIndex is valid will come later. 1273 // We set the bit here because we may exit early from this 1274 // function if we encounter some other error. 1275 CoveredArgs.set(argIndex); 1276 } 1277 1278 // Check for using an Objective-C specific conversion specifier 1279 // in a non-ObjC literal. 1280 if (!IsObjCLiteral && CS.isObjCArg()) { 1281 return HandleInvalidConversionSpecifier(FS, startSpecifier, specifierLen); 1282 } 1283 1284 // Are we using '%n'? Issue a warning about this being 1285 // a possible security issue. 1286 if (CS.getKind() == ConversionSpecifier::OutIntPtrArg) { 1287 S.Diag(getLocationOfByte(CS.getStart()), diag::warn_printf_write_back) 1288 << getFormatSpecifierRange(startSpecifier, specifierLen); 1289 // Continue checking the other format specifiers. 1290 return true; 1291 } 1292 1293 if (CS.getKind() == ConversionSpecifier::VoidPtrArg) { 1294 if (FS.getPrecision().getHowSpecified() != OptionalAmount::NotSpecified) 1295 S.Diag(getLocationOfByte(CS.getStart()), 1296 diag::warn_printf_nonsensical_precision) 1297 << CS.getCharacters() 1298 << getFormatSpecifierRange(startSpecifier, specifierLen); 1299 } 1300 if (CS.getKind() == ConversionSpecifier::VoidPtrArg || 1301 CS.getKind() == ConversionSpecifier::CStrArg) { 1302 // FIXME: Instead of using "0", "+", etc., eventually get them from 1303 // the FormatSpecifier. 1304 if (FS.hasLeadingZeros()) 1305 HandleFlags(FS, "0", CS.getCharacters(), startSpecifier, specifierLen); 1306 if (FS.hasPlusPrefix()) 1307 HandleFlags(FS, "+", CS.getCharacters(), startSpecifier, specifierLen); 1308 if (FS.hasSpacePrefix()) 1309 HandleFlags(FS, " ", CS.getCharacters(), startSpecifier, specifierLen); 1310 } 1311 1312 // The remaining checks depend on the data arguments. 1313 if (HasVAListArg) 1314 return true; 1315 1316 if (argIndex >= NumDataArgs) { 1317 if (FS.usesPositionalArg()) { 1318 S.Diag(getLocationOfByte(CS.getStart()), 1319 diag::warn_printf_positional_arg_exceeds_data_args) 1320 << (argIndex+1) << NumDataArgs 1321 << getFormatSpecifierRange(startSpecifier, specifierLen); 1322 } 1323 else { 1324 S.Diag(getLocationOfByte(CS.getStart()), 1325 diag::warn_printf_insufficient_data_args) 1326 << getFormatSpecifierRange(startSpecifier, specifierLen); 1327 } 1328 1329 // Don't do any more checking. 1330 return false; 1331 } 1332 1333 // Now type check the data expression that matches the 1334 // format specifier. 1335 const Expr *Ex = getDataArg(argIndex); 1336 const analyze_printf::ArgTypeResult &ATR = FS.getArgType(S.Context); 1337 if (ATR.isValid() && !ATR.matchesType(S.Context, Ex->getType())) { 1338 // Check if we didn't match because of an implicit cast from a 'char' 1339 // or 'short' to an 'int'. This is done because printf is a varargs 1340 // function. 1341 if (const ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(Ex)) 1342 if (ICE->getType() == S.Context.IntTy) 1343 if (ATR.matchesType(S.Context, ICE->getSubExpr()->getType())) 1344 return true; 1345 1346 S.Diag(getLocationOfByte(CS.getStart()), 1347 diag::warn_printf_conversion_argument_type_mismatch) 1348 << ATR.getRepresentativeType(S.Context) << Ex->getType() 1349 << getFormatSpecifierRange(startSpecifier, specifierLen) 1350 << Ex->getSourceRange(); 1351 } 1352 1353 return true; 1354} 1355 1356void CheckPrintfHandler::DoneProcessing() { 1357 // Does the number of data arguments exceed the number of 1358 // format conversions in the format string? 1359 if (!HasVAListArg) { 1360 // Find any arguments that weren't covered. 1361 CoveredArgs.flip(); 1362 signed notCoveredArg = CoveredArgs.find_first(); 1363 if (notCoveredArg >= 0) { 1364 assert((unsigned)notCoveredArg < NumDataArgs); 1365 S.Diag(getDataArg((unsigned) notCoveredArg)->getLocStart(), 1366 diag::warn_printf_data_arg_not_used) 1367 << getFormatStringRange(); 1368 } 1369 } 1370} 1371 1372void Sema::CheckPrintfString(const StringLiteral *FExpr, 1373 const Expr *OrigFormatExpr, 1374 const CallExpr *TheCall, bool HasVAListArg, 1375 unsigned format_idx, unsigned firstDataArg) { 1376 1377 // CHECK: is the format string a wide literal? 1378 if (FExpr->isWide()) { 1379 Diag(FExpr->getLocStart(), 1380 diag::warn_printf_format_string_is_wide_literal) 1381 << OrigFormatExpr->getSourceRange(); 1382 return; 1383 } 1384 1385 // Str - The format string. NOTE: this is NOT null-terminated! 1386 const char *Str = FExpr->getStrData(); 1387 1388 // CHECK: empty format string? 1389 unsigned StrLen = FExpr->getByteLength(); 1390 1391 if (StrLen == 0) { 1392 Diag(FExpr->getLocStart(), diag::warn_printf_empty_format_string) 1393 << OrigFormatExpr->getSourceRange(); 1394 return; 1395 } 1396 1397 CheckPrintfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg, 1398 TheCall->getNumArgs() - firstDataArg, 1399 isa<ObjCStringLiteral>(OrigFormatExpr), Str, 1400 HasVAListArg, TheCall, format_idx); 1401 1402 if (!analyze_printf::ParseFormatString(H, Str, Str + StrLen)) 1403 H.DoneProcessing(); 1404} 1405 1406//===--- CHECK: Return Address of Stack Variable --------------------------===// 1407 1408static DeclRefExpr* EvalVal(Expr *E); 1409static DeclRefExpr* EvalAddr(Expr* E); 1410 1411/// CheckReturnStackAddr - Check if a return statement returns the address 1412/// of a stack variable. 1413void 1414Sema::CheckReturnStackAddr(Expr *RetValExp, QualType lhsType, 1415 SourceLocation ReturnLoc) { 1416 1417 // Perform checking for returned stack addresses. 