SemaChecking.cpp revision 8fee33e7925d771f4b410744735a52122ce5b315
1//===--- SemaChecking.cpp - Extra Semantic Checking -----------------------===// 2// 3// The LLVM Compiler Infrastructure 4// 5// This file is distributed under the University of Illinois Open Source 6// License. See LICENSE.TXT for details. 7// 8//===----------------------------------------------------------------------===// 9// 10// This file implements extra semantic analysis beyond what is enforced 11// by the C type system. 12// 13//===----------------------------------------------------------------------===// 14 15#include "clang/Sema/Sema.h" 16#include "clang/Sema/SemaInternal.h" 17#include "clang/Sema/ScopeInfo.h" 18#include "clang/Analysis/Analyses/FormatString.h" 19#include "clang/AST/ASTContext.h" 20#include "clang/AST/CharUnits.h" 21#include "clang/AST/DeclCXX.h" 22#include "clang/AST/DeclObjC.h" 23#include "clang/AST/ExprCXX.h" 24#include "clang/AST/ExprObjC.h" 25#include "clang/AST/DeclObjC.h" 26#include "clang/AST/StmtCXX.h" 27#include "clang/AST/StmtObjC.h" 28#include "clang/Lex/LiteralSupport.h" 29#include "clang/Lex/Preprocessor.h" 30#include "llvm/ADT/BitVector.h" 31#include "llvm/ADT/STLExtras.h" 32#include "llvm/Support/raw_ostream.h" 33#include "clang/Basic/TargetBuiltins.h" 34#include "clang/Basic/TargetInfo.h" 35#include "clang/Basic/ConvertUTF.h" 36 37#include <limits> 38using namespace clang; 39using namespace sema; 40 41/// getLocationOfStringLiteralByte - Return a source location that points to the 42/// specified byte of the specified string literal. 43/// 44/// Strings are amazingly complex. They can be formed from multiple tokens and 45/// can have escape sequences in them in addition to the usual trigraph and 46/// escaped newline business. This routine handles this complexity. 47/// 48SourceLocation Sema::getLocationOfStringLiteralByte(const StringLiteral *SL, 49 unsigned ByteNo) const { 50 assert(!SL->isWide() && "This doesn't work for wide strings yet"); 51 52 // Loop over all of the tokens in this string until we find the one that 53 // contains the byte we're looking for. 54 unsigned TokNo = 0; 55 while (1) { 56 assert(TokNo < SL->getNumConcatenated() && "Invalid byte number!"); 57 SourceLocation StrTokLoc = SL->getStrTokenLoc(TokNo); 58 59 // Get the spelling of the string so that we can get the data that makes up 60 // the string literal, not the identifier for the macro it is potentially 61 // expanded through. 62 SourceLocation StrTokSpellingLoc = SourceMgr.getSpellingLoc(StrTokLoc); 63 64 // Re-lex the token to get its length and original spelling. 65 std::pair<FileID, unsigned> LocInfo = 66 SourceMgr.getDecomposedLoc(StrTokSpellingLoc); 67 bool Invalid = false; 68 llvm::StringRef Buffer = SourceMgr.getBufferData(LocInfo.first, &Invalid); 69 if (Invalid) 70 return StrTokSpellingLoc; 71 72 const char *StrData = Buffer.data()+LocInfo.second; 73 74 // Create a langops struct and enable trigraphs. This is sufficient for 75 // relexing tokens. 76 LangOptions LangOpts; 77 LangOpts.Trigraphs = true; 78 79 // Create a lexer starting at the beginning of this token. 80 Lexer TheLexer(StrTokSpellingLoc, LangOpts, Buffer.begin(), StrData, 81 Buffer.end()); 82 Token TheTok; 83 TheLexer.LexFromRawLexer(TheTok); 84 85 // Use the StringLiteralParser to compute the length of the string in bytes. 86 StringLiteralParser SLP(&TheTok, 1, PP, /*Complain=*/false); 87 unsigned TokNumBytes = SLP.GetStringLength(); 88 89 // If the byte is in this token, return the location of the byte. 90 if (ByteNo < TokNumBytes || 91 (ByteNo == TokNumBytes && TokNo == SL->getNumConcatenated())) { 92 unsigned Offset = 93 StringLiteralParser::getOffsetOfStringByte(TheTok, ByteNo, PP, 94 /*Complain=*/false); 95 96 // Now that we know the offset of the token in the spelling, use the 97 // preprocessor to get the offset in the original source. 98 return PP.AdvanceToTokenCharacter(StrTokLoc, Offset); 99 } 100 101 // Move to the next string token. 102 ++TokNo; 103 ByteNo -= TokNumBytes; 104 } 105} 106 107/// CheckablePrintfAttr - does a function call have a "printf" attribute 108/// and arguments that merit checking? 109bool Sema::CheckablePrintfAttr(const FormatAttr *Format, CallExpr *TheCall) { 110 if (Format->getType() == "printf") return true; 111 if (Format->getType() == "printf0") { 112 // printf0 allows null "format" string; if so don't check format/args 113 unsigned format_idx = Format->getFormatIdx() - 1; 114 // Does the index refer to the implicit object argument? 115 if (isa<CXXMemberCallExpr>(TheCall)) { 116 if (format_idx == 0) 117 return false; 118 --format_idx; 119 } 120 if (format_idx < TheCall->getNumArgs()) { 121 Expr *Format = TheCall->getArg(format_idx)->IgnoreParenCasts(); 122 if (!Format->isNullPointerConstant(Context, 123 Expr::NPC_ValueDependentIsNull)) 124 return true; 125 } 126 } 127 return false; 128} 129 130ExprResult 131Sema::CheckBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) { 132 ExprResult TheCallResult(Owned(TheCall)); 133 134 switch (BuiltinID) { 135 case Builtin::BI__builtin___CFStringMakeConstantString: 136 assert(TheCall->getNumArgs() == 1 && 137 "Wrong # arguments to builtin CFStringMakeConstantString"); 138 if (CheckObjCString(TheCall->getArg(0))) 139 return ExprError(); 140 break; 141 case Builtin::BI__builtin_stdarg_start: 142 case Builtin::BI__builtin_va_start: 143 if (SemaBuiltinVAStart(TheCall)) 144 return ExprError(); 145 break; 146 case Builtin::BI__builtin_isgreater: 147 case Builtin::BI__builtin_isgreaterequal: 148 case Builtin::BI__builtin_isless: 149 case Builtin::BI__builtin_islessequal: 150 case Builtin::BI__builtin_islessgreater: 151 case Builtin::BI__builtin_isunordered: 152 if (SemaBuiltinUnorderedCompare(TheCall)) 153 return ExprError(); 154 break; 155 case Builtin::BI__builtin_fpclassify: 156 if (SemaBuiltinFPClassification(TheCall, 6)) 157 return ExprError(); 158 break; 159 case Builtin::BI__builtin_isfinite: 160 case Builtin::BI__builtin_isinf: 161 case Builtin::BI__builtin_isinf_sign: 162 case Builtin::BI__builtin_isnan: 163 case Builtin::BI__builtin_isnormal: 164 if (SemaBuiltinFPClassification(TheCall, 1)) 165 return ExprError(); 166 break; 167 case Builtin::BI__builtin_return_address: 168 case Builtin::BI__builtin_frame_address: { 169 llvm::APSInt Result; 170 if (SemaBuiltinConstantArg(TheCall, 0, Result)) 171 return ExprError(); 172 break; 173 } 174 case Builtin::BI__builtin_eh_return_data_regno: { 175 llvm::APSInt Result; 176 if (SemaBuiltinConstantArg(TheCall, 0, Result)) 177 return ExprError(); 178 break; 179 } 180 case Builtin::BI__builtin_shufflevector: 181 return SemaBuiltinShuffleVector(TheCall); 182 // TheCall will be freed by the smart pointer here, but that's fine, since 183 // SemaBuiltinShuffleVector guts it, but then doesn't release it. 184 case Builtin::BI__builtin_prefetch: 185 if (SemaBuiltinPrefetch(TheCall)) 186 return ExprError(); 187 break; 188 case Builtin::BI__builtin_object_size: 189 if (SemaBuiltinObjectSize(TheCall)) 190 return ExprError(); 191 break; 192 case Builtin::BI__builtin_longjmp: 193 if (SemaBuiltinLongjmp(TheCall)) 194 return ExprError(); 195 break; 196 case Builtin::BI__sync_fetch_and_add: 197 case Builtin::BI__sync_fetch_and_sub: 198 case Builtin::BI__sync_fetch_and_or: 199 case Builtin::BI__sync_fetch_and_and: 200 case Builtin::BI__sync_fetch_and_xor: 201 case Builtin::BI__sync_add_and_fetch: 202 case Builtin::BI__sync_sub_and_fetch: 203 case Builtin::BI__sync_and_and_fetch: 204 case Builtin::BI__sync_or_and_fetch: 205 case Builtin::BI__sync_xor_and_fetch: 206 case Builtin::BI__sync_val_compare_and_swap: 207 case Builtin::BI__sync_bool_compare_and_swap: 208 case Builtin::BI__sync_lock_test_and_set: 209 case Builtin::BI__sync_lock_release: 210 return SemaBuiltinAtomicOverloaded(move(TheCallResult)); 211 } 212 213 // Since the target specific builtins for each arch overlap, only check those 214 // of the arch we are compiling for. 215 if (BuiltinID >= Builtin::FirstTSBuiltin) { 216 switch (Context.Target.getTriple().getArch()) { 217 case llvm::Triple::arm: 218 case llvm::Triple::thumb: 219 if (CheckARMBuiltinFunctionCall(BuiltinID, TheCall)) 220 return ExprError(); 221 break; 222 case llvm::Triple::x86: 223 case llvm::Triple::x86_64: 224 if (CheckX86BuiltinFunctionCall(BuiltinID, TheCall)) 225 return ExprError(); 226 break; 227 default: 228 break; 229 } 230 } 231 232 return move(TheCallResult); 233} 234 235bool Sema::CheckX86BuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) { 236 switch (BuiltinID) { 237 case X86::BI__builtin_ia32_palignr128: 238 case X86::BI__builtin_ia32_palignr: { 239 llvm::APSInt Result; 240 if (SemaBuiltinConstantArg(TheCall, 2, Result)) 241 return true; 242 break; 243 } 244 } 245 return false; 246} 247 248// Get the valid immediate range for the specified NEON type code. 249static unsigned RFT(unsigned t, bool shift = false) { 250 bool quad = t & 0x10; 251 252 switch (t & 0x7) { 253 case 0: // i8 254 return shift ? 7 : (8 << (int)quad) - 1; 255 case 1: // i16 256 return shift ? 15 : (4 << (int)quad) - 1; 257 case 2: // i32 258 return shift ? 31 : (2 << (int)quad) - 1; 259 case 3: // i64 260 return shift ? 63 : (1 << (int)quad) - 1; 261 case 4: // f32 262 assert(!shift && "cannot shift float types!"); 263 return (2 << (int)quad) - 1; 264 case 5: // poly8 265 assert(!shift && "cannot shift polynomial types!"); 266 return (8 << (int)quad) - 1; 267 case 6: // poly16 268 assert(!shift && "cannot shift polynomial types!"); 269 return (4 << (int)quad) - 1; 270 case 7: // float16 271 assert(!shift && "cannot shift float types!"); 272 return (4 << (int)quad) - 1; 273 } 274 return 0; 275} 276 277bool Sema::CheckARMBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) { 278 llvm::APSInt Result; 279 280 unsigned mask = 0; 281 unsigned TV = 0; 282 switch (BuiltinID) { 283#define GET_NEON_OVERLOAD_CHECK 284#include "clang/Basic/arm_neon.inc" 285#undef GET_NEON_OVERLOAD_CHECK 286 } 287 288 // For NEON intrinsics which are overloaded on vector element type, validate 289 // the immediate which specifies which variant to emit. 290 if (mask) { 291 unsigned ArgNo = TheCall->getNumArgs()-1; 292 if (SemaBuiltinConstantArg(TheCall, ArgNo, Result)) 293 return true; 294 295 TV = Result.getLimitedValue(32); 296 if ((TV > 31) || (mask & (1 << TV)) == 0) 297 return Diag(TheCall->getLocStart(), diag::err_invalid_neon_type_code) 298 << TheCall->getArg(ArgNo)->getSourceRange(); 299 } 300 301 // For NEON intrinsics which take an immediate value as part of the 302 // instruction, range check them here. 303 unsigned i = 0, l = 0, u = 0; 304 switch (BuiltinID) { 305 default: return false; 306 case ARM::BI__builtin_arm_ssat: i = 1; l = 1; u = 31; break; 307 case ARM::BI__builtin_arm_usat: i = 1; u = 31; break; 308 case ARM::BI__builtin_arm_vcvtr_f: 309 case ARM::BI__builtin_arm_vcvtr_d: i = 1; u = 1; break; 310#define GET_NEON_IMMEDIATE_CHECK 311#include "clang/Basic/arm_neon.inc" 312#undef GET_NEON_IMMEDIATE_CHECK 313 }; 314 315 // Check that the immediate argument is actually a constant. 316 if (SemaBuiltinConstantArg(TheCall, i, Result)) 317 return true; 318 319 // Range check against the upper/lower values for this isntruction. 320 unsigned Val = Result.getZExtValue(); 321 if (Val < l || Val > (u + l)) 322 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 323 << l << u+l << TheCall->getArg(i)->getSourceRange(); 324 325 // FIXME: VFP Intrinsics should error if VFP not present. 326 return false; 327} 328 329/// CheckFunctionCall - Check a direct function call for various correctness 330/// and safety properties not strictly enforced by the C type system. 331bool Sema::CheckFunctionCall(FunctionDecl *FDecl, CallExpr *TheCall) { 332 // Get the IdentifierInfo* for the called function. 333 IdentifierInfo *FnInfo = FDecl->getIdentifier(); 334 335 // None of the checks below are needed for functions that don't have 336 // simple names (e.g., C++ conversion functions). 337 if (!FnInfo) 338 return false; 339 340 // FIXME: This mechanism should be abstracted to be less fragile and 341 // more efficient. For example, just map function ids to custom 342 // handlers. 343 344 // Printf checking. 345 if (const FormatAttr *Format = FDecl->getAttr<FormatAttr>()) { 346 const bool b = Format->getType() == "scanf"; 347 if (b || CheckablePrintfAttr(Format, TheCall)) { 348 bool HasVAListArg = Format->getFirstArg() == 0; 349 CheckPrintfScanfArguments(TheCall, HasVAListArg, 350 Format->getFormatIdx() - 1, 351 HasVAListArg ? 0 : Format->getFirstArg() - 1, 352 !b); 353 } 354 } 355 356 specific_attr_iterator<NonNullAttr> 357 i = FDecl->specific_attr_begin<NonNullAttr>(), 358 e = FDecl->specific_attr_end<NonNullAttr>(); 359 360 for (; i != e; ++i) 361 CheckNonNullArguments(*i, TheCall); 362 363 return false; 364} 365 366bool Sema::CheckBlockCall(NamedDecl *NDecl, CallExpr *TheCall) { 367 // Printf checking. 368 const FormatAttr *Format = NDecl->getAttr<FormatAttr>(); 369 if (!Format) 370 return false; 371 372 const VarDecl *V = dyn_cast<VarDecl>(NDecl); 373 if (!V) 374 return false; 375 376 QualType Ty = V->getType(); 377 if (!Ty->isBlockPointerType()) 378 return false; 379 380 const bool b = Format->getType() == "scanf"; 381 if (!b && !CheckablePrintfAttr(Format, TheCall)) 382 return false; 383 384 bool HasVAListArg = Format->getFirstArg() == 0; 385 CheckPrintfScanfArguments(TheCall, HasVAListArg, Format->getFormatIdx() - 1, 386 HasVAListArg ? 