SemaChecking.cpp revision c82faca0be59b072b38bed375e21bd24f1f72647
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 and scanf checking. 345 for (specific_attr_iterator<FormatAttr> 346 i = FDecl->specific_attr_begin<FormatAttr>(), 347 e = FDecl->specific_attr_end<FormatAttr>(); i != e ; ++i) { 348 349 const FormatAttr *Format = *i; 350 const bool b = Format->getType() == "scanf"; 351 if (b || CheckablePrintfAttr(Format, TheCall)) { 352 bool HasVAListArg = Format->getFirstArg() == 0; 353 CheckPrintfScanfArguments(TheCall, HasVAListArg, 354 Format->getFormatIdx() - 1, 355 HasVAListArg ? 0 : Format->getFirstArg() - 1, 356 !b); 357 } 358 } 359 360 for (specific_attr_iterator<NonNullAttr> 361 i = FDecl->specific_attr_begin<NonNullAttr>(), 362 e = FDecl->specific_attr_end<NonNullAttr>(); i != e; ++i) { 363 CheckNonNullArguments(*i, TheCall); 364 } 365 366 return false; 367} 368 369bool Sema::CheckBlockCall(NamedDecl *NDecl, CallExpr *TheCall) { 370 // Printf checking. 371 const FormatAttr *Format = NDecl->getAttr<FormatAttr>(); 372 if (!Format) 373 return false; 374 375 const VarDecl *V = dyn_cast<VarDecl>(NDecl); 376 if (!V) 377 return false; 378 379 QualType Ty = V->getType(); 380 if (!Ty->isBlockPointerType()) 381 return false; 382 383 const bool b = Format->getType() == "scanf"; 384 if (!b && !CheckablePrintfAttr(Format, TheCall)) 385 return false; 386 387 bool HasVAListArg = Format->getFirstArg() == 0; 388 CheckPrintfScanfArguments(TheCall, HasVAListArg, Format->getFormatIdx() - 1, 389 HasVAListArg ? 0 : Format->getFirstArg() - 1, !b); 390 391 return false; 392} 393 394/// SemaBuiltinAtomicOverloaded - We have a call to a function like 395/// __sync_fetch_and_add, which is an overloaded function based on the pointer 396/// type of its first argument. The main ActOnCallExpr routines have already 397/// promoted the types of arguments because all of these calls are prototyped as 398/// void(...). 399/// 400/// This function goes through and does final semantic checking for these 401/// builtins, 402ExprResult 403Sema::SemaBuiltinAtomicOverloaded(ExprResult TheCallResult) { 404 CallExpr *TheCall = (CallExpr *)TheCallResult.get(); 405 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 406 FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl()); 407 408 // Ensure that we have at least one argument to do type inference from. 409 if (TheCall->getNumArgs() < 1) { 410 Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args_at_least) 411 << 0 << 1 << TheCall->getNumArgs() 412 << TheCall->getCallee()->getSourceRange(); 413 return ExprError(); 414 } 415 416 // Inspect the first argument of the atomic builtin. This should always be 417 // a pointer type, whose element is an integral scalar or pointer type. 418 // Because it is a pointer type, we don't have to worry about any implicit 419 // casts here. 420 // FIXME: We don't allow floating point scalars as input. 421 Expr *FirstArg = TheCall->getArg(0); 422 if (!FirstArg->getType()->isPointerType()) { 423 Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer) 424 << FirstArg->getType() << FirstArg->getSourceRange(); 425 return ExprError(); 426 } 427 428 QualType ValType = 429 FirstArg->getType()->getAs<PointerType>()->getPointeeType(); 430 if (!ValType->isIntegerType() && !ValType->isPointerType() && 431 !ValType->isBlockPointerType()) { 432 Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer_intptr) 433 << FirstArg->getType() << FirstArg->getSourceRange(); 434 return ExprError(); 435 } 436 437 // The majority of builtins return a value, but a few have special return 438 // types, so allow them to override appropriately below. 