1418 if (lhsType->isPointerType() || lhsType->isBlockPointerType()) { 1419 if (DeclRefExpr *DR = EvalAddr(RetValExp)) 1420 Diag(DR->getLocStart(), diag::warn_ret_stack_addr) 1421 << DR->getDecl()->getDeclName() << RetValExp->getSourceRange(); 1422 1423 // Skip over implicit cast expressions when checking for block expressions. 1424 RetValExp = RetValExp->IgnoreParenCasts(); 1425 1426 if (BlockExpr *C = dyn_cast<BlockExpr>(RetValExp)) 1427 if (C->hasBlockDeclRefExprs()) 1428 Diag(C->getLocStart(), diag::err_ret_local_block) 1429 << C->getSourceRange(); 1430 1431 if (AddrLabelExpr *ALE = dyn_cast<AddrLabelExpr>(RetValExp)) 1432 Diag(ALE->getLocStart(), diag::warn_ret_addr_label) 1433 << ALE->getSourceRange(); 1434 1435 } else if (lhsType->isReferenceType()) { 1436 // Perform checking for stack values returned by reference. 1437 // Check for a reference to the stack 1438 if (DeclRefExpr *DR = EvalVal(RetValExp)) 1439 Diag(DR->getLocStart(), diag::warn_ret_stack_ref) 1440 << DR->getDecl()->getDeclName() << RetValExp->getSourceRange(); 1441 } 1442} 1443 1444/// EvalAddr - EvalAddr and EvalVal are mutually recursive functions that 1445/// check if the expression in a return statement evaluates to an address 1446/// to a location on the stack. The recursion is used to traverse the 1447/// AST of the return expression, with recursion backtracking when we 1448/// encounter a subexpression that (1) clearly does not lead to the address 1449/// of a stack variable or (2) is something we cannot determine leads to 1450/// the address of a stack variable based on such local checking. 1451/// 1452/// EvalAddr processes expressions that are pointers that are used as 1453/// references (and not L-values). EvalVal handles all other values. 1454/// At the base case of the recursion is a check for a DeclRefExpr* in 1455/// the refers to a stack variable. 1456/// 1457/// This implementation handles: 1458/// 1459/// * pointer-to-pointer casts 1460/// * implicit conversions from array references to pointers 1461/// * taking the address of fields 1462/// * arbitrary interplay between "&" and "*" operators 1463/// * pointer arithmetic from an address of a stack variable 1464/// * taking the address of an array element where the array is on the stack 1465static DeclRefExpr* EvalAddr(Expr *E) { 1466 // We should only be called for evaluating pointer expressions. 1467 assert((E->getType()->isAnyPointerType() || 1468 E->getType()->isBlockPointerType() || 1469 E->getType()->isObjCQualifiedIdType()) && 1470 "EvalAddr only works on pointers"); 1471 1472 // Our "symbolic interpreter" is just a dispatch off the currently 1473 // viewed AST node. We then recursively traverse the AST by calling 1474 // EvalAddr and EvalVal appropriately. 1475 switch (E->getStmtClass()) { 1476 case Stmt::ParenExprClass: 1477 // Ignore parentheses. 1478 return EvalAddr(cast<ParenExpr>(E)->getSubExpr()); 1479 1480 case Stmt::UnaryOperatorClass: { 1481 // The only unary operator that make sense to handle here 1482 // is AddrOf. All others don't make sense as pointers. 1483 UnaryOperator *U = cast<UnaryOperator>(E); 1484 1485 if (U->getOpcode() == UnaryOperator::AddrOf) 1486 return EvalVal(U->getSubExpr()); 1487 else 1488 return NULL; 1489 } 1490 1491 case Stmt::BinaryOperatorClass: { 1492 // Handle pointer arithmetic. All other binary operators are not valid 1493 // in this context. 1494 BinaryOperator *B = cast<BinaryOperator>(E); 1495 BinaryOperator::Opcode op = B->getOpcode(); 1496 1497 if (op != BinaryOperator::Add && op != BinaryOperator::Sub) 1498 return NULL; 1499 1500 Expr *Base = B->getLHS(); 1501 1502 // Determine which argument is the real pointer base. It could be 1503 // the RHS argument instead of the LHS. 1504 if (!Base->getType()->isPointerType()) Base = B->getRHS(); 1505 1506 assert (Base->getType()->isPointerType()); 1507 return EvalAddr(Base); 1508 } 1509 1510 // For conditional operators we need to see if either the LHS or RHS are 1511 // valid DeclRefExpr*s. If one of them is valid, we return it. 1512 case Stmt::ConditionalOperatorClass: { 1513 ConditionalOperator *C = cast<ConditionalOperator>(E); 1514 1515 // Handle the GNU extension for missing LHS. 1516 if (Expr *lhsExpr = C->getLHS()) 1517 if (DeclRefExpr* LHS = EvalAddr(lhsExpr)) 1518 return LHS; 1519 1520 return EvalAddr(C->getRHS()); 1521 } 1522 1523 // For casts, we need to handle conversions from arrays to 1524 // pointer values, and pointer-to-pointer conversions. 1525 case Stmt::ImplicitCastExprClass: 1526 case Stmt::CStyleCastExprClass: 1527 case Stmt::CXXFunctionalCastExprClass: { 1528 Expr* SubExpr = cast<CastExpr>(E)->getSubExpr(); 1529 QualType T = SubExpr->getType(); 1530 1531 if (SubExpr->getType()->isPointerType() || 1532 SubExpr->getType()->isBlockPointerType() || 1533 SubExpr->getType()->isObjCQualifiedIdType()) 1534 return EvalAddr(SubExpr); 1535 else if (T->isArrayType()) 1536 return EvalVal(SubExpr); 1537 else 1538 return 0; 1539 } 1540 1541 // C++ casts. For dynamic casts, static casts, and const casts, we 1542 // are always converting from a pointer-to-pointer, so we just blow 1543 // through the cast. In the case the dynamic cast doesn't fail (and 1544 // return NULL), we take the conservative route and report cases 1545 // where we return the address of a stack variable. For Reinterpre 1546 // FIXME: The comment about is wrong; we're not always converting 1547 // from pointer to pointer. I'm guessing that this code should also 1548 // handle references to objects. 