0 : Format->getFirstArg() - 1, !b); 387 388 return false; 389} 390 391/// SemaBuiltinAtomicOverloaded - We have a call to a function like 392/// __sync_fetch_and_add, which is an overloaded function based on the pointer 393/// type of its first argument. The main ActOnCallExpr routines have already 394/// promoted the types of arguments because all of these calls are prototyped as 395/// void(...). 396/// 397/// This function goes through and does final semantic checking for these 398/// builtins, 399ExprResult 400Sema::SemaBuiltinAtomicOverloaded(ExprResult TheCallResult) { 401 CallExpr *TheCall = (CallExpr *)TheCallResult.get(); 402 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 403 FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl()); 404 405 // Ensure that we have at least one argument to do type inference from. 406 if (TheCall->getNumArgs() < 1) { 407 Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args_at_least) 408 << 0 << 1 << TheCall->getNumArgs() 409 << TheCall->getCallee()->getSourceRange(); 410 return ExprError(); 411 } 412 413 // Inspect the first argument of the atomic builtin. This should always be 414 // a pointer type, whose element is an integral scalar or pointer type. 415 // Because it is a pointer type, we don't have to worry about any implicit 416 // casts here. 417 // FIXME: We don't allow floating point scalars as input. 418 Expr *FirstArg = TheCall->getArg(0); 419 if (!FirstArg->getType()->isPointerType()) { 420 Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer) 421 << FirstArg->getType() << FirstArg->getSourceRange(); 422 return ExprError(); 423 } 424 425 QualType ValType = 426 FirstArg->getType()->getAs<PointerType>()->getPointeeType(); 427 if (!ValType->isIntegerType() && !ValType->isPointerType() && 428 !ValType->isBlockPointerType()) { 429 Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer_intptr) 430 << FirstArg->getType() << FirstArg->getSourceRange(); 431 return ExprError(); 432 } 433 434 // The majority of builtins return a value, but a few have special return 435 // types, so allow them to override appropriately below. 436 QualType ResultType = ValType; 437 438 // We need to figure out which concrete builtin this maps onto. For example, 439 // __sync_fetch_and_add with a 2 byte object turns into 440 // __sync_fetch_and_add_2. 441#define BUILTIN_ROW(x) \ 442 { Builtin::BI##x##_1, Builtin::BI##x##_2, Builtin::BI##x##_4, \ 443 Builtin::BI##x##_8, Builtin::BI##x##_16 } 444 445 static const unsigned BuiltinIndices[][5] = { 446 BUILTIN_ROW(__sync_fetch_and_add), 447 BUILTIN_ROW(__sync_fetch_and_sub), 448 BUILTIN_ROW(__sync_fetch_and_or), 449 BUILTIN_ROW(__sync_fetch_and_and), 450 BUILTIN_ROW(__sync_fetch_and_xor), 451 452 BUILTIN_ROW(__sync_add_and_fetch), 453 BUILTIN_ROW(__sync_sub_and_fetch), 454 BUILTIN_ROW(__sync_and_and_fetch), 455 BUILTIN_ROW(__sync_or_and_fetch), 456 BUILTIN_ROW(__sync_xor_and_fetch), 457 458 BUILTIN_ROW(__sync_val_compare_and_swap), 459 BUILTIN_ROW(__sync_bool_compare_and_swap), 460 BUILTIN_ROW(__sync_lock_test_and_set), 461 BUILTIN_ROW(__sync_lock_release) 462 }; 463#undef BUILTIN_ROW 464 465 // Determine the index of the size. 466 unsigned SizeIndex; 467 switch (Context.getTypeSizeInChars(ValType).getQuantity()) { 468 case 1: SizeIndex = 0; break; 469 case 2: SizeIndex = 1; break; 470 case 4: SizeIndex = 2; break; 471 case 8: SizeIndex = 3; break; 472 case 16: SizeIndex = 4; break; 473 default: 474 Diag(DRE->getLocStart(), diag::err_atomic_builtin_pointer_size) 475 << FirstArg->getType() << FirstArg->getSourceRange(); 476 return ExprError(); 477 } 478 479 // Each of these builtins has one pointer argument, followed by some number of 480 // values (0, 1 or 2) followed by a potentially empty varags list of stuff 481 // that we ignore. Find out which row of BuiltinIndices to read from as well 482 // as the number of fixed args. 483 unsigned BuiltinID = FDecl->getBuiltinID(); 484 unsigned BuiltinIndex, NumFixed = 1; 485 switch (BuiltinID) { 486 default: assert(0 && "Unknown overloaded atomic builtin!"); 487 case Builtin::BI__sync_fetch_and_add: BuiltinIndex = 0; break; 488 case Builtin::BI__sync_fetch_and_sub: BuiltinIndex = 1; break; 489 case Builtin::BI__sync_fetch_and_or: BuiltinIndex = 2; break; 490 case Builtin::BI__sync_fetch_and_and: BuiltinIndex = 3; break; 491 case Builtin::BI__sync_fetch_and_xor: BuiltinIndex = 4; break; 492 493 case Builtin::BI__sync_add_and_fetch: BuiltinIndex = 5; break; 494 case Builtin::BI__sync_sub_and_fetch: BuiltinIndex = 6; break; 495 case Builtin::BI__sync_and_and_fetch: BuiltinIndex = 7; break; 496 case Builtin::BI__sync_or_and_fetch: BuiltinIndex = 8; break; 497 case Builtin::BI__sync_xor_and_fetch: BuiltinIndex = 9; break; 498 499 case Builtin::BI__sync_val_compare_and_swap: 500 BuiltinIndex = 10; 501 NumFixed = 2; 502 break; 503 case Builtin::BI__sync_bool_compare_and_swap: 504 BuiltinIndex = 11; 505 NumFixed = 2; 506 ResultType = Context.BoolTy; 507 break; 508 case Builtin::BI__sync_lock_test_and_set: BuiltinIndex = 12; break; 509 case Builtin::BI__sync_lock_release: 510 BuiltinIndex = 13; 511 NumFixed = 0; 512 ResultType = Context.VoidTy; 513 break; 514 } 515 516 // Now that we know how many fixed arguments we expect, first check that we 517 // have at least that many. 518 if (TheCall->getNumArgs() < 1+NumFixed) { 519 Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args_at_least) 520 << 0 << 1+NumFixed << TheCall->getNumArgs() 521 << TheCall->getCallee()->getSourceRange(); 522 return ExprError(); 523 } 524 525 // Get the decl for the concrete builtin from this, we can tell what the 526 // concrete integer type we should convert to is. 527 unsigned NewBuiltinID = BuiltinIndices[BuiltinIndex][SizeIndex]; 528 const char *NewBuiltinName = Context.BuiltinInfo.GetName(NewBuiltinID); 529 IdentifierInfo *NewBuiltinII = PP.getIdentifierInfo(NewBuiltinName); 530 FunctionDecl *NewBuiltinDecl = 531 cast<FunctionDecl>(LazilyCreateBuiltin(NewBuiltinII, NewBuiltinID, 532 TUScope, false, DRE->getLocStart())); 533 534 // The first argument --- the pointer --- has a fixed type; we 535 // deduce the types of the rest of the arguments accordingly. Walk 536 // the remaining arguments, converting them to the deduced value type. 537 for (unsigned i = 0; i != NumFixed; ++i) { 538 Expr *Arg = TheCall->getArg(i+1); 539 540 // If the argument is an implicit cast, then there was a promotion due to 541 // "...", just remove it now. 542 if (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(Arg)) { 543 Arg = ICE->getSubExpr(); 544 ICE->setSubExpr(0); 545 TheCall->setArg(i+1, Arg); 546 } 547 548 // GCC does an implicit conversion to the pointer or integer ValType. This 549 // can fail in some cases (1i -> int**), check for this error case now. 550 CastKind Kind = CK_Unknown; 551 CXXCastPath BasePath; 552 if (CheckCastTypes(Arg->getSourceRange(), ValType, Arg, Kind, BasePath)) 553 return ExprError(); 554 555 // Okay, we have something that *can* be converted to the right type. Check 556 // to see if there is a potentially weird extension going on here. This can 557 // happen when you do an atomic operation on something like an char* and 558 // pass in 42. The 42 gets converted to char. This is even more strange 559 // for things like 45.123 -> char, etc. 560 // FIXME: Do this check. 561 ImpCastExprToType(Arg, ValType, Kind, VK_RValue, &BasePath); 562 TheCall->setArg(i+1, Arg); 563 } 564 565 // Switch the DeclRefExpr to refer to the new decl. 566 DRE->setDecl(NewBuiltinDecl); 567 DRE->setType(NewBuiltinDecl->getType()); 568 569 // Set the callee in the CallExpr. 570 // FIXME: This leaks the original parens and implicit casts. 571 Expr *PromotedCall = DRE; 572 UsualUnaryConversions(PromotedCall); 573 TheCall->setCallee(PromotedCall); 574 575 // Change the result type of the call to match the original value type. This 576 // is arbitrary, but the codegen for these builtins ins design to handle it 577 // gracefully. 578 TheCall->setType(ResultType); 579 580 return move(TheCallResult); 581} 582 583 584/// CheckObjCString - Checks that the argument to the builtin 585/// CFString constructor is correct 586/// Note: It might also make sense to do the UTF-16 conversion here (would 587/// simplify the backend). 588bool Sema::CheckObjCString(Expr *Arg) { 589 Arg = Arg->IgnoreParenCasts(); 590 StringLiteral *Literal = dyn_cast<StringLiteral>(Arg); 591 592 if (!Literal || Literal->isWide()) { 593 Diag(Arg->getLocStart(), diag::err_cfstring_literal_not_string_constant) 594 << Arg->getSourceRange(); 595 return true; 596 } 597 598 size_t NulPos = Literal->getString().find('\0'); 599 if (NulPos != llvm::StringRef::npos) { 600 Diag(getLocationOfStringLiteralByte(Literal, NulPos), 601 diag::warn_cfstring_literal_contains_nul_character) 602 << Arg->getSourceRange(); 603 } 604 if (Literal->containsNonAsciiOrNull()) { 605 llvm::StringRef String = Literal->getString(); 606 unsigned NumBytes = String.size(); 607 llvm::SmallVector<UTF16, 128> ToBuf(NumBytes); 608 const UTF8 *FromPtr = (UTF8 *)String.data(); 609 UTF16 *ToPtr = &ToBuf[0]; 610 611 ConversionResult Result = ConvertUTF8toUTF16(&FromPtr, FromPtr + NumBytes, 612 &ToPtr, ToPtr + NumBytes, 613 strictConversion); 614 // Check for conversion failure. 615 if (Result != conversionOK) 616 Diag(Arg->getLocStart(), 617 diag::warn_cfstring_truncated) << Arg->getSourceRange(); 618 } 619 return false; 620} 621 622/// SemaBuiltinVAStart - Check the arguments to __builtin_va_start for validity. 623/// Emit an error and return true on failure, return false on success. 624bool Sema::SemaBuiltinVAStart(CallExpr *TheCall) { 625 Expr *Fn = TheCall->getCallee(); 626 if (TheCall->getNumArgs() > 2) { 627 Diag(TheCall->getArg(2)->getLocStart(), 628 diag::err_typecheck_call_too_many_args) 629 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 630 << Fn->getSourceRange() 631 << SourceRange(TheCall->getArg(2)->getLocStart(), 632 (*(TheCall->arg_end()-1))->getLocEnd()); 633 return true; 634 } 635 636 if (TheCall->getNumArgs() < 2) { 637 return Diag(TheCall->getLocEnd(), 638 diag::err_typecheck_call_too_few_args_at_least) 639 << 0 /*function call*/ << 2 << TheCall->getNumArgs(); 640 } 641 642 // Determine whether the current function is variadic or not. 643 BlockScopeInfo *CurBlock = getCurBlock(); 644 bool isVariadic; 645 if (CurBlock) 646 isVariadic = CurBlock->TheDecl->isVariadic(); 647 else if (FunctionDecl *FD = getCurFunctionDecl()) 648 isVariadic = FD->isVariadic(); 649 else 650 isVariadic = getCurMethodDecl()->isVariadic(); 651 652 if (!isVariadic) { 653 Diag(Fn->getLocStart(), diag::err_va_start_used_in_non_variadic_function); 654 return true; 655 } 656 657 // Verify that the second argument to the builtin is the last argument of the 658 // current function or method. 659 bool SecondArgIsLastNamedArgument = false; 660 const Expr *Arg = TheCall->getArg(1)->IgnoreParenCasts(); 661 662 if (const DeclRefExpr *DR = dyn_cast<DeclRefExpr>(Arg)) { 663 if (const ParmVarDecl *PV = dyn_cast<ParmVarDecl>(DR->getDecl())) { 664 // FIXME: This isn't correct for methods (results in bogus warning). 665 // Get the last formal in the current function. 666 const ParmVarDecl *LastArg; 667 if (CurBlock) 668 LastArg = *(CurBlock->TheDecl->param_end()-1); 669 else if (FunctionDecl *FD = getCurFunctionDecl()) 670 LastArg = *(FD->param_end()-1); 671 else 672 LastArg = *(getCurMethodDecl()->param_end()-1); 673 SecondArgIsLastNamedArgument = PV == LastArg; 674 } 675 } 676 677 if (!SecondArgIsLastNamedArgument) 678 Diag(TheCall->getArg(1)->getLocStart(), 679 diag::warn_second_parameter_of_va_start_not_last_named_argument); 680 return false; 681} 682 683/// SemaBuiltinUnorderedCompare - Handle functions like __builtin_isgreater and 684/// friends. This is declared to take (...), so we have to check everything. 685bool Sema::SemaBuiltinUnorderedCompare(CallExpr *TheCall) { 686 if (TheCall->getNumArgs() < 2) 687 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 688 << 0 << 2 << TheCall->getNumArgs()/*function call*/; 689 if (TheCall->getNumArgs() > 2) 690 return Diag(TheCall->getArg(2)->getLocStart(), 691 diag::err_typecheck_call_too_many_args) 692 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 693 << SourceRange(TheCall->getArg(2)->getLocStart(), 694 (*(TheCall->arg_end()-1))->getLocEnd()); 695 696 Expr *OrigArg0 = TheCall->getArg(0); 697 Expr *OrigArg1 = TheCall->getArg(1); 698 699 // Do standard promotions between the two arguments, returning their common 700 // type. 701 QualType Res = UsualArithmeticConversions(OrigArg0, OrigArg1, false); 702 703 // Make sure any conversions are pushed back into the call; this is 704 // type safe since unordered compare builtins are declared as "_Bool 705 // foo(...)". 706 TheCall->setArg(0, OrigArg0); 707 TheCall->setArg(1, OrigArg1); 708 709 if (OrigArg0->isTypeDependent() || OrigArg1->isTypeDependent()) 710 return false; 711 712 // If the common type isn't a real floating type, then the arguments were 713 // invalid for this operation. 