439 QualType ResultType = ValType; 440 441 // We need to figure out which concrete builtin this maps onto. For example, 442 // __sync_fetch_and_add with a 2 byte object turns into 443 // __sync_fetch_and_add_2. 444#define BUILTIN_ROW(x) \ 445 { Builtin::BI##x##_1, Builtin::BI##x##_2, Builtin::BI##x##_4, \ 446 Builtin::BI##x##_8, Builtin::BI##x##_16 } 447 448 static const unsigned BuiltinIndices[][5] = { 449 BUILTIN_ROW(__sync_fetch_and_add), 450 BUILTIN_ROW(__sync_fetch_and_sub), 451 BUILTIN_ROW(__sync_fetch_and_or), 452 BUILTIN_ROW(__sync_fetch_and_and), 453 BUILTIN_ROW(__sync_fetch_and_xor), 454 455 BUILTIN_ROW(__sync_add_and_fetch), 456 BUILTIN_ROW(__sync_sub_and_fetch), 457 BUILTIN_ROW(__sync_and_and_fetch), 458 BUILTIN_ROW(__sync_or_and_fetch), 459 BUILTIN_ROW(__sync_xor_and_fetch), 460 461 BUILTIN_ROW(__sync_val_compare_and_swap), 462 BUILTIN_ROW(__sync_bool_compare_and_swap), 463 BUILTIN_ROW(__sync_lock_test_and_set), 464 BUILTIN_ROW(__sync_lock_release) 465 }; 466#undef BUILTIN_ROW 467 468 // Determine the index of the size. 469 unsigned SizeIndex; 470 switch (Context.getTypeSizeInChars(ValType).getQuantity()) { 471 case 1: SizeIndex = 0; break; 472 case 2: SizeIndex = 1; break; 473 case 4: SizeIndex = 2; break; 474 case 8: SizeIndex = 3; break; 475 case 16: SizeIndex = 4; break; 476 default: 477 Diag(DRE->getLocStart(), diag::err_atomic_builtin_pointer_size) 478 << FirstArg->getType() << FirstArg->getSourceRange(); 479 return ExprError(); 480 } 481 482 // Each of these builtins has one pointer argument, followed by some number of 483 // values (0, 1 or 2) followed by a potentially empty varags list of stuff 484 // that we ignore. Find out which row of BuiltinIndices to read from as well 485 // as the number of fixed args. 486 unsigned BuiltinID = FDecl->getBuiltinID(); 487 unsigned BuiltinIndex, NumFixed = 1; 488 switch (BuiltinID) { 489 default: assert(0 && "Unknown overloaded atomic builtin!"); 490 case Builtin::BI__sync_fetch_and_add: BuiltinIndex = 0; break; 491 case Builtin::BI__sync_fetch_and_sub: BuiltinIndex = 1; break; 492 case Builtin::BI__sync_fetch_and_or: BuiltinIndex = 2; break; 493 case Builtin::BI__sync_fetch_and_and: BuiltinIndex = 3; break; 494 case Builtin::BI__sync_fetch_and_xor: BuiltinIndex = 4; break; 495 496 case Builtin::BI__sync_add_and_fetch: BuiltinIndex = 5; break; 497 case Builtin::BI__sync_sub_and_fetch: BuiltinIndex = 6; break; 498 case Builtin::BI__sync_and_and_fetch: BuiltinIndex = 7; break; 499 case Builtin::BI__sync_or_and_fetch: BuiltinIndex = 8; break; 500 case Builtin::BI__sync_xor_and_fetch: BuiltinIndex = 9; break; 501 502 case Builtin::BI__sync_val_compare_and_swap: 503 BuiltinIndex = 10; 504 NumFixed = 2; 505 break; 506 case Builtin::BI__sync_bool_compare_and_swap: 507 BuiltinIndex = 11; 508 NumFixed = 2; 509 ResultType = Context.BoolTy; 510 break; 511 case Builtin::BI__sync_lock_test_and_set: BuiltinIndex = 12; break; 512 case Builtin::BI__sync_lock_release: 513 BuiltinIndex = 13; 514 NumFixed = 0; 515 ResultType = Context.VoidTy; 516 break; 517 } 518 519 // Now that we know how many fixed arguments we expect, first check that we 520 // have at least that many. 521 if (TheCall->getNumArgs() < 1+NumFixed) { 522 Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args_at_least) 523 << 0 << 1+NumFixed << TheCall->getNumArgs() 524 << TheCall->getCallee()->getSourceRange(); 525 return ExprError(); 526 } 527 528 // Get the decl for the concrete builtin from this, we can tell what the 529 // concrete integer type we should convert to is. 530 unsigned NewBuiltinID = BuiltinIndices[BuiltinIndex][SizeIndex]; 531 const char *NewBuiltinName = Context.BuiltinInfo.GetName(NewBuiltinID); 532 IdentifierInfo *NewBuiltinII = PP.getIdentifierInfo(NewBuiltinName); 533 FunctionDecl *NewBuiltinDecl = 534 cast<FunctionDecl>(LazilyCreateBuiltin(NewBuiltinII, NewBuiltinID, 