1549 case Stmt::CXXStaticCastExprClass: 1550 case Stmt::CXXDynamicCastExprClass: 1551 case Stmt::CXXConstCastExprClass: 1552 case Stmt::CXXReinterpretCastExprClass: { 1553 Expr *S = cast<CXXNamedCastExpr>(E)->getSubExpr(); 1554 if (S->getType()->isPointerType() || S->getType()->isBlockPointerType()) 1555 return EvalAddr(S); 1556 else 1557 return NULL; 1558 } 1559 1560 // Everything else: we simply don't reason about them. 1561 default: 1562 return NULL; 1563 } 1564} 1565 1566 1567/// EvalVal - This function is complements EvalAddr in the mutual recursion. 1568/// See the comments for EvalAddr for more details. 1569static DeclRefExpr* EvalVal(Expr *E) { 1570 1571 // We should only be called for evaluating non-pointer expressions, or 1572 // expressions with a pointer type that are not used as references but instead 1573 // are l-values (e.g., DeclRefExpr with a pointer type). 1574 1575 // Our "symbolic interpreter" is just a dispatch off the currently 1576 // viewed AST node. We then recursively traverse the AST by calling 1577 // EvalAddr and EvalVal appropriately. 1578 switch (E->getStmtClass()) { 1579 case Stmt::DeclRefExprClass: { 1580 // DeclRefExpr: the base case. When we hit a DeclRefExpr we are looking 1581 // at code that refers to a variable's name. We check if it has local 1582 // storage within the function, and if so, return the expression. 1583 DeclRefExpr *DR = cast<DeclRefExpr>(E); 1584 1585 if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl())) 1586 if (V->hasLocalStorage() && !V->getType()->isReferenceType()) return DR; 1587 1588 return NULL; 1589 } 1590 1591 case Stmt::ParenExprClass: 1592 // Ignore parentheses. 1593 return EvalVal(cast<ParenExpr>(E)->getSubExpr()); 1594 1595 case Stmt::UnaryOperatorClass: { 1596 // The only unary operator that make sense to handle here 1597 // is Deref. All others don't resolve to a "name." This includes 1598 // handling all sorts of rvalues passed to a unary operator. 1599 UnaryOperator *U = cast<UnaryOperator>(E); 1600 1601 if (U->getOpcode() == UnaryOperator::Deref) 1602 return EvalAddr(U->getSubExpr()); 1603 1604 return NULL; 1605 } 1606 1607 case Stmt::ArraySubscriptExprClass: { 1608 // Array subscripts are potential references to data on the stack. We 1609 // retrieve the DeclRefExpr* for the array variable if it indeed 1610 // has local storage. 1611 return EvalAddr(cast<ArraySubscriptExpr>(E)->getBase()); 1612 } 1613 1614 case Stmt::ConditionalOperatorClass: { 1615 // For conditional operators we need to see if either the LHS or RHS are 1616 // non-NULL DeclRefExpr's. If one is non-NULL, we return it. 1617 ConditionalOperator *C = cast<ConditionalOperator>(E); 1618 1619 // Handle the GNU extension for missing LHS. 1620 if (Expr *lhsExpr = C->getLHS()) 1621 if (DeclRefExpr *LHS = EvalVal(lhsExpr)) 1622 return LHS; 1623 1624 return EvalVal(C->getRHS()); 1625 } 1626 1627 // Accesses to members are potential references to data on the stack. 1628 case Stmt::MemberExprClass: { 1629 MemberExpr *M = cast<MemberExpr>(E); 1630 1631 // Check for indirect access. We only want direct field accesses. 1632 if (!M->isArrow()) 1633 return EvalVal(M->getBase()); 1634 else 1635 return NULL; 1636 } 1637 1638 // Everything else: we simply don't reason about them. 1639 default: 1640 return NULL; 1641 } 1642} 1643 1644//===--- CHECK: Floating-Point comparisons (-Wfloat-equal) ---------------===// 1645 1646/// Check for comparisons of floating point operands using != and ==. 1647/// Issue a warning if these are no self-comparisons, as they are not likely 1648/// to do what the programmer intended. 1649void Sema::CheckFloatComparison(SourceLocation loc, Expr* lex, Expr *rex) { 1650 bool EmitWarning = true; 1651 1652 Expr* LeftExprSansParen = lex->IgnoreParens(); 1653 Expr* RightExprSansParen = rex->IgnoreParens(); 1654 1655 // Special case: check for x == x (which is OK). 1656 // Do not emit warnings for such cases. 1657 if (DeclRefExpr* DRL = dyn_cast<DeclRefExpr>(LeftExprSansParen)) 1658 if (DeclRefExpr* DRR = dyn_cast<DeclRefExpr>(RightExprSansParen)) 1659 if (DRL->getDecl() == DRR->getDecl()) 1660 EmitWarning = false; 1661 1662 1663 // Special case: check for comparisons against literals that can be exactly 1664 // represented by APFloat. In such cases, do not emit a warning. This 1665 // is a heuristic: often comparison against such literals are used to 1666 // detect if a value in a variable has not changed. This clearly can 1667 // lead to false negatives. 1668 if (EmitWarning) { 1669 if (FloatingLiteral* FLL = dyn_cast<FloatingLiteral>(LeftExprSansParen)) { 1670 if (FLL->isExact()) 1671 EmitWarning = false; 1672 } else 1673 if (FloatingLiteral* FLR = dyn_cast<FloatingLiteral>(RightExprSansParen)){ 1674 if (FLR->isExact()) 1675 EmitWarning = false; 1676 } 1677 } 1678 1679 // Check for comparisons with builtin types. 1680 if (EmitWarning) 1681 if (CallExpr* CL = dyn_cast<CallExpr>(LeftExprSansParen)) 1682 if (CL->isBuiltinCall(Context)) 1683 EmitWarning = false; 1684 1685 if (EmitWarning) 1686 if (CallExpr* CR = dyn_cast<CallExpr>(RightExprSansParen)) 1687 if (CR->isBuiltinCall(Context)) 1688 EmitWarning = false; 1689 1690 // Emit the diagnostic. 