714 if (!Res->isRealFloatingType()) 715 return Diag(OrigArg0->getLocStart(), 716 diag::err_typecheck_call_invalid_ordered_compare) 717 << OrigArg0->getType() << OrigArg1->getType() 718 << SourceRange(OrigArg0->getLocStart(), OrigArg1->getLocEnd()); 719 720 return false; 721} 722 723/// SemaBuiltinSemaBuiltinFPClassification - Handle functions like 724/// __builtin_isnan and friends. This is declared to take (...), so we have 725/// to check everything. We expect the last argument to be a floating point 726/// value. 727bool Sema::SemaBuiltinFPClassification(CallExpr *TheCall, unsigned NumArgs) { 728 if (TheCall->getNumArgs() < NumArgs) 729 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 730 << 0 << NumArgs << TheCall->getNumArgs()/*function call*/; 731 if (TheCall->getNumArgs() > NumArgs) 732 return Diag(TheCall->getArg(NumArgs)->getLocStart(), 733 diag::err_typecheck_call_too_many_args) 734 << 0 /*function call*/ << NumArgs << TheCall->getNumArgs() 735 << SourceRange(TheCall->getArg(NumArgs)->getLocStart(), 736 (*(TheCall->arg_end()-1))->getLocEnd()); 737 738 Expr *OrigArg = TheCall->getArg(NumArgs-1); 739 740 if (OrigArg->isTypeDependent()) 741 return false; 742 743 // This operation requires a non-_Complex floating-point number. 744 if (!OrigArg->getType()->isRealFloatingType()) 745 return Diag(OrigArg->getLocStart(), 746 diag::err_typecheck_call_invalid_unary_fp) 747 << OrigArg->getType() << OrigArg->getSourceRange(); 748 749 // If this is an implicit conversion from float -> double, remove it. 750 if (ImplicitCastExpr *Cast = dyn_cast<ImplicitCastExpr>(OrigArg)) { 751 Expr *CastArg = Cast->getSubExpr(); 752 if (CastArg->getType()->isSpecificBuiltinType(BuiltinType::Float)) { 753 assert(Cast->getType()->isSpecificBuiltinType(BuiltinType::Double) && 754 "promotion from float to double is the only expected cast here"); 755 Cast->setSubExpr(0); 756 TheCall->setArg(NumArgs-1, CastArg); 757 OrigArg = CastArg; 758 } 759 } 760 761 return false; 762} 763 764/// SemaBuiltinShuffleVector - Handle __builtin_shufflevector. 765// This is declared to take (...), so we have to check everything. 766ExprResult Sema::SemaBuiltinShuffleVector(CallExpr *TheCall) { 767 if (TheCall->getNumArgs() < 2) 768 return ExprError(Diag(TheCall->getLocEnd(), 769 diag::err_typecheck_call_too_few_args_at_least) 770 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 771 << TheCall->getSourceRange()); 772 773 // Determine which of the following types of shufflevector we're checking: 774 // 1) unary, vector mask: (lhs, mask) 775 // 2) binary, vector mask: (lhs, rhs, mask) 776 // 3) binary, scalar mask: (lhs, rhs, index, ..., index) 777 QualType resType = TheCall->getArg(0)->getType(); 778 unsigned numElements = 0; 779 780 if (!TheCall->getArg(0)->isTypeDependent() && 781 !TheCall->getArg(1)->isTypeDependent()) { 782 QualType LHSType = TheCall->getArg(0)->getType(); 783 QualType RHSType = TheCall->getArg(1)->getType(); 784 785 if (!LHSType->isVectorType() || !RHSType->isVectorType()) { 786 Diag(TheCall->getLocStart(), diag::err_shufflevector_non_vector) 787 << SourceRange(TheCall->getArg(0)->getLocStart(), 788 TheCall->getArg(1)->getLocEnd()); 789 return ExprError(); 790 } 791 792 numElements = LHSType->getAs<VectorType>()->getNumElements(); 793 unsigned numResElements = TheCall->getNumArgs() - 2; 794 795 // Check to see if we have a call with 2 vector arguments, the unary shuffle 796 // with mask. If so, verify that RHS is an integer vector type with the 797 // same number of elts as lhs. 798 if (TheCall->getNumArgs() == 2) { 799 if (!RHSType->hasIntegerRepresentation() || 800 RHSType->getAs<VectorType>()->getNumElements() != numElements) 801 Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector) 802 << SourceRange(TheCall->getArg(1)->getLocStart(), 803 TheCall->getArg(1)->getLocEnd()); 804 numResElements = numElements; 805 } 806 else if (!Context.hasSameUnqualifiedType(LHSType, RHSType)) { 807 Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector) 808 << SourceRange(TheCall->getArg(0)->getLocStart(), 809 TheCall->getArg(1)->getLocEnd()); 810 return ExprError(); 811 } else if (numElements != numResElements) { 812 QualType eltType = LHSType->getAs<VectorType>()->getElementType(); 813 resType = Context.getVectorType(eltType, numResElements, 814 VectorType::NotAltiVec); 815 } 816 } 817 818 for (unsigned i = 2; i < TheCall->getNumArgs(); i++) { 819 if (TheCall->getArg(i)->isTypeDependent() || 820 TheCall->getArg(i)->isValueDependent()) 821 continue; 822 823 llvm::APSInt Result(32); 824 if (!TheCall->getArg(i)->isIntegerConstantExpr(Result, Context)) 825 return ExprError(Diag(TheCall->getLocStart(), 826 diag::err_shufflevector_nonconstant_argument) 827 << TheCall->getArg(i)->getSourceRange()); 828 829 if (Result.getActiveBits() > 64 || Result.getZExtValue() >= numElements*2) 830 return ExprError(Diag(TheCall->getLocStart(), 831 diag::err_shufflevector_argument_too_large) 832 << TheCall->getArg(i)->getSourceRange()); 833 } 834 835 llvm::SmallVector<Expr*, 32> exprs; 836 837 for (unsigned i = 0, e = TheCall->getNumArgs(); i != e; i++) { 838 exprs.push_back(TheCall->getArg(i)); 839 TheCall->setArg(i, 0); 840 } 841 842 return Owned(new (Context) ShuffleVectorExpr(Context, exprs.begin(), 843 exprs.size(), resType, 844 TheCall->getCallee()->getLocStart(), 845 TheCall->getRParenLoc())); 846} 847 848/// SemaBuiltinPrefetch - Handle __builtin_prefetch. 849// This is declared to take (const void*, ...) and can take two 850// optional constant int args. 851bool Sema::SemaBuiltinPrefetch(CallExpr *TheCall) { 852 unsigned NumArgs = TheCall->getNumArgs(); 853 854 if (NumArgs > 3) 855 return Diag(TheCall->getLocEnd(), 856 diag::err_typecheck_call_too_many_args_at_most) 857 << 0 /*function call*/ << 3 << NumArgs 858 << TheCall->getSourceRange(); 859 860 // Argument 0 is checked for us and the remaining arguments must be 861 // constant integers. 862 for (unsigned i = 1; i != NumArgs; ++i) { 863 Expr *Arg = TheCall->getArg(i); 864 865 llvm::APSInt Result; 866 if (SemaBuiltinConstantArg(TheCall, i, Result)) 867 return true; 868 869 // FIXME: gcc issues a warning and rewrites these to 0. These 870 // seems especially odd for the third argument since the default 871 // is 3. 872 if (i == 1) { 873 if (Result.getLimitedValue() > 1) 874 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 875 << "0" << "1" << Arg->getSourceRange(); 876 } else { 877 if (Result.getLimitedValue() > 3) 878 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 879 << "0" << "3" << Arg->getSourceRange(); 880 } 881 } 882 883 return false; 884} 885 886/// SemaBuiltinConstantArg - Handle a check if argument ArgNum of CallExpr 887/// TheCall is a constant expression. 888bool Sema::SemaBuiltinConstantArg(CallExpr *TheCall, int ArgNum, 889 llvm::APSInt &Result) { 890 Expr *Arg = TheCall->getArg(ArgNum); 891 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 892 FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl()); 893 894 if (Arg->isTypeDependent() || Arg->isValueDependent()) return false; 895 896 if (!Arg->isIntegerConstantExpr(Result, Context)) 897 return Diag(TheCall->getLocStart(), diag::err_constant_integer_arg_type) 898 << FDecl->getDeclName() << Arg->getSourceRange(); 899 900 return false; 901} 902 903/// SemaBuiltinObjectSize - Handle __builtin_object_size(void *ptr, 904/// int type). This simply type checks that type is one of the defined 905/// constants (0-3). 906// For compatability check 0-3, llvm only handles 0 and 2. 907bool Sema::SemaBuiltinObjectSize(CallExpr *TheCall) { 908 llvm::APSInt Result; 909 910 // Check constant-ness first. 911 if (SemaBuiltinConstantArg(TheCall, 1, Result)) 912 return true; 913 914 Expr *Arg = TheCall->getArg(1); 915 if (Result.getSExtValue() < 0 || Result.getSExtValue() > 3) { 916 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 917 << "0" << "3" << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 918 } 919 920 return false; 921} 922 923/// SemaBuiltinLongjmp - Handle __builtin_longjmp(void *env[5], int val). 924/// This checks that val is a constant 1. 925bool Sema::SemaBuiltinLongjmp(CallExpr *TheCall) { 926 Expr *Arg = TheCall->getArg(1); 927 llvm::APSInt Result; 928 929 // TODO: This is less than ideal. Overload this to take a value. 930 if (SemaBuiltinConstantArg(TheCall, 1, Result)) 931 return true; 932 933 if (Result != 1) 934 return Diag(TheCall->getLocStart(), diag::err_builtin_longjmp_invalid_val) 935 << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 936 937 return false; 938} 939 940// Handle i > 1 ? "x" : "y", recursivelly 941bool Sema::SemaCheckStringLiteral(const Expr *E, const CallExpr *TheCall, 942 bool HasVAListArg, 943 unsigned format_idx, unsigned firstDataArg, 944 bool isPrintf) { 945 946 if (E->isTypeDependent() || E->isValueDependent()) 947 return false; 948 949 switch (E->getStmtClass()) { 950 case Stmt::ConditionalOperatorClass: { 951 const ConditionalOperator *C = cast<ConditionalOperator>(E); 952 return SemaCheckStringLiteral(C->getTrueExpr(), TheCall, HasVAListArg, 953 format_idx, firstDataArg, isPrintf) 954 && SemaCheckStringLiteral(C->getRHS(), TheCall, HasVAListArg, 955 format_idx, firstDataArg, isPrintf); 956 } 957 958 case Stmt::ImplicitCastExprClass: { 959 const ImplicitCastExpr *Expr = cast<ImplicitCastExpr>(E); 960 return SemaCheckStringLiteral(Expr->getSubExpr(), TheCall, HasVAListArg, 961 format_idx, firstDataArg, isPrintf); 962 } 963 964 case Stmt::ParenExprClass: { 965 const ParenExpr *Expr = cast<ParenExpr>(E); 966 return SemaCheckStringLiteral(Expr->getSubExpr(), TheCall, HasVAListArg, 967 format_idx, firstDataArg, isPrintf); 968 } 969 970 case Stmt::DeclRefExprClass: { 971 const DeclRefExpr *DR = cast<DeclRefExpr>(E); 972 973 // As an exception, do not flag errors for variables binding to 974 // const string literals. 975 if (const VarDecl *VD = dyn_cast<VarDecl>(DR->getDecl())) { 976 bool isConstant = false; 977 QualType T = DR->getType(); 978 979 if (const ArrayType *AT = Context.getAsArrayType(T)) { 980 isConstant = AT->getElementType().isConstant(Context); 981 } else if (const PointerType *PT = T->getAs<PointerType>()) { 982 isConstant = T.isConstant(Context) && 983 PT->getPointeeType().isConstant(Context); 984 } 985 986 if (isConstant) { 987 if (const Expr *Init = VD->getAnyInitializer()) 988 return SemaCheckStringLiteral(Init, TheCall, 989 HasVAListArg, format_idx, firstDataArg, 990 isPrintf); 991 } 992 993 // For vprintf* functions (i.e., HasVAListArg==true), we add a 994 // special check to see if the format string is a function parameter 995 // of the function calling the printf function. If the function 996 // has an attribute indicating it is a printf-like function, then we 997 // should suppress warnings concerning non-literals being used in a call 998 // to a vprintf function. For example: 999 // 1000 // void 1001 // logmessage(char const *fmt __attribute__ (format (printf, 1, 2)), ...){ 1002 // va_list ap; 1003 // va_start(ap, fmt); 1004 // vprintf(fmt, ap); // Do NOT emit a warning about "fmt". 1005 // ... 1006 // 1007 // 1008 // FIXME: We don't have full attribute support yet, so just check to see 1009 // if the argument is a DeclRefExpr that references a parameter. We'll 1010 // add proper support for checking the attribute later. 1011 if (HasVAListArg) 1012 if (isa<ParmVarDecl>(VD)) 1013 return true; 1014 } 1015 1016 return false; 1017 } 1018 1019 case Stmt::CallExprClass: { 1020 const CallExpr *CE = cast<CallExpr>(E); 1021 if (const ImplicitCastExpr *ICE 1022 = dyn_cast<ImplicitCastExpr>(CE->getCallee())) { 1023 if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(ICE->getSubExpr())) { 1024 if (const FunctionDecl *FD = dyn_cast<FunctionDecl>(DRE->getDecl())) { 1025 if (const FormatArgAttr *FA = FD->getAttr<FormatArgAttr>()) { 1026 unsigned ArgIndex = FA->getFormatIdx(); 1027 const Expr *Arg = CE->getArg(ArgIndex - 1); 1028 1029 return SemaCheckStringLiteral(Arg, TheCall, HasVAListArg, 1030 format_idx, firstDataArg, isPrintf); 1031 } 1032 } 1033 } 1034 } 1035 1036 return false; 1037 } 1038 case Stmt::ObjCStringLiteralClass: 1039 case Stmt::StringLiteralClass: { 1040 const StringLiteral *StrE = NULL; 1041 1042 if (const ObjCStringLiteral *ObjCFExpr = dyn_cast<ObjCStringLiteral>(E)) 1043 StrE = ObjCFExpr->getString(); 1044 else 1045 StrE = cast<StringLiteral>(E); 1046 1047 if (StrE) { 1048 CheckFormatString(StrE, E, TheCall, HasVAListArg, format_idx, 1049 firstDataArg, isPrintf); 1050 return true; 1051 } 1052 1053 return false; 1054 } 1055 1056 default: 1057 return false; 1058 } 1059} 1060 1061void 1062Sema::CheckNonNullArguments(const NonNullAttr *NonNull, 1063 const CallExpr *TheCall) { 1064 for (NonNullAttr::args_iterator i = NonNull->args_begin(), 1065 e = NonNull->args_end(); 1066 i != e; ++i) { 1067 const Expr *ArgExpr = TheCall->getArg(*i); 1068 if (ArgExpr->isNullPointerConstant(Context, 1069 Expr::NPC_ValueDependentIsNotNull)) 1070 Diag(TheCall->getCallee()->getLocStart(), diag::warn_null_arg) 1071 << ArgExpr->getSourceRange(); 1072 } 1073} 1074 1075/// CheckPrintfScanfArguments - Check calls to printf and scanf (and similar 1076/// functions) for correct use of format strings. 1077void 1078Sema::CheckPrintfScanfArguments(const CallExpr *TheCall, bool HasVAListArg, 1079 unsigned format_idx, unsigned firstDataArg, 1080 bool isPrintf) { 1081 1082 const Expr *Fn = TheCall->getCallee(); 1083 1084 // The way the format attribute works in GCC, the implicit this argument 1085 // of member functions is counted. However, it doesn't appear in our own 1086 // lists, so decrement format_idx in that case. 1087 if (isa<CXXMemberCallExpr>(TheCall)) { 1088 // Catch a format attribute mistakenly referring to the object argument. 1089 if (format_idx == 0) 1090 return; 1091 --format_idx; 1092 if(firstDataArg != 0) 1093 --firstDataArg; 1094 } 1095 1096 // CHECK: printf/scanf-like function is called with no format string. 1097 if (format_idx >= TheCall->getNumArgs()) { 1098 Diag(TheCall->getRParenLoc(), diag::warn_missing_format_string) 1099 << Fn->getSourceRange(); 1100 return; 1101 } 1102 1103 const Expr *OrigFormatExpr = TheCall->getArg(format_idx)->IgnoreParenCasts(); 1104 1105 // CHECK: format string is not a string literal. 1106 // 1107 // Dynamically generated format strings are difficult to 1108 // automatically vet at compile time. Requiring that format strings 1109 // are string literals: (1) permits the checking of format strings by 1110 // the compiler and thereby (2) can practically remove the source of 1111 // many format string exploits. 1112 1113 // Format string can be either ObjC string (e.g. @"%d") or 1114 // C string (e.g. "%d") 1115 // ObjC string uses the same format specifiers as C string, so we can use 1116 // the same format string checking logic for both ObjC and C strings. 1117 if (SemaCheckStringLiteral(OrigFormatExpr, TheCall, HasVAListArg, format_idx, 1118 firstDataArg, isPrintf)) 1119 return; // Literal format string found, check done! 1120 1121 // If there are no arguments specified, warn with -Wformat-security, otherwise 1122 // warn only with -Wformat-nonliteral. 1123 if (TheCall->getNumArgs() == format_idx+1) 1124 Diag(TheCall->getArg(format_idx)->getLocStart(), 1125 diag::warn_format_nonliteral_noargs) 1126 << OrigFormatExpr->getSourceRange(); 1127 else 1128 Diag(TheCall->getArg(format_idx)->getLocStart(), 1129 diag::warn_format_nonliteral) 1130 << OrigFormatExpr->getSourceRange(); 1131} 1132 1133namespace { 1134class CheckFormatHandler : public analyze_format_string::FormatStringHandler { 1135protected: 1136 Sema &S; 1137 const StringLiteral *FExpr; 1138 const Expr *OrigFormatExpr; 1139 const unsigned FirstDataArg; 1140 const unsigned NumDataArgs; 1141 const bool IsObjCLiteral; 1142 const char *Beg; // Start of format string. 1143 const bool HasVAListArg; 1144 const CallExpr *TheCall; 1145 unsigned FormatIdx; 1146 llvm::BitVector CoveredArgs; 1147 bool usesPositionalArgs; 1148 bool atFirstArg; 1149public: 1150 CheckFormatHandler(Sema &s, const StringLiteral *fexpr, 1151 const Expr *origFormatExpr, unsigned firstDataArg, 1152 unsigned numDataArgs, bool isObjCLiteral, 1153 const char *beg, bool hasVAListArg, 1154 const CallExpr *theCall, unsigned formatIdx) 1155 : S(s), FExpr(fexpr), OrigFormatExpr(origFormatExpr), 1156 FirstDataArg(firstDataArg), 1157 NumDataArgs(numDataArgs), 1158 IsObjCLiteral(isObjCLiteral), Beg(beg), 1159 HasVAListArg(hasVAListArg), 1160 TheCall(theCall), FormatIdx(formatIdx), 1161 usesPositionalArgs(false), atFirstArg(true) { 1162 CoveredArgs.resize(numDataArgs); 1163 CoveredArgs.reset(); 1164 } 1165 1166 void DoneProcessing(); 1167 1168 void HandleIncompleteSpecifier(const char *startSpecifier, 1169 unsigned specifierLen); 1170 1171 virtual void HandleInvalidPosition(const char *startSpecifier, 1172 unsigned specifierLen, 1173 analyze_format_string::PositionContext p); 1174 1175 virtual void HandleZeroPosition(const char *startPos, unsigned posLen); 1176 1177 void HandleNullChar(const char *nullCharacter); 1178 1179protected: 1180 bool HandleInvalidConversionSpecifier(unsigned argIndex, SourceLocation Loc, 1181 const char *startSpec, 1182 unsigned specifierLen, 1183 const char *csStart, unsigned csLen); 1184 1185 SourceRange getFormatStringRange(); 1186 CharSourceRange getSpecifierRange(const char *startSpecifier, 1187 unsigned specifierLen); 1188 SourceLocation getLocationOfByte(const char *x); 1189 1190 const Expr *getDataArg(unsigned i) const; 1191 1192 bool CheckNumArgs(const analyze_format_string::FormatSpecifier &FS, 1193 const analyze_format_string::ConversionSpecifier &CS, 1194 const char *startSpecifier, unsigned specifierLen, 1195 unsigned argIndex); 1196}; 1197} 1198 1199SourceRange CheckFormatHandler::getFormatStringRange() { 1200 return OrigFormatExpr->getSourceRange(); 1201} 1202 1203CharSourceRange CheckFormatHandler:: 1204getSpecifierRange(const char *startSpecifier, unsigned specifierLen) { 1205 SourceLocation Start = getLocationOfByte(startSpecifier); 1206 SourceLocation End = getLocationOfByte(startSpecifier + specifierLen - 1); 1207 1208 // Advance the end SourceLocation by one due to half-open ranges. 1209 End = End.getFileLocWithOffset(1); 1210 1211 return CharSourceRange::getCharRange(Start, End); 1212} 1213 1214SourceLocation CheckFormatHandler::getLocationOfByte(const char *x) { 1215 return S.getLocationOfStringLiteralByte(FExpr, x - Beg); 1216} 1217 1218void CheckFormatHandler::HandleIncompleteSpecifier(const char *startSpecifier, 1219 unsigned specifierLen){ 1220 SourceLocation Loc = getLocationOfByte(startSpecifier); 1221 S.Diag(Loc, diag::warn_printf_incomplete_specifier) 1222 << getSpecifierRange(startSpecifier, specifierLen); 1223} 1224 1225void 1226CheckFormatHandler::HandleInvalidPosition(const char *startPos, unsigned posLen, 1227 analyze_format_string::PositionContext p) { 1228 SourceLocation Loc = getLocationOfByte(startPos); 1229 S.Diag(Loc, diag::warn_format_invalid_positional_specifier) 1230 << (unsigned) p << getSpecifierRange(startPos, posLen); 1231} 1232 1233void CheckFormatHandler::HandleZeroPosition(const char *startPos, 1234 unsigned posLen) { 1235 SourceLocation Loc = getLocationOfByte(startPos); 1236 S.Diag(Loc, diag::warn_format_zero_positional_specifier) 1237 << getSpecifierRange(startPos, posLen); 1238} 1239 1240void CheckFormatHandler::HandleNullChar(const char *nullCharacter) { 1241 // The presence of a null character is likely an error. 1242 S.Diag(getLocationOfByte(nullCharacter), 1243 diag::warn_printf_format_string_contains_null_char) 1244 << getFormatStringRange(); 1245} 1246 1247const Expr *CheckFormatHandler::getDataArg(unsigned i) const { 1248 return TheCall->getArg(FirstDataArg + i); 1249} 1250 1251void CheckFormatHandler::DoneProcessing() { 1252 // Does the number of data arguments exceed the number of 1253 // format conversions in the format string? 1254 if (!HasVAListArg) { 1255 // Find any arguments that weren't covered. 1256 CoveredArgs.flip(); 1257 signed notCoveredArg = CoveredArgs.find_first(); 1258 if (notCoveredArg >= 0) { 1259 assert((unsigned)notCoveredArg < NumDataArgs); 1260 S.Diag(getDataArg((unsigned) notCoveredArg)->getLocStart(), 1261 diag::warn_printf_data_arg_not_used) 1262 << getFormatStringRange(); 1263 } 1264 } 1265} 1266 1267bool 1268CheckFormatHandler::HandleInvalidConversionSpecifier(unsigned argIndex, 1269 SourceLocation Loc, 1270 const char *startSpec, 1271 unsigned specifierLen, 1272 const char *csStart, 1273 unsigned csLen) { 1274 1275 bool keepGoing = true; 1276 if (argIndex < NumDataArgs) { 1277 // Consider the argument coverered, even though the specifier doesn't 1278 // make sense. 1279 CoveredArgs.set(argIndex); 1280 } 1281 else { 1282 // If argIndex exceeds the number of data arguments we 1283 // don't issue a warning because that is just a cascade of warnings (and 1284 // they may have intended '%%' anyway). We don't want to continue processing 1285 // the format string after this point, however, as we will like just get 1286 // gibberish when trying to match arguments. 1287 keepGoing = false; 1288 } 1289 1290 S.Diag(Loc, diag::warn_format_invalid_conversion) 1291 << llvm::StringRef(csStart, csLen) 1292 << getSpecifierRange(startSpec, specifierLen); 1293 1294 return keepGoing; 1295} 1296 1297bool 1298CheckFormatHandler::CheckNumArgs( 1299 const analyze_format_string::FormatSpecifier &FS, 1300 const analyze_format_string::ConversionSpecifier &CS, 1301 const char *startSpecifier, unsigned specifierLen, unsigned argIndex) { 1302 1303 if (argIndex >= NumDataArgs) { 1304 if (FS.usesPositionalArg()) { 1305 S.Diag(getLocationOfByte(CS.getStart()), 1306 diag::warn_printf_positional_arg_exceeds_data_args) 1307 << (argIndex+1) << NumDataArgs 1308 << getSpecifierRange(startSpecifier, specifierLen); 1309 } 1310 else { 1311 S.Diag(getLocationOfByte(CS.getStart()), 1312 diag::warn_printf_insufficient_data_args) 1313 << getSpecifierRange(startSpecifier, specifierLen); 1314 } 1315 1316 return false; 1317 } 1318 return true; 1319} 1320 1321//===--- CHECK: Printf format string checking ------------------------------===// 1322 1323namespace { 1324class CheckPrintfHandler : public CheckFormatHandler { 1325public: 1326 CheckPrintfHandler(Sema &s, const StringLiteral *fexpr, 1327 const Expr *origFormatExpr, unsigned firstDataArg, 1328 unsigned numDataArgs, bool isObjCLiteral, 1329 const char *beg, bool hasVAListArg, 1330 const CallExpr *theCall, unsigned formatIdx) 1331 : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg, 1332 numDataArgs, isObjCLiteral, beg, hasVAListArg, 1333 theCall, formatIdx) {} 1334 1335 1336 bool HandleInvalidPrintfConversionSpecifier( 1337 const analyze_printf::PrintfSpecifier &FS, 1338 const char *startSpecifier, 1339 unsigned specifierLen); 1340 1341 bool HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier &FS, 1342 const char *startSpecifier, 1343 unsigned specifierLen); 1344 1345 bool HandleAmount(const analyze_format_string::OptionalAmount &Amt, unsigned k, 1346 const char *startSpecifier, unsigned specifierLen); 1347 void HandleInvalidAmount(const analyze_printf::PrintfSpecifier &FS, 1348 const analyze_printf::OptionalAmount &Amt, 1349 unsigned type, 1350 const char *startSpecifier, unsigned specifierLen); 1351 void HandleFlag(const analyze_printf::PrintfSpecifier &FS, 1352 const analyze_printf::OptionalFlag &flag, 1353 const char *startSpecifier, unsigned specifierLen); 1354 void HandleIgnoredFlag(const analyze_printf::PrintfSpecifier &FS, 1355 const analyze_printf::OptionalFlag &ignoredFlag, 1356 const analyze_printf::OptionalFlag &flag, 1357 const char *startSpecifier, unsigned specifierLen); 1358}; 1359} 1360 1361bool CheckPrintfHandler::HandleInvalidPrintfConversionSpecifier( 1362 const analyze_printf::PrintfSpecifier &FS, 1363 const char *startSpecifier, 1364 unsigned specifierLen) { 1365 const analyze_printf::PrintfConversionSpecifier &CS = 1366 FS.getConversionSpecifier(); 1367 1368 return HandleInvalidConversionSpecifier(FS.getArgIndex(), 1369 getLocationOfByte(CS.getStart()), 1370 startSpecifier, specifierLen, 1371 CS.getStart(), CS.getLength()); 1372} 1373 1374bool CheckPrintfHandler::HandleAmount( 1375 const analyze_format_string::OptionalAmount &Amt, 1376 unsigned k, const char *startSpecifier, 1377 unsigned specifierLen) { 1378 1379 if (Amt.hasDataArgument()) { 1380 if (!HasVAListArg) { 1381 unsigned argIndex = Amt.getArgIndex(); 1382 if (argIndex >= NumDataArgs) { 1383 S.Diag(getLocationOfByte(Amt.getStart()), 1384 diag::warn_printf_asterisk_missing_arg) 1385 << k << getSpecifierRange(startSpecifier, specifierLen); 1386 // Don't do any more checking. We will just emit 1387 // spurious errors. 1388 return false; 1389 } 1390 1391 // Type check the data argument. It should be an 'int'. 1392 // Although not in conformance with C99, we also allow the argument to be 1393 // an 'unsigned int' as that is a reasonably safe case. GCC also 1394 // doesn't emit a warning for that case. 1395 CoveredArgs.set(argIndex); 1396 const Expr *Arg = getDataArg(argIndex); 1397 QualType T = Arg->getType(); 1398 1399 const analyze_printf::ArgTypeResult &ATR = Amt.getArgType(S.Context); 1400 assert(ATR.isValid()); 1401 1402 if (!ATR.matchesType(S.Context, T)) { 1403 S.Diag(getLocationOfByte(Amt.getStart()), 1404 diag::warn_printf_asterisk_wrong_type) 1405 << k 1406 << ATR.getRepresentativeType(S.Context) << T 1407 << getSpecifierRange(startSpecifier, specifierLen) 1408 << Arg->getSourceRange(); 1409 // Don't do any more checking. We will just emit 1410 // spurious errors. 1411 return false; 1412 } 1413 } 1414 } 1415 return true; 1416} 1417 1418void CheckPrintfHandler::HandleInvalidAmount( 1419 const analyze_printf::PrintfSpecifier &FS, 1420 const analyze_printf::OptionalAmount &Amt, 1421 unsigned type, 1422 const char *startSpecifier, 1423 unsigned specifierLen) { 1424 const analyze_printf::PrintfConversionSpecifier &CS = 1425 FS.getConversionSpecifier(); 1426 switch (Amt.getHowSpecified()) { 1427 case analyze_printf::OptionalAmount::Constant: 1428 S.Diag(getLocationOfByte(Amt.getStart()), 1429 diag::warn_printf_nonsensical_optional_amount) 1430 << type 1431 << CS.toString() 1432 << getSpecifierRange(startSpecifier, specifierLen) 1433 << FixItHint::CreateRemoval(getSpecifierRange(Amt.getStart(), 1434 Amt.getConstantLength())); 1435 break; 1436 1437 default: 1438 S.Diag(getLocationOfByte(Amt.getStart()), 1439 diag::warn_printf_nonsensical_optional_amount) 1440 << type 1441 << CS.toString() 1442 << getSpecifierRange(startSpecifier, specifierLen); 1443 break; 1444 } 1445} 1446 1447void CheckPrintfHandler::HandleFlag(const analyze_printf::PrintfSpecifier &FS, 1448 const analyze_printf::OptionalFlag &flag, 1449 const char *startSpecifier, 1450 unsigned specifierLen) { 1451 // Warn about pointless flag with a fixit removal. 