1691 if (EmitWarning) 1692 Diag(loc, diag::warn_floatingpoint_eq) 1693 << lex->getSourceRange() << rex->getSourceRange(); 1694} 1695 1696//===--- CHECK: Integer mixed-sign comparisons (-Wsign-compare) --------===// 1697//===--- CHECK: Lossy implicit conversions (-Wconversion) --------------===// 1698 1699namespace { 1700 1701/// Structure recording the 'active' range of an integer-valued 1702/// expression. 1703struct IntRange { 1704 /// The number of bits active in the int. 1705 unsigned Width; 1706 1707 /// True if the int is known not to have negative values. 1708 bool NonNegative; 1709 1710 IntRange() {} 1711 IntRange(unsigned Width, bool NonNegative) 1712 : Width(Width), NonNegative(NonNegative) 1713 {} 1714 1715 // Returns the range of the bool type. 1716 static IntRange forBoolType() { 1717 return IntRange(1, true); 1718 } 1719 1720 // Returns the range of an integral type. 1721 static IntRange forType(ASTContext &C, QualType T) { 1722 return forCanonicalType(C, T->getCanonicalTypeInternal().getTypePtr()); 1723 } 1724 1725 // Returns the range of an integeral type based on its canonical 1726 // representation. 1727 static IntRange forCanonicalType(ASTContext &C, const Type *T) { 1728 assert(T->isCanonicalUnqualified()); 1729 1730 if (const VectorType *VT = dyn_cast<VectorType>(T)) 1731 T = VT->getElementType().getTypePtr(); 1732 if (const ComplexType *CT = dyn_cast<ComplexType>(T)) 1733 T = CT->getElementType().getTypePtr(); 1734 1735 if (const EnumType *ET = dyn_cast<EnumType>(T)) { 1736 EnumDecl *Enum = ET->getDecl(); 1737 unsigned NumPositive = Enum->getNumPositiveBits(); 1738 unsigned NumNegative = Enum->getNumNegativeBits(); 1739 1740 return IntRange(std::max(NumPositive, NumNegative), NumNegative == 0); 1741 } 1742 1743 const BuiltinType *BT = cast<BuiltinType>(T); 1744 assert(BT->isInteger()); 1745 1746 return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger()); 1747 } 1748 1749 // Returns the supremum of two ranges: i.e. their conservative merge. 1750 static IntRange join(IntRange L, IntRange R) { 1751 return IntRange(std::max(L.Width, R.Width), 1752 L.NonNegative && R.NonNegative); 1753 } 1754 1755 // Returns the infinum of two ranges: i.e. their aggressive merge. 1756 static IntRange meet(IntRange L, IntRange R) { 1757 return IntRange(std::min(L.Width, R.Width), 1758 L.NonNegative || R.NonNegative); 1759 } 1760}; 1761 1762IntRange GetValueRange(ASTContext &C, llvm::APSInt &value, unsigned MaxWidth) { 1763 if (value.isSigned() && value.isNegative()) 1764 return IntRange(value.getMinSignedBits(), false); 1765 1766 if (value.getBitWidth() > MaxWidth) 1767 value.trunc(MaxWidth); 1768 1769 // isNonNegative() just checks the sign bit without considering 1770 // signedness. 1771 return IntRange(value.getActiveBits(), true); 1772} 1773 1774IntRange GetValueRange(ASTContext &C, APValue &result, QualType Ty, 1775 unsigned MaxWidth) { 1776 if (result.isInt()) 1777 return GetValueRange(C, result.getInt(), MaxWidth); 1778 1779 if (result.isVector()) { 1780 IntRange R = GetValueRange(C, result.getVectorElt(0), Ty, MaxWidth); 1781 for (unsigned i = 1, e = result.getVectorLength(); i != e; ++i) { 1782 IntRange El = GetValueRange(C, result.getVectorElt(i), Ty, MaxWidth); 1783 R = IntRange::join(R, El); 1784 } 1785 return R; 1786 } 1787 1788 if (result.isComplexInt()) { 1789 IntRange R = GetValueRange(C, result.getComplexIntReal(), MaxWidth); 1790 IntRange I = GetValueRange(C, result.getComplexIntImag(), MaxWidth); 1791 return IntRange::join(R, I); 1792 } 1793 1794 // This can happen with lossless casts to intptr_t of "based" lvalues. 1795 // Assume it might use arbitrary bits. 1796 // FIXME: The only reason we need to pass the type in here is to get 1797 // the sign right on this one case. It would be nice if APValue 1798 // preserved this. 1799 assert(result.isLValue()); 1800 return IntRange(MaxWidth, Ty->isUnsignedIntegerType()); 1801} 1802 1803/// Pseudo-evaluate the given integer expression, estimating the 1804/// range of values it might take. 1805/// 1806/// \param MaxWidth - the width to which the value will be truncated 1807IntRange GetExprRange(ASTContext &C, Expr *E, unsigned MaxWidth) { 1808 E = E->IgnoreParens(); 1809 1810 // Try a full evaluation first. 1811 Expr::EvalResult result; 1812 if (E->Evaluate(result, C)) 1813 return GetValueRange(C, result.Val, E->getType(), MaxWidth); 1814 1815 // I think we only want to look through implicit casts here; if the 1816 // user has an explicit widening cast, we should treat the value as 1817 // being of the new, wider type. 1818 if (ImplicitCastExpr *CE = dyn_cast<ImplicitCastExpr>(E)) { 1819 if (CE->getCastKind() == CastExpr::CK_NoOp) 1820 return GetExprRange(C, CE->getSubExpr(), MaxWidth); 1821 1822 IntRange OutputTypeRange = IntRange::forType(C, CE->getType()); 1823 1824 bool isIntegerCast = (CE->getCastKind() == CastExpr::CK_IntegralCast); 1825 if (!isIntegerCast && CE->getCastKind() == CastExpr::CK_Unknown) 1826 isIntegerCast = CE->getSubExpr()->getType()->isIntegerType(); 1827 1828 // Assume that non-integer casts can span the full range of the type. 1829 if (!isIntegerCast) 1830 return OutputTypeRange; 1831 1832 IntRange SubRange 1833 = GetExprRange(C, CE->getSubExpr(), 1834 std::min(MaxWidth, OutputTypeRange.Width)); 1835 1836 // Bail out if the subexpr's range is as wide as the cast type. 1837 if (SubRange.Width >= OutputTypeRange.Width) 1838 return OutputTypeRange; 1839 1840 // Otherwise, we take the smaller width, and we're non-negative if 1841 // either the output type or the subexpr is. 1842 return IntRange(SubRange.Width, 1843 SubRange.NonNegative || OutputTypeRange.NonNegative); 1844 } 1845 1846 if (ConditionalOperator *CO = dyn_cast<ConditionalOperator>(E)) { 1847 // If we can fold the condition, just take that operand. 