1452 const analyze_printf::PrintfConversionSpecifier &CS = 1453 FS.getConversionSpecifier(); 1454 S.Diag(getLocationOfByte(flag.getPosition()), 1455 diag::warn_printf_nonsensical_flag) 1456 << flag.toString() << CS.toString() 1457 << getSpecifierRange(startSpecifier, specifierLen) 1458 << FixItHint::CreateRemoval(getSpecifierRange(flag.getPosition(), 1)); 1459} 1460 1461void CheckPrintfHandler::HandleIgnoredFlag( 1462 const analyze_printf::PrintfSpecifier &FS, 1463 const analyze_printf::OptionalFlag &ignoredFlag, 1464 const analyze_printf::OptionalFlag &flag, 1465 const char *startSpecifier, 1466 unsigned specifierLen) { 1467 // Warn about ignored flag with a fixit removal. 1468 S.Diag(getLocationOfByte(ignoredFlag.getPosition()), 1469 diag::warn_printf_ignored_flag) 1470 << ignoredFlag.toString() << flag.toString() 1471 << getSpecifierRange(startSpecifier, specifierLen) 1472 << FixItHint::CreateRemoval(getSpecifierRange( 1473 ignoredFlag.getPosition(), 1)); 1474} 1475 1476bool 1477CheckPrintfHandler::HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier 1478 &FS, 1479 const char *startSpecifier, 1480 unsigned specifierLen) { 1481 1482 using namespace analyze_format_string; 1483 using namespace analyze_printf; 1484 const PrintfConversionSpecifier &CS = FS.getConversionSpecifier(); 1485 1486 if (FS.consumesDataArgument()) { 1487 if (atFirstArg) { 1488 atFirstArg = false; 1489 usesPositionalArgs = FS.usesPositionalArg(); 1490 } 1491 else if (usesPositionalArgs != FS.usesPositionalArg()) { 1492 // Cannot mix-and-match positional and non-positional arguments. 1493 S.Diag(getLocationOfByte(CS.getStart()), 1494 diag::warn_format_mix_positional_nonpositional_args) 1495 << getSpecifierRange(startSpecifier, specifierLen); 1496 return false; 1497 } 1498 } 1499 1500 // First check if the field width, precision, and conversion specifier 1501 // have matching data arguments. 1502 if (!HandleAmount(FS.getFieldWidth(), /* field width */ 0, 1503 startSpecifier, specifierLen)) { 1504 return false; 1505 } 1506 1507 if (!HandleAmount(FS.getPrecision(), /* precision */ 1, 1508 startSpecifier, specifierLen)) { 1509 return false; 1510 } 1511 1512 if (!CS.consumesDataArgument()) { 1513 // FIXME: Technically specifying a precision or field width here 1514 // makes no sense. Worth issuing a warning at some point. 1515 return true; 1516 } 1517 1518 // Consume the argument. 1519 unsigned argIndex = FS.getArgIndex(); 1520 if (argIndex < NumDataArgs) { 1521 // The check to see if the argIndex is valid will come later. 1522 // We set the bit here because we may exit early from this 1523 // function if we encounter some other error. 1524 CoveredArgs.set(argIndex); 1525 } 1526 1527 // Check for using an Objective-C specific conversion specifier 1528 // in a non-ObjC literal. 1529 if (!IsObjCLiteral && CS.isObjCArg()) { 1530 return HandleInvalidPrintfConversionSpecifier(FS, startSpecifier, 1531 specifierLen); 1532 } 1533 1534 // Check for invalid use of field width 1535 if (!FS.hasValidFieldWidth()) { 1536 HandleInvalidAmount(FS, FS.getFieldWidth(), /* field width */ 0, 1537 startSpecifier, specifierLen); 1538 } 1539 1540 // Check for invalid use of precision 1541 if (!FS.hasValidPrecision()) { 1542 HandleInvalidAmount(FS, FS.getPrecision(), /* precision */ 1, 1543 startSpecifier, specifierLen); 1544 } 1545 1546 // Check each flag does not conflict with any other component. 1547 if (!FS.hasValidLeadingZeros()) 1548 HandleFlag(FS, FS.hasLeadingZeros(), startSpecifier, specifierLen); 1549 if (!FS.hasValidPlusPrefix()) 1550 HandleFlag(FS, FS.hasPlusPrefix(), startSpecifier, specifierLen); 1551 if (!FS.hasValidSpacePrefix()) 1552 HandleFlag(FS, FS.hasSpacePrefix(), startSpecifier, specifierLen); 1553 if (!FS.hasValidAlternativeForm()) 1554 HandleFlag(FS, FS.hasAlternativeForm(), startSpecifier, specifierLen); 1555 if (!FS.hasValidLeftJustified()) 1556 HandleFlag(FS, FS.isLeftJustified(), startSpecifier, specifierLen); 1557 1558 // Check that flags are not ignored by another flag 1559 if (FS.hasSpacePrefix() && FS.hasPlusPrefix()) // ' ' ignored by '+' 1560 HandleIgnoredFlag(FS, FS.hasSpacePrefix(), FS.hasPlusPrefix(), 1561 startSpecifier, specifierLen); 1562 if (FS.hasLeadingZeros() && FS.isLeftJustified()) // '0' ignored by '-' 1563 HandleIgnoredFlag(FS, FS.hasLeadingZeros(), FS.isLeftJustified(), 1564 startSpecifier, specifierLen); 1565 1566 // Check the length modifier is valid with the given conversion specifier. 1567 const LengthModifier &LM = FS.getLengthModifier(); 1568 if (!FS.hasValidLengthModifier()) 1569 S.Diag(getLocationOfByte(LM.getStart()), 1570 diag::warn_format_nonsensical_length) 1571 << LM.toString() << CS.toString() 1572 << getSpecifierRange(startSpecifier, specifierLen) 1573 << FixItHint::CreateRemoval(getSpecifierRange(LM.getStart(), 1574 LM.getLength())); 1575 1576 // Are we using '%n'? 1577 if (CS.getKind() == ConversionSpecifier::nArg) { 1578 // Issue a warning about this being a possible security issue. 1579 S.Diag(getLocationOfByte(CS.getStart()), diag::warn_printf_write_back) 1580 << getSpecifierRange(startSpecifier, specifierLen); 1581 // Continue checking the other format specifiers. 1582 return true; 1583 } 1584 1585 // The remaining checks depend on the data arguments. 1586 if (HasVAListArg) 1587 return true; 1588 1589 if (!CheckNumArgs(FS, CS, startSpecifier, specifierLen, argIndex)) 1590 return false; 1591 1592 // Now type check the data expression that matches the 1593 // format specifier. 1594 const Expr *Ex = getDataArg(argIndex); 1595 const analyze_printf::ArgTypeResult &ATR = FS.getArgType(S.Context); 1596 if (ATR.isValid() && !ATR.matchesType(S.Context, Ex->getType())) { 1597 // Check if we didn't match because of an implicit cast from a 'char' 1598 // or 'short' to an 'int'. This is done because printf is a varargs 1599 // function. 1600 if (const ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(Ex)) 1601 if (ICE->getType() == S.Context.IntTy) 1602 if (ATR.matchesType(S.Context, ICE->getSubExpr()->getType())) 1603 return true; 1604 1605 // We may be able to offer a FixItHint if it is a supported type. 1606 PrintfSpecifier fixedFS = FS; 1607 bool success = fixedFS.fixType(Ex->getType()); 1608 1609 if (success) { 1610 // Get the fix string from the fixed format specifier 1611 llvm::SmallString<128> buf; 1612 llvm::raw_svector_ostream os(buf); 1613 fixedFS.toString(os); 1614 1615 // FIXME: getRepresentativeType() perhaps should return a string 1616 // instead of a QualType to better handle when the representative 1617 // type is 'wint_t' (which is defined in the system headers). 1618 S.Diag(getLocationOfByte(CS.getStart()), 1619 diag::warn_printf_conversion_argument_type_mismatch) 1620 << ATR.getRepresentativeType(S.Context) << Ex->getType() 1621 << getSpecifierRange(startSpecifier, specifierLen) 1622 << Ex->getSourceRange() 1623 << FixItHint::CreateReplacement( 1624 getSpecifierRange(startSpecifier, specifierLen), 1625 os.str()); 1626 } 1627 else { 1628 S.Diag(getLocationOfByte(CS.getStart()), 1629 diag::warn_printf_conversion_argument_type_mismatch) 1630 << ATR.getRepresentativeType(S.Context) << Ex->getType() 1631 << getSpecifierRange(startSpecifier, specifierLen) 1632 << Ex->getSourceRange(); 1633 } 1634 } 1635 1636 return true; 1637} 1638 1639//===--- CHECK: Scanf format string checking ------------------------------===// 1640 1641namespace { 1642class CheckScanfHandler : public CheckFormatHandler { 1643public: 1644 CheckScanfHandler(Sema &s, const StringLiteral *fexpr, 1645 const Expr *origFormatExpr, unsigned firstDataArg, 1646 unsigned numDataArgs, bool isObjCLiteral, 1647 const char *beg, bool hasVAListArg, 1648 const CallExpr *theCall, unsigned formatIdx) 1649 : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg, 1650 numDataArgs, isObjCLiteral, beg, hasVAListArg, 1651 theCall, formatIdx) {} 1652 1653 bool HandleScanfSpecifier(const analyze_scanf::ScanfSpecifier &FS, 1654 const char *startSpecifier, 1655 unsigned specifierLen); 1656 1657 bool HandleInvalidScanfConversionSpecifier( 1658 const analyze_scanf::ScanfSpecifier &FS, 1659 const char *startSpecifier, 1660 unsigned specifierLen); 1661 1662 void HandleIncompleteScanList(const char *start, const char *end); 1663}; 1664} 1665 1666void CheckScanfHandler::HandleIncompleteScanList(const char *start, 1667 const char *end) { 1668 S.Diag(getLocationOfByte(end), diag::warn_scanf_scanlist_incomplete) 1669 << getSpecifierRange(start, end - start); 1670} 1671 1672bool CheckScanfHandler::HandleInvalidScanfConversionSpecifier( 1673 const analyze_scanf::ScanfSpecifier &FS, 1674 const char *startSpecifier, 1675 unsigned specifierLen) { 1676 1677 const analyze_scanf::ScanfConversionSpecifier &CS = 1678 FS.getConversionSpecifier(); 1679 1680 return HandleInvalidConversionSpecifier(FS.getArgIndex(), 1681 getLocationOfByte(CS.getStart()), 1682 startSpecifier, specifierLen, 1683 CS.getStart(), CS.getLength()); 1684} 1685 1686bool CheckScanfHandler::HandleScanfSpecifier( 1687 const analyze_scanf::ScanfSpecifier &FS, 1688 const char *startSpecifier, 1689 unsigned specifierLen) { 1690 1691 using namespace analyze_scanf; 1692 using namespace analyze_format_string; 1693 1694 const ScanfConversionSpecifier &CS = FS.getConversionSpecifier(); 1695 1696 // Handle case where '%' and '*' don't consume an argument. These shouldn't 1697 // be used to decide if we are using positional arguments consistently. 1698 if (FS.consumesDataArgument()) { 1699 if (atFirstArg) { 1700 atFirstArg = false; 1701 usesPositionalArgs = FS.usesPositionalArg(); 1702 } 1703 else if (usesPositionalArgs != FS.usesPositionalArg()) { 1704 // Cannot mix-and-match positional and non-positional arguments. 1705 S.Diag(getLocationOfByte(CS.getStart()), 1706 diag::warn_format_mix_positional_nonpositional_args) 1707 << getSpecifierRange(startSpecifier, specifierLen); 1708 return false; 1709 } 1710 } 1711 1712 // Check if the field with is non-zero. 1713 const OptionalAmount &Amt = FS.getFieldWidth(); 1714 if (Amt.getHowSpecified() == OptionalAmount::Constant) { 1715 if (Amt.getConstantAmount() == 0) { 1716 const CharSourceRange &R = getSpecifierRange(Amt.getStart(), 1717 Amt.getConstantLength()); 1718 S.Diag(getLocationOfByte(Amt.getStart()), 1719 diag::warn_scanf_nonzero_width) 1720 << R << FixItHint::CreateRemoval(R); 1721 } 1722 } 1723 1724 if (!FS.consumesDataArgument()) { 1725 // FIXME: Technically specifying a precision or field width here 1726 // makes no sense. Worth issuing a warning at some point. 1727 return true; 1728 } 1729 1730 // Consume the argument. 1731 unsigned argIndex = FS.getArgIndex(); 1732 if (argIndex < NumDataArgs) { 1733 // The check to see if the argIndex is valid will come later. 1734 // We set the bit here because we may exit early from this 1735 // function if we encounter some other error. 1736 CoveredArgs.set(argIndex); 1737 } 1738 1739 // Check the length modifier is valid with the given conversion specifier. 1740 const LengthModifier &LM = FS.getLengthModifier(); 1741 if (!FS.hasValidLengthModifier()) { 1742 S.Diag(getLocationOfByte(LM.getStart()), 1743 diag::warn_format_nonsensical_length) 1744 << LM.toString() << CS.toString() 1745 << getSpecifierRange(startSpecifier, specifierLen) 1746 << FixItHint::CreateRemoval(getSpecifierRange(LM.getStart(), 1747 LM.getLength())); 1748 } 1749 1750 // The remaining checks depend on the data arguments. 1751 if (HasVAListArg) 1752 return true; 1753 1754 if (!CheckNumArgs(FS, CS, startSpecifier, specifierLen, argIndex)) 1755 return false; 1756 1757 // FIXME: Check that the argument type matches the format specifier. 1758 1759 return true; 1760} 1761 1762void Sema::CheckFormatString(const StringLiteral *FExpr, 1763 const Expr *OrigFormatExpr, 1764 const CallExpr *TheCall, bool HasVAListArg, 1765 unsigned format_idx, unsigned firstDataArg, 1766 bool isPrintf) { 1767 1768 // CHECK: is the format string a wide literal? 1769 if (FExpr->isWide()) { 1770 Diag(FExpr->getLocStart(), 1771 diag::warn_format_string_is_wide_literal) 1772 << OrigFormatExpr->getSourceRange(); 1773 return; 1774 } 1775 1776 // Str - The format string. NOTE: this is NOT null-terminated! 1777 llvm::StringRef StrRef = FExpr->getString(); 1778 const char *Str = StrRef.data(); 1779 unsigned StrLen = StrRef.size(); 1780 1781 // CHECK: empty format string? 1782 if (StrLen == 0) { 1783 Diag(FExpr->getLocStart(), diag::warn_empty_format_string) 1784 << OrigFormatExpr->getSourceRange(); 1785 return; 1786 } 1787 1788 if (isPrintf) { 1789 CheckPrintfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg, 1790 TheCall->getNumArgs() - firstDataArg, 1791 isa<ObjCStringLiteral>(OrigFormatExpr), Str, 1792 HasVAListArg, TheCall, format_idx); 1793 1794 if (!analyze_format_string::ParsePrintfString(H, Str, Str + StrLen)) 1795 H.DoneProcessing(); 1796 } 1797 else { 1798 CheckScanfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg, 1799 TheCall->getNumArgs() - firstDataArg, 1800 isa<ObjCStringLiteral>(OrigFormatExpr), Str, 1801 HasVAListArg, TheCall, format_idx); 1802 1803 if (!analyze_format_string::ParseScanfString(H, Str, Str + StrLen)) 1804 H.DoneProcessing(); 1805 } 1806} 1807 1808//===--- CHECK: Return Address of Stack Variable --------------------------===// 1809 1810static DeclRefExpr* EvalVal(Expr *E); 1811static DeclRefExpr* EvalAddr(Expr* E); 1812 1813/// CheckReturnStackAddr - Check if a return statement returns the address 1814/// of a stack variable. 1815void 1816Sema::CheckReturnStackAddr(Expr *RetValExp, QualType lhsType, 1817 SourceLocation ReturnLoc) { 1818 1819 // Perform checking for returned stack addresses. 1820 if (lhsType->isPointerType() || lhsType->isBlockPointerType()) { 1821 if (DeclRefExpr *DR = EvalAddr(RetValExp)) 1822 Diag(DR->getLocStart(), diag::warn_ret_stack_addr) 1823 << DR->getDecl()->getDeclName() << RetValExp->getSourceRange(); 1824 1825 // Skip over implicit cast expressions when checking for block expressions. 1826 RetValExp = RetValExp->IgnoreParenCasts(); 1827 1828 if (BlockExpr *C = dyn_cast<BlockExpr>(RetValExp)) 1829 if (C->hasBlockDeclRefExprs()) 1830 Diag(C->getLocStart(), diag::err_ret_local_block) 1831 << C->getSourceRange(); 1832 1833 if (AddrLabelExpr *ALE = dyn_cast<AddrLabelExpr>(RetValExp)) 1834 Diag(ALE->getLocStart(), diag::warn_ret_addr_label) 1835 << ALE->getSourceRange(); 1836 1837 } else if (lhsType->isReferenceType()) { 1838 // Perform checking for stack values returned by reference. 1839 // Check for a reference to the stack 1840 if (DeclRefExpr *DR = EvalVal(RetValExp)) 1841 Diag(DR->getLocStart(), diag::warn_ret_stack_ref) 1842 << DR->getDecl()->getDeclName() << RetValExp->getSourceRange(); 1843 } 1844} 1845 1846/// EvalAddr - EvalAddr and EvalVal are mutually recursive functions that 1847/// check if the expression in a return statement evaluates to an address 1848/// to a location on the stack. The recursion is used to traverse the 1849/// AST of the return expression, with recursion backtracking when we 1850/// encounter a subexpression that (1) clearly does not lead to the address 1851/// of a stack variable or (2) is something we cannot determine leads to 1852/// the address of a stack variable based on such local checking. 1853/// 1854/// EvalAddr processes expressions that are pointers that are used as 1855/// references (and not L-values). EvalVal handles all other values. 1856/// At the base case of the recursion is a check for a DeclRefExpr* in 1857/// the refers to a stack variable. 1858/// 1859/// This implementation handles: 1860/// 1861/// * pointer-to-pointer casts 1862/// * implicit conversions from array references to pointers 1863/// * taking the address of fields 1864/// * arbitrary interplay between "&" and "*" operators 1865/// * pointer arithmetic from an address of a stack variable 1866/// * taking the address of an array element where the array is on the stack 1867static DeclRefExpr* EvalAddr(Expr *E) { 1868 // We should only be called for evaluating pointer expressions. 1869 assert((E->getType()->isAnyPointerType() || 1870 E->getType()->isBlockPointerType() || 1871 E->getType()->isObjCQualifiedIdType()) && 1872 "EvalAddr only works on pointers"); 1873 1874 // Our "symbolic interpreter" is just a dispatch off the currently 1875 // viewed AST node. We then recursively traverse the AST by calling 1876 // EvalAddr and EvalVal appropriately. 1877 switch (E->getStmtClass()) { 1878 case Stmt::ParenExprClass: 1879 // Ignore parentheses. 1880 return EvalAddr(cast<ParenExpr>(E)->getSubExpr()); 1881 1882 case Stmt::UnaryOperatorClass: { 1883 // The only unary operator that make sense to handle here 1884 // is AddrOf. All others don't make sense as pointers. 1885 UnaryOperator *U = cast<UnaryOperator>(E); 1886 1887 if (U->getOpcode() == UO_AddrOf) 1888 return EvalVal(U->getSubExpr()); 1889 else 1890 return NULL; 1891 } 1892 1893 case Stmt::BinaryOperatorClass: { 1894 // Handle pointer arithmetic. All other binary operators are not valid 1895 // in this context. 1896 BinaryOperator *B = cast<BinaryOperator>(E); 1897 BinaryOperatorKind op = B->getOpcode(); 1898 1899 if (op != BO_Add && op != BO_Sub) 1900 return NULL; 1901 1902 Expr *Base = B->getLHS(); 1903 1904 // Determine which argument is the real pointer base. It could be 1905 // the RHS argument instead of the LHS. 1906 if (!Base->getType()->isPointerType()) Base = B->getRHS(); 1907 1908 assert (Base->getType()->isPointerType()); 1909 return EvalAddr(Base); 1910 } 1911 1912 // For conditional operators we need to see if either the LHS or RHS are 1913 // valid DeclRefExpr*s. If one of them is valid, we return it. 1914 case Stmt::ConditionalOperatorClass: { 1915 ConditionalOperator *C = cast<ConditionalOperator>(E); 1916 1917 // Handle the GNU extension for missing LHS. 1918 if (Expr *lhsExpr = C->getLHS()) 1919 if (DeclRefExpr* LHS = EvalAddr(lhsExpr)) 1920 return LHS; 1921 1922 return EvalAddr(C->getRHS()); 1923 } 1924 1925 // For casts, we need to handle conversions from arrays to 1926 // pointer values, and pointer-to-pointer conversions. 1927 case Stmt::ImplicitCastExprClass: 1928 case Stmt::CStyleCastExprClass: 1929 case Stmt::CXXFunctionalCastExprClass: { 1930 Expr* SubExpr = cast<CastExpr>(E)->getSubExpr(); 1931 QualType T = SubExpr->getType(); 1932 1933 if (SubExpr->getType()->isPointerType() || 1934 SubExpr->getType()->isBlockPointerType() || 1935 SubExpr->getType()->isObjCQualifiedIdType()) 1936 return EvalAddr(SubExpr); 1937 else if (T->isArrayType()) 1938 return EvalVal(SubExpr); 1939 else 1940 return 0; 1941 } 1942 1943 // C++ casts. For dynamic casts, static casts, and const casts, we 1944 // are always converting from a pointer-to-pointer, so we just blow 1945 // through the cast. In the case the dynamic cast doesn't fail (and 1946 // return NULL), we take the conservative route and report cases 1947 // where we return the address of a stack variable. For Reinterpre 1948 // FIXME: The comment about is wrong; we're not always converting 1949 // from pointer to pointer. I'm guessing that this code should also 1950 // handle references to objects. 1951 case Stmt::CXXStaticCastExprClass: 1952 case Stmt::CXXDynamicCastExprClass: 1953 case Stmt::CXXConstCastExprClass: 1954 case Stmt::CXXReinterpretCastExprClass: { 1955 Expr *S = cast<CXXNamedCastExpr>(E)->getSubExpr(); 1956 if (S->getType()->isPointerType() || S->getType()->isBlockPointerType()) 1957 return EvalAddr(S); 1958 else 1959 return NULL; 1960 } 1961 1962 // Everything else: we simply don't reason about them. 1963 default: 1964 return NULL; 1965 } 1966} 1967 1968 1969/// EvalVal - This function is complements EvalAddr in the mutual recursion. 1970/// See the comments for EvalAddr for more details. 1971static DeclRefExpr* EvalVal(Expr *E) { 1972do { 1973 // We should only be called for evaluating non-pointer expressions, or 1974 // expressions with a pointer type that are not used as references but instead 1975 // are l-values (e.g., DeclRefExpr with a pointer type). 1976 1977 // Our "symbolic interpreter" is just a dispatch off the currently 1978 // viewed AST node. We then recursively traverse the AST by calling 1979 // EvalAddr and EvalVal appropriately. 1980 switch (E->getStmtClass()) { 1981 case Stmt::ImplicitCastExprClass: { 1982 ImplicitCastExpr *IE = cast<ImplicitCastExpr>(E); 1983 if (IE->getValueKind() == VK_LValue) { 1984 E = IE->getSubExpr(); 1985 continue; 1986 } 1987 return NULL; 1988 } 1989 1990 case Stmt::DeclRefExprClass: { 1991 // DeclRefExpr: the base case. When we hit a DeclRefExpr we are looking 1992 // at code that refers to a variable's name. We check if it has local 1993 // storage within the function, and if so, return the expression. 1994 DeclRefExpr *DR = cast<DeclRefExpr>(E); 1995 1996 if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl())) 1997 if (V->hasLocalStorage() && !V->getType()->isReferenceType()) return DR; 1998 1999 return NULL; 2000 } 2001 2002 case Stmt::ParenExprClass: { 2003 // Ignore parentheses. 2004 E = cast<ParenExpr>(E)->getSubExpr(); 2005 continue; 2006 } 2007 2008 case Stmt::UnaryOperatorClass: { 2009 // The only unary operator that make sense to handle here 2010 // is Deref. All others don't resolve to a "name." This includes 2011 // handling all sorts of rvalues passed to a unary operator. 2012 UnaryOperator *U = cast<UnaryOperator>(E); 2013 2014 if (U->getOpcode() == UO_Deref) 2015 return EvalAddr(U->getSubExpr()); 2016 2017 return NULL; 2018 } 2019 2020 case Stmt::ArraySubscriptExprClass: { 2021 // Array subscripts are potential references to data on the stack. We 2022 // retrieve the DeclRefExpr* for the array variable if it indeed 2023 // has local storage. 2024 return EvalAddr(cast<ArraySubscriptExpr>(E)->getBase()); 2025 } 2026 2027 case Stmt::ConditionalOperatorClass: { 2028 // For conditional operators we need to see if either the LHS or RHS are 2029 // non-NULL DeclRefExpr's. If one is non-NULL, we return it. 2030 ConditionalOperator *C = cast<ConditionalOperator>(E); 2031 2032 // Handle the GNU extension for missing LHS. 2033 if (Expr *lhsExpr = C->getLHS()) 2034 if (DeclRefExpr *LHS = EvalVal(lhsExpr)) 2035 return LHS; 2036 2037 return EvalVal(C->getRHS()); 2038 } 2039 2040 // Accesses to members are potential references to data on the stack. 2041 case Stmt::MemberExprClass: { 2042 MemberExpr *M = cast<MemberExpr>(E); 2043 2044 // Check for indirect access. We only want direct field accesses. 2045 if (M->isArrow()) 2046 return NULL; 2047 2048 // Check whether the member type is itself a reference, in which case 2049 // we're not going to refer to the member, but to what the member refers to. 2050 if (M->getMemberDecl()->getType()->isReferenceType()) 2051 return NULL; 2052 2053 return EvalVal(M->getBase()); 2054 } 2055 2056 // Everything else: we simply don't reason about them. 2057 default: 2058 return NULL; 2059 } 2060} while (true); 2061} 2062 2063//===--- CHECK: Floating-Point comparisons (-Wfloat-equal) ---------------===// 2064 2065/// Check for comparisons of floating point operands using != and ==. 2066/// Issue a warning if these are no self-comparisons, as they are not likely 2067/// to do what the programmer intended. 2068void Sema::CheckFloatComparison(SourceLocation loc, Expr* lex, Expr *rex) { 2069 bool EmitWarning = true; 2070 2071 Expr* LeftExprSansParen = lex->IgnoreParens(); 2072 Expr* RightExprSansParen = rex->IgnoreParens(); 2073 2074 // Special case: check for x == x (which is OK). 2075 // Do not emit warnings for such cases. 2076 if (DeclRefExpr* DRL = dyn_cast<DeclRefExpr>(LeftExprSansParen)) 2077 if (DeclRefExpr* DRR = dyn_cast<DeclRefExpr>(RightExprSansParen)) 2078 if (DRL->getDecl() == DRR->getDecl()) 2079 EmitWarning = false; 2080 2081 2082 // Special case: check for comparisons against literals that can be exactly 2083 // represented by APFloat. In such cases, do not emit a warning. This 2084 // is a heuristic: often comparison against such literals are used to 2085 // detect if a value in a variable has not changed. This clearly can 2086 // lead to false negatives. 2087 if (EmitWarning) { 2088 if (FloatingLiteral* FLL = dyn_cast<FloatingLiteral>(LeftExprSansParen)) { 2089 if (FLL->isExact()) 2090 EmitWarning = false; 2091 } else 2092 if (FloatingLiteral* FLR = dyn_cast<FloatingLiteral>(RightExprSansParen)){ 2093 if (FLR->isExact()) 2094 EmitWarning = false; 2095 } 2096 } 2097 2098 // Check for comparisons with builtin types. 2099 if (EmitWarning) 2100 if (CallExpr* CL = dyn_cast<CallExpr>(LeftExprSansParen)) 2101 if (CL->isBuiltinCall(Context)) 2102 EmitWarning = false; 2103 2104 if (EmitWarning) 2105 if (CallExpr* CR = dyn_cast<CallExpr>(RightExprSansParen)) 2106 if (CR->isBuiltinCall(Context)) 2107 EmitWarning = false; 2108 2109 // Emit the diagnostic. 2110 if (EmitWarning) 2111 Diag(loc, diag::warn_floatingpoint_eq) 2112 << lex->getSourceRange() << rex->getSourceRange(); 2113} 2114 2115//===--- CHECK: Integer mixed-sign comparisons (-Wsign-compare) --------===// 2116//===--- CHECK: Lossy implicit conversions (-Wconversion) --------------===// 2117 2118namespace { 2119 2120/// Structure recording the 'active' range of an integer-valued 2121/// expression. 2122struct IntRange { 2123 /// The number of bits active in the int. 2124 unsigned Width; 2125 2126 /// True if the int is known not to have negative values. 2127 bool NonNegative; 2128 2129 IntRange(unsigned Width, bool NonNegative) 2130 : Width(Width), NonNegative(NonNegative) 2131 {} 2132 2133 // Returns the range of the bool type. 2134 static IntRange forBoolType() { 2135 return IntRange(1, true); 2136 } 2137 2138 // Returns the range of an integral type. 2139 static IntRange forType(ASTContext &C, QualType T) { 2140 return forCanonicalType(C, T->getCanonicalTypeInternal().getTypePtr()); 2141 } 2142 2143 // Returns the range of an integeral type based on its canonical 2144 // representation. 2145 static IntRange forCanonicalType(ASTContext &C, const Type *T) { 2146 assert(T->isCanonicalUnqualified()); 2147 2148 if (const VectorType *VT = dyn_cast<VectorType>(T)) 2149 T = VT->getElementType().getTypePtr(); 2150 if (const ComplexType *CT = dyn_cast<ComplexType>(T)) 2151 T = CT->getElementType().getTypePtr(); 2152 2153 if (const EnumType *ET = dyn_cast<EnumType>(T)) { 2154 EnumDecl *Enum = ET->getDecl(); 2155 unsigned NumPositive = Enum->getNumPositiveBits(); 2156 unsigned NumNegative = Enum->getNumNegativeBits(); 2157 2158 return IntRange(std::max(NumPositive, NumNegative), NumNegative == 0); 2159 } 2160 2161 const BuiltinType *BT = cast<BuiltinType>(T); 2162 assert(BT->isInteger()); 2163 2164 return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger()); 2165 } 2166 2167 // Returns the supremum of two ranges: i.e. their conservative merge. 