1848 bool CondResult; 1849 if (CO->getCond()->EvaluateAsBooleanCondition(CondResult, C)) 1850 return GetExprRange(C, CondResult ? CO->getTrueExpr() 1851 : CO->getFalseExpr(), 1852 MaxWidth); 1853 1854 // Otherwise, conservatively merge. 1855 IntRange L = GetExprRange(C, CO->getTrueExpr(), MaxWidth); 1856 IntRange R = GetExprRange(C, CO->getFalseExpr(), MaxWidth); 1857 return IntRange::join(L, R); 1858 } 1859 1860 if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) { 1861 switch (BO->getOpcode()) { 1862 1863 // Boolean-valued operations are single-bit and positive. 1864 case BinaryOperator::LAnd: 1865 case BinaryOperator::LOr: 1866 case BinaryOperator::LT: 1867 case BinaryOperator::GT: 1868 case BinaryOperator::LE: 1869 case BinaryOperator::GE: 1870 case BinaryOperator::EQ: 1871 case BinaryOperator::NE: 1872 return IntRange::forBoolType(); 1873 1874 // The type of these compound assignments is the type of the LHS, 1875 // so the RHS is not necessarily an integer. 1876 case BinaryOperator::MulAssign: 1877 case BinaryOperator::DivAssign: 1878 case BinaryOperator::RemAssign: 1879 case BinaryOperator::AddAssign: 1880 case BinaryOperator::SubAssign: 1881 return IntRange::forType(C, E->getType()); 1882 1883 // Operations with opaque sources are black-listed. 1884 case BinaryOperator::PtrMemD: 1885 case BinaryOperator::PtrMemI: 1886 return IntRange::forType(C, E->getType()); 1887 1888 // Bitwise-and uses the *infinum* of the two source ranges. 1889 case BinaryOperator::And: 1890 case BinaryOperator::AndAssign: 1891 return IntRange::meet(GetExprRange(C, BO->getLHS(), MaxWidth), 1892 GetExprRange(C, BO->getRHS(), MaxWidth)); 1893 1894 // Left shift gets black-listed based on a judgement call. 1895 case BinaryOperator::Shl: 1896 // ...except that we want to treat '1 << (blah)' as logically 1897 // positive. It's an important idiom. 1898 if (IntegerLiteral *I 1899 = dyn_cast<IntegerLiteral>(BO->getLHS()->IgnoreParenCasts())) { 1900 if (I->getValue() == 1) { 1901 IntRange R = IntRange::forType(C, E->getType()); 1902 return IntRange(R.Width, /*NonNegative*/ true); 1903 } 1904 } 1905 // fallthrough 1906 1907 case BinaryOperator::ShlAssign: 1908 return IntRange::forType(C, E->getType()); 1909 1910 // Right shift by a constant can narrow its left argument. 1911 case BinaryOperator::Shr: 1912 case BinaryOperator::ShrAssign: { 1913 IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth); 1914 1915 // If the shift amount is a positive constant, drop the width by 1916 // that much. 1917 llvm::APSInt shift; 1918 if (BO->getRHS()->isIntegerConstantExpr(shift, C) && 1919 shift.isNonNegative()) { 1920 unsigned zext = shift.getZExtValue(); 1921 if (zext >= L.Width) 1922 L.Width = (L.NonNegative ? 0 : 1); 1923 else 1924 L.Width -= zext; 1925 } 1926 1927 return L; 1928 } 1929 1930 // Comma acts as its right operand. 1931 case BinaryOperator::Comma: 1932 return GetExprRange(C, BO->getRHS(), MaxWidth); 1933 1934 // Black-list pointer subtractions. 1935 case BinaryOperator::Sub: 1936 if (BO->getLHS()->getType()->isPointerType()) 1937 return IntRange::forType(C, E->getType()); 1938 // fallthrough 1939 1940 default: 1941 break; 1942 } 1943 1944 // Treat every other operator as if it were closed on the 1945 // narrowest type that encompasses both operands. 1946 IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth); 1947 IntRange R = GetExprRange(C, BO->getRHS(), MaxWidth); 1948 return IntRange::join(L, R); 1949 } 1950 1951 if (UnaryOperator *UO = dyn_cast<UnaryOperator>(E)) { 1952 switch (UO->getOpcode()) { 1953 // Boolean-valued operations are white-listed. 1954 case UnaryOperator::LNot: 1955 return IntRange::forBoolType(); 1956 1957 // Operations with opaque sources are black-listed. 1958 case UnaryOperator::Deref: 1959 case UnaryOperator::AddrOf: // should be impossible 1960 case UnaryOperator::OffsetOf: 1961 return IntRange::forType(C, E->getType()); 1962 1963 default: 1964 return GetExprRange(C, UO->getSubExpr(), MaxWidth); 1965 } 1966 } 1967 1968 if (dyn_cast<OffsetOfExpr>(E)) { 1969 IntRange::forType(C, E->getType()); 1970 } 1971 1972 FieldDecl *BitField = E->getBitField(); 1973 if (BitField) { 1974 llvm::APSInt BitWidthAP = BitField->getBitWidth()->EvaluateAsInt(C); 1975 unsigned BitWidth = BitWidthAP.getZExtValue(); 1976 1977 return IntRange(BitWidth, BitField->getType()->isUnsignedIntegerType()); 1978 } 1979 1980 return IntRange::forType(C, E->getType()); 1981} 1982 1983IntRange GetExprRange(ASTContext &C, Expr *E) { 1984 return GetExprRange(C, E, C.getIntWidth(E->getType())); 1985} 1986 1987/// Checks whether the given value, which currently has the given 1988/// source semantics, has the same value when coerced through the 1989/// target semantics. 