2168 static IntRange join(IntRange L, IntRange R) { 2169 return IntRange(std::max(L.Width, R.Width), 2170 L.NonNegative && R.NonNegative); 2171 } 2172 2173 // Returns the infinum of two ranges: i.e. their aggressive merge. 2174 static IntRange meet(IntRange L, IntRange R) { 2175 return IntRange(std::min(L.Width, R.Width), 2176 L.NonNegative || R.NonNegative); 2177 } 2178}; 2179 2180IntRange GetValueRange(ASTContext &C, llvm::APSInt &value, unsigned MaxWidth) { 2181 if (value.isSigned() && value.isNegative()) 2182 return IntRange(value.getMinSignedBits(), false); 2183 2184 if (value.getBitWidth() > MaxWidth) 2185 value.trunc(MaxWidth); 2186 2187 // isNonNegative() just checks the sign bit without considering 2188 // signedness. 2189 return IntRange(value.getActiveBits(), true); 2190} 2191 2192IntRange GetValueRange(ASTContext &C, APValue &result, QualType Ty, 2193 unsigned MaxWidth) { 2194 if (result.isInt()) 2195 return GetValueRange(C, result.getInt(), MaxWidth); 2196 2197 if (result.isVector()) { 2198 IntRange R = GetValueRange(C, result.getVectorElt(0), Ty, MaxWidth); 2199 for (unsigned i = 1, e = result.getVectorLength(); i != e; ++i) { 2200 IntRange El = GetValueRange(C, result.getVectorElt(i), Ty, MaxWidth); 2201 R = IntRange::join(R, El); 2202 } 2203 return R; 2204 } 2205 2206 if (result.isComplexInt()) { 2207 IntRange R = GetValueRange(C, result.getComplexIntReal(), MaxWidth); 2208 IntRange I = GetValueRange(C, result.getComplexIntImag(), MaxWidth); 2209 return IntRange::join(R, I); 2210 } 2211 2212 // This can happen with lossless casts to intptr_t of "based" lvalues. 2213 // Assume it might use arbitrary bits. 2214 // FIXME: The only reason we need to pass the type in here is to get 2215 // the sign right on this one case. It would be nice if APValue 2216 // preserved this. 2217 assert(result.isLValue()); 2218 return IntRange(MaxWidth, Ty->isUnsignedIntegerType()); 2219} 2220 2221/// Pseudo-evaluate the given integer expression, estimating the 2222/// range of values it might take. 2223/// 2224/// \param MaxWidth - the width to which the value will be truncated 2225IntRange GetExprRange(ASTContext &C, Expr *E, unsigned MaxWidth) { 2226 E = E->IgnoreParens(); 2227 2228 // Try a full evaluation first. 2229 Expr::EvalResult result; 2230 if (E->Evaluate(result, C)) 2231 return GetValueRange(C, result.Val, E->getType(), MaxWidth); 2232 2233 // I think we only want to look through implicit casts here; if the 2234 // user has an explicit widening cast, we should treat the value as 2235 // being of the new, wider type. 2236 if (ImplicitCastExpr *CE = dyn_cast<ImplicitCastExpr>(E)) { 2237 if (CE->getCastKind() == CK_NoOp) 2238 return GetExprRange(C, CE->getSubExpr(), MaxWidth); 2239 2240 IntRange OutputTypeRange = IntRange::forType(C, CE->getType()); 2241 2242 bool isIntegerCast = (CE->getCastKind() == CK_IntegralCast); 2243 if (!isIntegerCast && CE->getCastKind() == CK_Unknown) 2244 isIntegerCast = CE->getSubExpr()->getType()->isIntegerType(); 2245 2246 // Assume that non-integer casts can span the full range of the type. 2247 if (!isIntegerCast) 2248 return OutputTypeRange; 2249 2250 IntRange SubRange 2251 = GetExprRange(C, CE->getSubExpr(), 2252 std::min(MaxWidth, OutputTypeRange.Width)); 2253 2254 // Bail out if the subexpr's range is as wide as the cast type. 2255 if (SubRange.Width >= OutputTypeRange.Width) 2256 return OutputTypeRange; 2257 2258 // Otherwise, we take the smaller width, and we're non-negative if 2259 // either the output type or the subexpr is. 2260 return IntRange(SubRange.Width, 2261 SubRange.NonNegative || OutputTypeRange.NonNegative); 2262 } 2263 2264 if (ConditionalOperator *CO = dyn_cast<ConditionalOperator>(E)) { 2265 // If we can fold the condition, just take that operand. 2266 bool CondResult; 2267 if (CO->getCond()->EvaluateAsBooleanCondition(CondResult, C)) 2268 return GetExprRange(C, CondResult ? CO->getTrueExpr() 2269 : CO->getFalseExpr(), 2270 MaxWidth); 2271 2272 // Otherwise, conservatively merge. 2273 IntRange L = GetExprRange(C, CO->getTrueExpr(), MaxWidth); 2274 IntRange R = GetExprRange(C, CO->getFalseExpr(), MaxWidth); 2275 return IntRange::join(L, R); 2276 } 2277 2278 if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) { 2279 switch (BO->getOpcode()) { 2280 2281 // Boolean-valued operations are single-bit and positive. 2282 case BO_LAnd: 2283 case BO_LOr: 2284 case BO_LT: 2285 case BO_GT: 2286 case BO_LE: 2287 case BO_GE: 2288 case BO_EQ: 2289 case BO_NE: 2290 return IntRange::forBoolType(); 2291 2292 // The type of these compound assignments is the type of the LHS, 2293 // so the RHS is not necessarily an integer. 2294 case BO_MulAssign: 2295 case BO_DivAssign: 2296 case BO_RemAssign: 2297 case BO_AddAssign: 2298 case BO_SubAssign: 2299 return IntRange::forType(C, E->getType()); 2300 2301 // Operations with opaque sources are black-listed. 2302 case BO_PtrMemD: 2303 case BO_PtrMemI: 2304 return IntRange::forType(C, E->getType()); 2305 2306 // Bitwise-and uses the *infinum* of the two source ranges. 2307 case BO_And: 2308 case BO_AndAssign: 2309 return IntRange::meet(GetExprRange(C, BO->getLHS(), MaxWidth), 2310 GetExprRange(C, BO->getRHS(), MaxWidth)); 2311 2312 // Left shift gets black-listed based on a judgement call. 2313 case BO_Shl: 2314 // ...except that we want to treat '1 << (blah)' as logically 2315 // positive. It's an important idiom. 2316 if (IntegerLiteral *I 2317 = dyn_cast<IntegerLiteral>(BO->getLHS()->IgnoreParenCasts())) { 2318 if (I->getValue() == 1) { 2319 IntRange R = IntRange::forType(C, E->getType()); 2320 return IntRange(R.Width, /*NonNegative*/ true); 2321 } 2322 } 2323 // fallthrough 2324 2325 case BO_ShlAssign: 2326 return IntRange::forType(C, E->getType()); 2327 2328 // Right shift by a constant can narrow its left argument. 2329 case BO_Shr: 2330 case BO_ShrAssign: { 2331 IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth); 2332 2333 // If the shift amount is a positive constant, drop the width by 2334 // that much. 2335 llvm::APSInt shift; 2336 if (BO->getRHS()->isIntegerConstantExpr(shift, C) && 2337 shift.isNonNegative()) { 2338 unsigned zext = shift.getZExtValue(); 2339 if (zext >= L.Width) 2340 L.Width = (L.NonNegative ? 0 : 1); 2341 else 2342 L.Width -= zext; 2343 } 2344 2345 return L; 2346 } 2347 2348 // Comma acts as its right operand. 2349 case BO_Comma: 2350 return GetExprRange(C, BO->getRHS(), MaxWidth); 2351 2352 // Black-list pointer subtractions. 2353 case BO_Sub: 2354 if (BO->getLHS()->getType()->isPointerType()) 2355 return IntRange::forType(C, E->getType()); 2356 // fallthrough 2357 2358 default: 2359 break; 2360 } 2361 2362 // Treat every other operator as if it were closed on the 2363 // narrowest type that encompasses both operands. 2364 IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth); 2365 IntRange R = GetExprRange(C, BO->getRHS(), MaxWidth); 2366 return IntRange::join(L, R); 2367 } 2368 2369 if (UnaryOperator *UO = dyn_cast<UnaryOperator>(E)) { 2370 switch (UO->getOpcode()) { 2371 // Boolean-valued operations are white-listed. 2372 case UO_LNot: 2373 return IntRange::forBoolType(); 2374 2375 // Operations with opaque sources are black-listed. 2376 case UO_Deref: 2377 case UO_AddrOf: // should be impossible 2378 return IntRange::forType(C, E->getType()); 2379 2380 default: 2381 return GetExprRange(C, UO->getSubExpr(), MaxWidth); 2382 } 2383 } 2384 2385 if (dyn_cast<OffsetOfExpr>(E)) { 2386 IntRange::forType(C, E->getType()); 2387 } 2388 2389 FieldDecl *BitField = E->getBitField(); 2390 if (BitField) { 2391 llvm::APSInt BitWidthAP = BitField->getBitWidth()->EvaluateAsInt(C); 2392 unsigned BitWidth = BitWidthAP.getZExtValue(); 2393 2394 return IntRange(BitWidth, BitField->getType()->isUnsignedIntegerType()); 2395 } 2396 2397 return IntRange::forType(C, E->getType()); 2398} 2399 2400IntRange GetExprRange(ASTContext &C, Expr *E) { 2401 return GetExprRange(C, E, C.getIntWidth(E->getType())); 2402} 2403 2404/// Checks whether the given value, which currently has the given 2405/// source semantics, has the same value when coerced through the 2406/// target semantics. 2407bool IsSameFloatAfterCast(const llvm::APFloat &value, 2408 const llvm::fltSemantics &Src, 2409 const llvm::fltSemantics &Tgt) { 2410 llvm::APFloat truncated = value; 2411 2412 bool ignored; 2413 truncated.convert(Src, llvm::APFloat::rmNearestTiesToEven, &ignored); 2414 truncated.convert(Tgt, llvm::APFloat::rmNearestTiesToEven, &ignored); 2415 2416 return truncated.bitwiseIsEqual(value); 2417} 2418 2419/// Checks whether the given value, which currently has the given 2420/// source semantics, has the same value when coerced through the 2421/// target semantics. 2422/// 2423/// The value might be a vector of floats (or a complex number). 2424bool IsSameFloatAfterCast(const APValue &value, 2425 const llvm::fltSemantics &Src, 2426 const llvm::fltSemantics &Tgt) { 2427 if (value.isFloat()) 2428 return IsSameFloatAfterCast(value.getFloat(), Src, Tgt); 2429 2430 if (value.isVector()) { 2431 for (unsigned i = 0, e = value.getVectorLength(); i != e; ++i) 2432 if (!IsSameFloatAfterCast(value.getVectorElt(i), Src, Tgt)) 2433 return false; 2434 return true; 2435 } 2436 2437 assert(value.isComplexFloat()); 2438 return (IsSameFloatAfterCast(value.getComplexFloatReal(), Src, Tgt) && 2439 IsSameFloatAfterCast(value.getComplexFloatImag(), Src, Tgt)); 2440} 2441 2442void AnalyzeImplicitConversions(Sema &S, Expr *E); 2443 2444bool IsZero(Sema &S, Expr *E) { 2445 llvm::APSInt Value; 2446 return E->isIntegerConstantExpr(Value, S.Context) && Value == 0; 2447} 2448 2449void CheckTrivialUnsignedComparison(Sema &S, BinaryOperator *E) { 2450 BinaryOperatorKind op = E->getOpcode(); 2451 if (op == BO_LT && IsZero(S, E->getRHS())) { 2452 S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison) 2453 << "< 0" << "false" 2454 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 2455 } else if (op == BO_GE && IsZero(S, E->getRHS())) { 2456 S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison) 2457 << ">= 0" << "true" 2458 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 2459 } else if (op == BO_GT && IsZero(S, E->getLHS())) { 2460 S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison) 2461 << "0 >" << "false" 2462 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 2463 } else if (op == BO_LE && IsZero(S, E->getLHS())) { 2464 S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison) 2465 << "0 <=" << "true" 2466 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 2467 } 2468} 2469 2470/// Analyze the operands of the given comparison. Implements the 2471/// fallback case from AnalyzeComparison. 2472void AnalyzeImpConvsInComparison(Sema &S, BinaryOperator *E) { 2473 AnalyzeImplicitConversions(S, E->getLHS()); 2474 AnalyzeImplicitConversions(S, E->getRHS()); 2475} 2476 2477/// \brief Implements -Wsign-compare. 2478/// 2479/// \param lex the left-hand expression 2480/// \param rex the right-hand expression 2481/// \param OpLoc the location of the joining operator 2482/// \param BinOpc binary opcode or 0 2483void AnalyzeComparison(Sema &S, BinaryOperator *E) { 2484 // The type the comparison is being performed in. 2485 QualType T = E->getLHS()->getType(); 2486 assert(S.Context.hasSameUnqualifiedType(T, E->getRHS()->getType()) 2487 && "comparison with mismatched types"); 2488 2489 // We don't do anything special if this isn't an unsigned integral 2490 // comparison: we're only interested in integral comparisons, and 2491 // signed comparisons only happen in cases we don't care to warn about. 2492 if (!T->hasUnsignedIntegerRepresentation()) 2493 return AnalyzeImpConvsInComparison(S, E); 2494 2495 Expr *lex = E->getLHS()->IgnoreParenImpCasts(); 2496 Expr *rex = E->getRHS()->IgnoreParenImpCasts(); 2497 2498 // Check to see if one of the (unmodified) operands is of different 2499 // signedness. 2500 Expr *signedOperand, *unsignedOperand; 2501 if (lex->getType()->hasSignedIntegerRepresentation()) { 2502 assert(!rex->getType()->hasSignedIntegerRepresentation() && 2503 "unsigned comparison between two signed integer expressions?"); 2504 signedOperand = lex; 2505 unsignedOperand = rex; 2506 } else if (rex->getType()->hasSignedIntegerRepresentation()) { 2507 signedOperand = rex; 2508 unsignedOperand = lex; 2509 } else { 2510 CheckTrivialUnsignedComparison(S, E); 2511 return AnalyzeImpConvsInComparison(S, E); 2512 } 2513 2514 // Otherwise, calculate the effective range of the signed operand. 2515 IntRange signedRange = GetExprRange(S.Context, signedOperand); 2516 2517 // Go ahead and analyze implicit conversions in the operands. Note 2518 // that we skip the implicit conversions on both sides. 2519 AnalyzeImplicitConversions(S, lex); 2520 AnalyzeImplicitConversions(S, rex); 2521 2522 // If the signed range is non-negative, -Wsign-compare won't fire, 2523 // but we should still check for comparisons which are always true 2524 // or false. 2525 if (signedRange.NonNegative) 2526 return CheckTrivialUnsignedComparison(S, E); 2527 2528 // For (in)equality comparisons, if the unsigned operand is a 2529 // constant which cannot collide with a overflowed signed operand, 2530 // then reinterpreting the signed operand as unsigned will not 2531 // change the result of the comparison. 2532 if (E->isEqualityOp()) { 2533 unsigned comparisonWidth = S.Context.getIntWidth(T); 2534 IntRange unsignedRange = GetExprRange(S.Context, unsignedOperand); 2535 2536 // We should never be unable to prove that the unsigned operand is 2537 // non-negative. 2538 assert(unsignedRange.NonNegative && "unsigned range includes negative?"); 2539 2540 if (unsignedRange.Width < comparisonWidth) 2541 return; 2542 } 2543 2544 S.Diag(E->getOperatorLoc(), diag::warn_mixed_sign_comparison) 2545 << lex->getType() << rex->getType() 2546 << lex->getSourceRange() << rex->getSourceRange(); 2547} 2548 2549/// Diagnose an implicit cast; purely a helper for CheckImplicitConversion. 2550void DiagnoseImpCast(Sema &S, Expr *E, QualType T, unsigned diag) { 2551 S.Diag(E->getExprLoc(), diag) << E->getType() << T << E->getSourceRange(); 2552} 2553 2554void CheckImplicitConversion(Sema &S, Expr *E, QualType T, 2555 bool *ICContext = 0) { 2556 if (E->isTypeDependent() || E->isValueDependent()) return; 2557 2558 const Type *Source = S.Context.getCanonicalType(E->getType()).getTypePtr(); 2559 const Type *Target = S.Context.getCanonicalType(T).getTypePtr(); 2560 if (Source == Target) return; 2561 if (Target->isDependentType()) return; 2562 2563 // Never diagnose implicit casts to bool. 2564 if (Target->isSpecificBuiltinType(BuiltinType::Bool)) 2565 return; 2566 2567 // Strip vector types. 2568 if (isa<VectorType>(Source)) { 2569 if (!isa<VectorType>(Target)) 2570 return DiagnoseImpCast(S, E, T, diag::warn_impcast_vector_scalar); 2571 2572 Source = cast<VectorType>(Source)->getElementType().getTypePtr(); 2573 Target = cast<VectorType>(Target)->getElementType().getTypePtr(); 2574 } 2575 2576 // Strip complex types. 2577 if (isa<ComplexType>(Source)) { 2578 if (!isa<ComplexType>(Target)) 2579 return DiagnoseImpCast(S, E, T, diag::warn_impcast_complex_scalar); 2580 2581 Source = cast<ComplexType>(Source)->getElementType().getTypePtr(); 2582 Target = cast<ComplexType>(Target)->getElementType().getTypePtr(); 2583 } 2584 2585 const BuiltinType *SourceBT = dyn_cast<BuiltinType>(Source); 2586 const BuiltinType *TargetBT = dyn_cast<BuiltinType>(Target); 2587 2588 // If the source is floating point... 2589 if (SourceBT && SourceBT->isFloatingPoint()) { 2590 // ...and the target is floating point... 2591 if (TargetBT && TargetBT->isFloatingPoint()) { 2592 // ...then warn if we're dropping FP rank. 2593 2594 // Builtin FP kinds are ordered by increasing FP rank. 2595 if (SourceBT->getKind() > TargetBT->getKind()) { 2596 // Don't warn about float constants that are precisely 2597 // representable in the target type. 2598 Expr::EvalResult result; 2599 if (E->Evaluate(result, S.Context)) { 2600 // Value might be a float, a float vector, or a float complex. 2601 if (IsSameFloatAfterCast(result.Val, 2602 S.Context.getFloatTypeSemantics(QualType(TargetBT, 0)), 2603 S.Context.getFloatTypeSemantics(QualType(SourceBT, 0)))) 2604 return; 2605 } 2606 2607 DiagnoseImpCast(S, E, T, diag::warn_impcast_float_precision); 2608 } 2609 return; 2610 } 2611 2612 // If the target is integral, always warn. 2613 if ((TargetBT && TargetBT->isInteger())) 2614 // TODO: don't warn for integer values? 2615 DiagnoseImpCast(S, E, T, diag::warn_impcast_float_integer); 2616 2617 return; 2618 } 2619 2620 if (!Source->isIntegerType() || !Target->isIntegerType()) 2621 return; 2622 2623 IntRange SourceRange = GetExprRange(S.Context, E); 2624 IntRange TargetRange = IntRange::forCanonicalType(S.Context, Target); 2625 2626 if (SourceRange.Width > TargetRange.Width) { 2627 // People want to build with -Wshorten-64-to-32 and not -Wconversion 2628 // and by god we'll let them. 2629 if (SourceRange.Width == 64 && TargetRange.Width == 32) 2630 return DiagnoseImpCast(S, E, T, diag::warn_impcast_integer_64_32); 2631 return DiagnoseImpCast(S, E, T, diag::warn_impcast_integer_precision); 2632 } 2633 2634 if ((TargetRange.NonNegative && !SourceRange.NonNegative) || 2635 (!TargetRange.NonNegative && SourceRange.NonNegative && 2636 SourceRange.Width == TargetRange.Width)) { 2637 unsigned DiagID = diag::warn_impcast_integer_sign; 2638 2639 // Traditionally, gcc has warned about this under -Wsign-compare. 2640 // We also want to warn about it in -Wconversion. 2641 // So if -Wconversion is off, use a completely identical diagnostic 2642 // in the sign-compare group. 2643 // The conditional-checking code will 2644 if (ICContext) { 2645 DiagID = diag::warn_impcast_integer_sign_conditional; 2646 *ICContext = true; 2647 } 2648 2649 return DiagnoseImpCast(S, E, T, DiagID); 2650 } 2651 2652 return; 2653} 2654 2655void CheckConditionalOperator(Sema &S, ConditionalOperator *E, QualType T); 2656 2657void CheckConditionalOperand(Sema &S, Expr *E, QualType T, 2658 bool &ICContext) { 2659 E = E->IgnoreParenImpCasts(); 2660 2661 if (isa<ConditionalOperator>(E)) 2662 return CheckConditionalOperator(S, cast<ConditionalOperator>(E), T); 2663 2664 AnalyzeImplicitConversions(S, E); 2665 if (E->getType() != T) 2666 return CheckImplicitConversion(S, E, T, &ICContext); 2667 return; 2668} 2669 2670void CheckConditionalOperator(Sema &S, ConditionalOperator *E, QualType T) { 2671 AnalyzeImplicitConversions(S, E->getCond()); 2672 2673 bool Suspicious = false; 2674 CheckConditionalOperand(S, E->getTrueExpr(), T, Suspicious); 2675 CheckConditionalOperand(S, E->getFalseExpr(), T, Suspicious); 2676 2677 // If -Wconversion would have warned about either of the candidates 2678 // for a signedness conversion to the context type... 2679 if (!Suspicious) return; 2680 2681 // ...but it's currently ignored... 2682 if (S.Diags.getDiagnosticLevel(diag::warn_impcast_integer_sign_conditional)) 2683 return; 2684 2685 // ...and -Wsign-compare isn't... 2686 if (!S.Diags.getDiagnosticLevel(diag::warn_mixed_sign_conditional)) 2687 return; 2688 2689 // ...then check whether it would have warned about either of the 2690 // candidates for a signedness conversion to the condition type. 2691 if (E->getType() != T) { 2692 Suspicious = false; 2693 CheckImplicitConversion(S, E->getTrueExpr()->IgnoreParenImpCasts(), 2694 E->getType(), &Suspicious); 2695 if (!Suspicious) 2696 CheckImplicitConversion(S, E->getFalseExpr()->IgnoreParenImpCasts(), 2697 E->getType(), &Suspicious); 2698 if (!Suspicious) 2699 return; 2700 } 2701 2702 // If so, emit a diagnostic under -Wsign-compare. 2703 Expr *lex = E->getTrueExpr()->IgnoreParenImpCasts(); 2704 Expr *rex = E->getFalseExpr()->IgnoreParenImpCasts(); 2705 S.Diag(E->getQuestionLoc(), diag::warn_mixed_sign_conditional) 2706 << lex->getType() << rex->getType() 2707 << lex->getSourceRange() << rex->getSourceRange(); 2708} 2709 2710/// AnalyzeImplicitConversions - Find and report any interesting 2711/// implicit conversions in the given expression. There are a couple 2712/// of competing diagnostics here, -Wconversion and -Wsign-compare. 2713void AnalyzeImplicitConversions(Sema &S, Expr *OrigE) { 2714 QualType T = OrigE->getType(); 2715 Expr *E = OrigE->IgnoreParenImpCasts(); 2716 2717 // For conditional operators, we analyze the arguments as if they 2718 // were being fed directly into the output. 2719 if (isa<ConditionalOperator>(E)) { 2720 ConditionalOperator *CO = cast<ConditionalOperator>(E); 2721 CheckConditionalOperator(S, CO, T); 2722 return; 2723 } 2724 2725 // Go ahead and check any implicit conversions we might have skipped. 2726 // The non-canonical typecheck is just an optimization; 2727 // CheckImplicitConversion will filter out dead implicit conversions. 2728 if (E->getType() != T) 2729 CheckImplicitConversion(S, E, T); 2730 2731 // Now continue drilling into this expression. 2732 2733 // Skip past explicit casts. 2734 if (isa<ExplicitCastExpr>(E)) { 2735 E = cast<ExplicitCastExpr>(E)->getSubExpr()->IgnoreParenImpCasts(); 2736 return AnalyzeImplicitConversions(S, E); 2737 } 2738 2739 // Do a somewhat different check with comparison operators. 2740 if (isa<BinaryOperator>(E) && cast<BinaryOperator>(E)->isComparisonOp()) 2741 return AnalyzeComparison(S, cast<BinaryOperator>(E)); 2742 2743 // These break the otherwise-useful invariant below. Fortunately, 2744 // we don't really need to recurse into them, because any internal 2745 // expressions should have been analyzed already when they were 2746 // built into statements. 2747 if (isa<StmtExpr>(E)) return; 2748 2749 // Don't descend into unevaluated contexts. 2750 if (isa<SizeOfAlignOfExpr>(E)) return; 2751 2752 // Now just recurse over the expression's children. 2753 for (Stmt::child_iterator I = E->child_begin(), IE = E->child_end(); 2754 I != IE; ++I) 2755 AnalyzeImplicitConversions(S, cast<Expr>(*I)); 2756} 2757 2758} // end anonymous namespace 2759 2760/// Diagnoses "dangerous" implicit conversions within the given 2761/// expression (which is a full expression). Implements -Wconversion 2762/// and -Wsign-compare. 2763void Sema::CheckImplicitConversions(Expr *E) { 2764 // Don't diagnose in unevaluated contexts. 2765 if (ExprEvalContexts.back().Context == Sema::Unevaluated) 2766 return; 2767 2768 // Don't diagnose for value- or type-dependent expressions. 2769 if (E->isTypeDependent() || E->isValueDependent()) 2770 return; 2771 2772 AnalyzeImplicitConversions(*this, E); 2773} 2774 2775/// CheckParmsForFunctionDef - Check that the parameters of the given 2776/// function are appropriate for the definition of a function. This 2777/// takes care of any checks that cannot be performed on the 2778/// declaration itself, e.g., that the types of each of the function 2779/// parameters are complete. 2780bool Sema::CheckParmsForFunctionDef(FunctionDecl *FD) { 2781 bool HasInvalidParm = false; 2782 for (unsigned p = 0, NumParams = FD->getNumParams(); p < NumParams; ++p) { 2783 ParmVarDecl *Param = FD->getParamDecl(p); 2784 2785 // C99 6.7.5.3p4: the parameters in a parameter type list in a 2786 // function declarator that is part of a function definition of 2787 // that function shall not have incomplete type. 2788 // 2789 // This is also C++ [dcl.fct]p6. 2790 if (!Param->isInvalidDecl() && 2791 RequireCompleteType(Param->getLocation(), Param->getType(), 2792 diag::err_typecheck_decl_incomplete_type)) { 2793 Param->setInvalidDecl(); 2794 HasInvalidParm = true; 2795 } 2796 2797 // C99 6.9.1p5: If the declarator includes a parameter type list, the 2798 // declaration of each parameter shall include an identifier. 2799 if (Param->getIdentifier() == 0 && 2800 !Param->isImplicit() && 2801 !getLangOptions().CPlusPlus) 2802 Diag(Param->getLocation(), diag::err_parameter_name_omitted); 2803 2804 // C99 6.7.5.3p12: 2805 // If the function declarator is not part of a definition of that 2806 // function, parameters may have incomplete type and may use the [*] 2807 // notation in their sequences of declarator specifiers to specify 2808 // variable length array types. 2809 QualType PType = Param->getOriginalType(); 2810 if (const ArrayType *AT = Context.getAsArrayType(PType)) { 2811 if (AT->getSizeModifier() == ArrayType::Star) { 2812 // FIXME: This diagnosic should point the the '[*]' if source-location 2813 // information is added for it. 2814 Diag(Param->getLocation(), diag::err_array_star_in_function_definition); 2815 } 2816 } 2817 } 2818 2819 return HasInvalidParm; 2820} 2821 2822/// CheckCastAlign - Implements -Wcast-align, which warns when a 2823/// pointer cast increases the alignment requirements. 2824void Sema::CheckCastAlign(Expr *Op, QualType T, SourceRange TRange) { 2825 // This is actually a lot of work to potentially be doing on every 2826 // cast; don't do it if we're ignoring -Wcast_align (as is the default). 2827 if (getDiagnostics().getDiagnosticLevel(diag::warn_cast_align) 2828 == Diagnostic::Ignored) 2829 return; 2830 2831 // Ignore dependent types. 2832 if (T->isDependentType() || Op->getType()->isDependentType()) 2833 return; 2834 2835 // Require that the destination be a pointer type. 2836 const PointerType *DestPtr = T->getAs<PointerType>(); 2837 if (!DestPtr) return; 2838 2839 // If the destination has alignment 1, we're done. 2840 QualType DestPointee = DestPtr->getPointeeType(); 2841 if (DestPointee->isIncompleteType()) return; 2842 CharUnits DestAlign = Context.getTypeAlignInChars(DestPointee); 2843 if (DestAlign.isOne()) return; 2844 2845 // Require that the source be a pointer type. 2846 const PointerType *SrcPtr = Op->getType()->getAs<PointerType>(); 2847 if (!SrcPtr) return; 2848 QualType SrcPointee = SrcPtr->getPointeeType(); 2849 2850 // Whitelist casts from cv void*. We already implicitly 2851 // whitelisted casts to cv void*, since they have alignment 1. 2852 // Also whitelist casts involving incomplete types, which implicitly 2853 // includes 'void'. 2854 if (SrcPointee->isIncompleteType()) return; 2855 2856 CharUnits SrcAlign = Context.getTypeAlignInChars(SrcPointee); 2857 if (SrcAlign >= DestAlign) return; 2858 2859 Diag(TRange.getBegin(), diag::warn_cast_align) 2860 << Op->getType() << T 2861 << static_cast<unsigned>(SrcAlign.getQuantity()) 2862 << static_cast<unsigned>(DestAlign.getQuantity()) 2863 << TRange << Op->getSourceRange(); 2864} 2865 2866