1990bool IsSameFloatAfterCast(const llvm::APFloat &value, 1991 const llvm::fltSemantics &Src, 1992 const llvm::fltSemantics &Tgt) { 1993 llvm::APFloat truncated = value; 1994 1995 bool ignored; 1996 truncated.convert(Src, llvm::APFloat::rmNearestTiesToEven, &ignored); 1997 truncated.convert(Tgt, llvm::APFloat::rmNearestTiesToEven, &ignored); 1998 1999 return truncated.bitwiseIsEqual(value); 2000} 2001 2002/// Checks whether the given value, which currently has the given 2003/// source semantics, has the same value when coerced through the 2004/// target semantics. 2005/// 2006/// The value might be a vector of floats (or a complex number). 2007bool IsSameFloatAfterCast(const APValue &value, 2008 const llvm::fltSemantics &Src, 2009 const llvm::fltSemantics &Tgt) { 2010 if (value.isFloat()) 2011 return IsSameFloatAfterCast(value.getFloat(), Src, Tgt); 2012 2013 if (value.isVector()) { 2014 for (unsigned i = 0, e = value.getVectorLength(); i != e; ++i) 2015 if (!IsSameFloatAfterCast(value.getVectorElt(i), Src, Tgt)) 2016 return false; 2017 return true; 2018 } 2019 2020 assert(value.isComplexFloat()); 2021 return (IsSameFloatAfterCast(value.getComplexFloatReal(), Src, Tgt) && 2022 IsSameFloatAfterCast(value.getComplexFloatImag(), Src, Tgt)); 2023} 2024 2025void AnalyzeImplicitConversions(Sema &S, Expr *E); 2026 2027bool IsZero(Sema &S, Expr *E) { 2028 llvm::APSInt Value; 2029 return E->isIntegerConstantExpr(Value, S.Context) && Value == 0; 2030} 2031 2032void CheckTrivialUnsignedComparison(Sema &S, BinaryOperator *E) { 2033 BinaryOperator::Opcode op = E->getOpcode(); 2034 if (op == BinaryOperator::LT && IsZero(S, E->getRHS())) { 2035 S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison) 2036 << "< 0" << "false" 2037 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 2038 } else if (op == BinaryOperator::GE && IsZero(S, E->getRHS())) { 2039 S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison) 2040 << ">= 0" << "true" 2041 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 2042 } else if (op == BinaryOperator::GT && IsZero(S, E->getLHS())) { 2043 S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison) 2044 << "0 >" << "false" 2045 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 2046 } else if (op == BinaryOperator::LE && IsZero(S, E->getLHS())) { 2047 S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison) 2048 << "0 <=" << "true" 2049 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 2050 } 2051} 2052 2053/// Analyze the operands of the given comparison. Implements the 2054/// fallback case from AnalyzeComparison. 2055void AnalyzeImpConvsInComparison(Sema &S, BinaryOperator *E) { 2056 AnalyzeImplicitConversions(S, E->getLHS()); 2057 AnalyzeImplicitConversions(S, E->getRHS()); 2058} 2059 2060/// \brief Implements -Wsign-compare. 2061/// 2062/// \param lex the left-hand expression 2063/// \param rex the right-hand expression 2064/// \param OpLoc the location of the joining operator 2065/// \param BinOpc binary opcode or 0 2066void AnalyzeComparison(Sema &S, BinaryOperator *E) { 2067 // The type the comparison is being performed in. 2068 QualType T = E->getLHS()->getType(); 2069 assert(S.Context.hasSameUnqualifiedType(T, E->getRHS()->getType()) 2070 && "comparison with mismatched types"); 2071 2072 // We don't do anything special if this isn't an unsigned integral 2073 // comparison: we're only interested in integral comparisons, and 2074 // signed comparisons only happen in cases we don't care to warn about. 2075 if (!T->isUnsignedIntegerType()) 2076 return AnalyzeImpConvsInComparison(S, E); 2077 2078 Expr *lex = E->getLHS()->IgnoreParenImpCasts(); 2079 Expr *rex = E->getRHS()->IgnoreParenImpCasts(); 2080 2081 // Check to see if one of the (unmodified) operands is of different 2082 // signedness. 2083 Expr *signedOperand, *unsignedOperand; 2084 if (lex->getType()->isSignedIntegerType()) { 2085 assert(!rex->getType()->isSignedIntegerType() && 2086 "unsigned comparison between two signed integer expressions?"); 2087 signedOperand = lex; 2088 unsignedOperand = rex; 2089 } else if (rex->getType()->isSignedIntegerType()) { 2090 signedOperand = rex; 2091 unsignedOperand = lex; 2092 } else { 2093 CheckTrivialUnsignedComparison(S, E); 2094 return AnalyzeImpConvsInComparison(S, E); 2095 } 2096 2097 // Otherwise, calculate the effective range of the signed operand. 2098 IntRange signedRange = GetExprRange(S.Context, signedOperand); 2099 2100 // Go ahead and analyze implicit conversions in the operands. Note 2101 // that we skip the implicit conversions on both sides. 2102 AnalyzeImplicitConversions(S, lex); 2103 AnalyzeImplicitConversions(S, rex); 2104 2105 // If the signed range is non-negative, -Wsign-compare won't fire, 2106 // but we should still check for comparisons which are always true 2107 // or false. 2108 if (signedRange.NonNegative) 2109 return CheckTrivialUnsignedComparison(S, E); 2110 2111 // For (in)equality comparisons, if the unsigned operand is a 2112 // constant which cannot collide with a overflowed signed operand, 2113 // then reinterpreting the signed operand as unsigned will not 2114 // change the result of the comparison. 2115 if (E->isEqualityOp()) { 2116 unsigned comparisonWidth = S.Context.getIntWidth(T); 2117 IntRange unsignedRange = GetExprRange(S.Context, unsignedOperand); 2118 2119 // We should never be unable to prove that the unsigned operand is 2120 // non-negative. 2121 assert(unsignedRange.NonNegative && "unsigned range includes negative?"); 2122 2123 if (unsignedRange.Width < comparisonWidth) 2124 return; 2125 } 2126 2127 S.Diag(E->getOperatorLoc(), diag::warn_mixed_sign_comparison) 2128 << lex->getType() << rex->getType() 2129 << lex->getSourceRange() << rex->getSourceRange(); 2130} 2131 2132/// Diagnose an implicit cast; purely a helper for CheckImplicitConversion. 2133void DiagnoseImpCast(Sema &S, Expr *E, QualType T, unsigned diag) { 2134 S.Diag(E->getExprLoc(), diag) << E->getType() << T << E->getSourceRange(); 2135} 2136 2137void CheckImplicitConversion(Sema &S, Expr *E, QualType T, 2138 bool *ICContext = 0) { 2139 if (E->isTypeDependent() || E->isValueDependent()) return; 2140 2141 const Type *Source = S.Context.getCanonicalType(E->getType()).getTypePtr(); 2142 const Type *Target = S.Context.getCanonicalType(T).getTypePtr(); 2143 if (Source == Target) return; 2144 if (Target->isDependentType()) return; 2145 2146 // Never diagnose implicit casts to bool. 2147 if (Target->isSpecificBuiltinType(BuiltinType::Bool)) 2148 return; 2149 2150 // Strip vector types. 2151 if (isa<VectorType>(Source)) { 2152 if (!isa<VectorType>(Target)) 2153 return DiagnoseImpCast(S, E, T, diag::warn_impcast_vector_scalar); 2154 2155 Source = cast<VectorType>(Source)->getElementType().getTypePtr(); 2156 Target = cast<VectorType>(Target)->getElementType().getTypePtr(); 2157 } 2158 2159 // Strip complex types. 2160 if (isa<ComplexType>(Source)) { 2161 if (!isa<ComplexType>(Target)) 2162 return DiagnoseImpCast(S, E, T, diag::warn_impcast_complex_scalar); 2163 2164 Source = cast<ComplexType>(Source)->getElementType().getTypePtr(); 2165 Target = cast<ComplexType>(Target)->getElementType().getTypePtr(); 2166 } 2167 2168 const BuiltinType *SourceBT = dyn_cast<BuiltinType>(Source); 2169 const BuiltinType *TargetBT = dyn_cast<BuiltinType>(Target); 2170 2171 // If the source is floating point... 2172 if (SourceBT && SourceBT->isFloatingPoint()) { 2173 // ...and the target is floating point... 2174 if (TargetBT && TargetBT->isFloatingPoint()) { 2175 // ...then warn if we're dropping FP rank. 2176 2177 // Builtin FP kinds are ordered by increasing FP rank. 2178 if (SourceBT->getKind() > TargetBT->getKind()) { 2179 // Don't warn about float constants that are precisely 2180 // representable in the target type. 2181 Expr::EvalResult result; 2182 if (E->Evaluate(result, S.Context)) { 2183 // Value might be a float, a float vector, or a float complex. 2184 if (IsSameFloatAfterCast(result.Val, 2185 S.Context.getFloatTypeSemantics(QualType(TargetBT, 0)), 2186 S.Context.getFloatTypeSemantics(QualType(SourceBT, 0)))) 2187 return; 2188 } 2189 2190 DiagnoseImpCast(S, E, T, diag::warn_impcast_float_precision); 2191 } 2192 return; 2193 } 2194 2195 // If the target is integral, always warn. 2196 if ((TargetBT && TargetBT->isInteger())) 2197 // TODO: don't warn for integer values? 2198 DiagnoseImpCast(S, E, T, diag::warn_impcast_float_integer); 2199 2200 return; 2201 } 2202 2203 if (!Source->isIntegerType() || !Target->isIntegerType()) 2204 return; 2205 2206 IntRange SourceRange = GetExprRange(S.Context, E); 2207 IntRange TargetRange = IntRange::forCanonicalType(S.Context, Target); 2208 2209 if (SourceRange.Width > TargetRange.Width) { 2210 // People want to build with -Wshorten-64-to-32 and not -Wconversion 2211 // and by god we'll let them. 2212 if (SourceRange.Width == 64 && TargetRange.Width == 32) 2213 return DiagnoseImpCast(S, E, T, diag::warn_impcast_integer_64_32); 2214 return DiagnoseImpCast(S, E, T, diag::warn_impcast_integer_precision); 2215 } 2216 2217 if ((TargetRange.NonNegative && !SourceRange.NonNegative) || 2218 (!TargetRange.NonNegative && SourceRange.NonNegative && 2219 SourceRange.Width == TargetRange.Width)) { 2220 unsigned DiagID = diag::warn_impcast_integer_sign; 2221 2222 // Traditionally, gcc has warned about this under -Wsign-compare. 2223 // We also want to warn about it in -Wconversion. 2224 // So if -Wconversion is off, use a completely identical diagnostic 2225 // in the sign-compare group. 2226 // The conditional-checking code will 2227 if (ICContext) { 2228 DiagID = diag::warn_impcast_integer_sign_conditional; 2229 *ICContext = true; 2230 } 2231 2232 return DiagnoseImpCast(S, E, T, DiagID); 2233 } 2234 2235 return; 2236} 2237 2238void CheckConditionalOperator(Sema &S, ConditionalOperator *E, QualType T); 2239 2240void CheckConditionalOperand(Sema &S, Expr *E, QualType T, 2241 bool &ICContext) { 2242 E = E->IgnoreParenImpCasts(); 2243 2244 if (isa<ConditionalOperator>(E)) 2245 return CheckConditionalOperator(S, cast<ConditionalOperator>(E), T); 2246 2247 AnalyzeImplicitConversions(S, E); 2248 if (E->getType() != T) 2249 return CheckImplicitConversion(S, E, T, &ICContext); 2250 return; 2251} 2252 2253void CheckConditionalOperator(Sema &S, ConditionalOperator *E, QualType T) { 2254 AnalyzeImplicitConversions(S, E->getCond()); 2255 2256 bool Suspicious = false; 2257 CheckConditionalOperand(S, E->getTrueExpr(), T, Suspicious); 2258 CheckConditionalOperand(S, E->getFalseExpr(), T, Suspicious); 2259 2260 // If -Wconversion would have warned about either of the candidates 2261 // for a signedness conversion to the context type... 2262 if (!Suspicious) return; 2263 2264 // ...but it's currently ignored... 2265 if (S.Diags.getDiagnosticLevel(diag::warn_impcast_integer_sign_conditional)) 2266 return; 2267 2268 // ...and -Wsign-compare isn't... 2269 if (!S.Diags.getDiagnosticLevel(diag::warn_mixed_sign_conditional)) 2270 return; 2271 2272 // ...then check whether it would have warned about either of the 2273 // candidates for a signedness conversion to the condition type. 2274 if (E->getType() != T) { 2275 Suspicious = false; 2276 CheckImplicitConversion(S, E->getTrueExpr()->IgnoreParenImpCasts(), 2277 E->getType(), &Suspicious); 2278 if (!Suspicious) 2279 CheckImplicitConversion(S, E->getFalseExpr()->IgnoreParenImpCasts(), 2280 E->getType(), &Suspicious); 2281 if (!Suspicious) 2282 return; 2283 } 2284 2285 // If so, emit a diagnostic under -Wsign-compare. 2286 Expr *lex = E->getTrueExpr()->IgnoreParenImpCasts(); 2287 Expr *rex = E->getFalseExpr()->IgnoreParenImpCasts(); 2288 S.Diag(E->getQuestionLoc(), diag::warn_mixed_sign_conditional) 2289 << lex->getType() << rex->getType() 2290 << lex->getSourceRange() << rex->getSourceRange(); 2291} 2292 2293/// AnalyzeImplicitConversions - Find and report any interesting 2294/// implicit conversions in the given expression. There are a couple 2295/// of competing diagnostics here, -Wconversion and -Wsign-compare. 2296void AnalyzeImplicitConversions(Sema &S, Expr *OrigE) { 2297 QualType T = OrigE->getType(); 2298 Expr *E = OrigE->IgnoreParenImpCasts(); 2299 2300 // For conditional operators, we analyze the arguments as if they 2301 // were being fed directly into the output. 2302 if (isa<ConditionalOperator>(E)) { 2303 ConditionalOperator *CO = cast<ConditionalOperator>(E); 2304 CheckConditionalOperator(S, CO, T); 2305 return; 2306 } 2307 2308 // Go ahead and check any implicit conversions we might have skipped. 2309 // The non-canonical typecheck is just an optimization; 2310 // CheckImplicitConversion will filter out dead implicit conversions. 2311 if (E->getType() != T) 2312 CheckImplicitConversion(S, E, T); 2313 2314 // Now continue drilling into this expression. 2315 2316 // Skip past explicit casts. 2317 if (isa<ExplicitCastExpr>(E)) { 2318 E = cast<ExplicitCastExpr>(E)->getSubExpr()->IgnoreParenImpCasts(); 2319 return AnalyzeImplicitConversions(S, E); 2320 } 2321 2322 // Do a somewhat different check with comparison operators. 2323 if (isa<BinaryOperator>(E) && cast<BinaryOperator>(E)->isComparisonOp()) 2324 return AnalyzeComparison(S, cast<BinaryOperator>(E)); 2325 2326 // These break the otherwise-useful invariant below. Fortunately, 2327 // we don't really need to recurse into them, because any internal 2328 // expressions should have been analyzed already when they were 2329 // built into statements. 2330 if (isa<StmtExpr>(E)) return; 2331 2332 // Don't descend into unevaluated contexts. 2333 if (isa<SizeOfAlignOfExpr>(E)) return; 2334 2335 // Now just recurse over the expression's children. 2336 for (Stmt::child_iterator I = E->child_begin(), IE = E->child_end(); 2337 I != IE; ++I) 2338 AnalyzeImplicitConversions(S, cast<Expr>(*I)); 2339} 2340 2341} // end anonymous namespace 2342 2343/// Diagnoses "dangerous" implicit conversions within the given 2344/// expression (which is a full expression). Implements -Wconversion 2345/// and -Wsign-compare. 2346void Sema::CheckImplicitConversions(Expr *E) { 2347 // Don't diagnose in unevaluated contexts. 2348 if (ExprEvalContexts.back().Context == Sema::Unevaluated) 2349 return; 2350 2351 // Don't diagnose for value- or type-dependent expressions. 2352 if (E->isTypeDependent() || E->isValueDependent()) 2353 return; 2354 2355 AnalyzeImplicitConversions(*this, E); 2356} 2357 2358/// CheckParmsForFunctionDef - Check that the parameters of the given 2359/// function are appropriate for the definition of a function. This 2360/// takes care of any checks that cannot be performed on the 2361/// declaration itself, e.g., that the types of each of the function 2362/// parameters are complete. 2363bool Sema::CheckParmsForFunctionDef(FunctionDecl *FD) { 2364 bool HasInvalidParm = false; 2365 for (unsigned p = 0, NumParams = FD->getNumParams(); p < NumParams; ++p) { 2366 ParmVarDecl *Param = FD->getParamDecl(p); 2367 2368 // C99 6.7.5.3p4: the parameters in a parameter type list in a 2369 // function declarator that is part of a function definition of 2370 // that function shall not have incomplete type. 2371 // 2372 // This is also C++ [dcl.fct]p6. 2373 if (!Param->isInvalidDecl() && 2374 RequireCompleteType(Param->getLocation(), Param->getType(), 2375 diag::err_typecheck_decl_incomplete_type)) { 2376 Param->setInvalidDecl(); 2377 HasInvalidParm = true; 2378 } 2379 2380 // C99 6.9.1p5: If the declarator includes a parameter type list, the 2381 // declaration of each parameter shall include an identifier. 2382 if (Param->getIdentifier() == 0 && 2383 !Param->isImplicit() && 2384 !getLangOptions().CPlusPlus) 2385 Diag(Param->getLocation(), diag::err_parameter_name_omitted); 2386 2387 // C99 6.7.5.3p12: 2388 // If the function declarator is not part of a definition of that 2389 // function, parameters may have incomplete type and may use the [*] 2390 // notation in their sequences of declarator specifiers to specify 2391 // variable length array types. 2392 QualType PType = Param->getOriginalType(); 2393 if (const ArrayType *AT = Context.getAsArrayType(PType)) { 2394 if (AT->getSizeModifier() == ArrayType::Star) { 2395 // FIXME: This diagnosic should point the the '[*]' if source-location 2396 // information is added for it. 2397 Diag(Param->getLocation(), diag::err_array_star_in_function_definition); 2398 } 2399 } 2400 } 2401 2402 return HasInvalidParm; 2403} 2404