535 TUScope, false, DRE->getLocStart())); 536 537 // The first argument --- the pointer --- has a fixed type; we 538 // deduce the types of the rest of the arguments accordingly. Walk 539 // the remaining arguments, converting them to the deduced value type. 540 for (unsigned i = 0; i != NumFixed; ++i) { 541 Expr *Arg = TheCall->getArg(i+1); 542 543 // If the argument is an implicit cast, then there was a promotion due to 544 // "...", just remove it now. 545 if (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(Arg)) { 546 Arg = ICE->getSubExpr(); 547 ICE->setSubExpr(0); 548 TheCall->setArg(i+1, Arg); 549 } 550 551 // GCC does an implicit conversion to the pointer or integer ValType. This 552 // can fail in some cases (1i -> int**), check for this error case now. 553 CastKind Kind = CK_Unknown; 554 CXXCastPath BasePath; 555 if (CheckCastTypes(Arg->getSourceRange(), ValType, Arg, Kind, BasePath)) 556 return ExprError(); 557 558 // Okay, we have something that *can* be converted to the right type. Check 559 // to see if there is a potentially weird extension going on here. This can 560 // happen when you do an atomic operation on something like an char* and 561 // pass in 42. The 42 gets converted to char. This is even more strange 562 // for things like 45.123 -> char, etc. 563 // FIXME: Do this check. 564 ImpCastExprToType(Arg, ValType, Kind, VK_RValue, &BasePath); 565 TheCall->setArg(i+1, Arg); 566 } 567 568 // Switch the DeclRefExpr to refer to the new decl. 569 DRE->setDecl(NewBuiltinDecl); 570 DRE->setType(NewBuiltinDecl->getType()); 571 572 // Set the callee in the CallExpr. 573 // FIXME: This leaks the original parens and implicit casts. 574 Expr *PromotedCall = DRE; 575 UsualUnaryConversions(PromotedCall); 576 TheCall->setCallee(PromotedCall); 577 578 // Change the result type of the call to match the original value type. This 579 // is arbitrary, but the codegen for these builtins ins design to handle it 580 // gracefully. 581 TheCall->setType(ResultType); 582 583 return move(TheCallResult); 584} 585 586 587/// CheckObjCString - Checks that the argument to the builtin 588/// CFString constructor is correct 589/// Note: It might also make sense to do the UTF-16 conversion here (would 590/// simplify the backend). 591bool Sema::CheckObjCString(Expr *Arg) { 592 Arg = Arg->IgnoreParenCasts(); 593 StringLiteral *Literal = dyn_cast<StringLiteral>(Arg); 594 595 if (!Literal || Literal->isWide()) { 596 Diag(Arg->getLocStart(), diag::err_cfstring_literal_not_string_constant) 597 << Arg->getSourceRange(); 598 return true; 599 } 600 601 size_t NulPos = Literal->getString().find('\0'); 602 if (NulPos != llvm::StringRef::npos) { 603 Diag(getLocationOfStringLiteralByte(Literal, NulPos), 604 diag::warn_cfstring_literal_contains_nul_character) 605 << Arg->getSourceRange(); 606 } 607 if (Literal->containsNonAsciiOrNull()) { 608 llvm::StringRef String = Literal->getString(); 609 unsigned NumBytes = String.size(); 610 llvm::SmallVector<UTF16, 128> ToBuf(NumBytes); 611 const UTF8 *FromPtr = (UTF8 *)String.data(); 612 UTF16 *ToPtr = &ToBuf[0]; 613 614 ConversionResult Result = ConvertUTF8toUTF16(&FromPtr, FromPtr + NumBytes, 615 &ToPtr, ToPtr + NumBytes, 616 strictConversion); 617 // Check for conversion failure. 618 if (Result != conversionOK) 619 Diag(Arg->getLocStart(), 620 diag::warn_cfstring_truncated) << Arg->getSourceRange(); 621 } 622 return false; 623} 624 625/// SemaBuiltinVAStart - Check the arguments to __builtin_va_start for validity. 626/// Emit an error and return true on failure, return false on success. 627bool Sema::SemaBuiltinVAStart(CallExpr *TheCall) { 628 Expr *Fn = TheCall->getCallee(); 629 if (TheCall->getNumArgs() > 2) { 630 Diag(TheCall->getArg(2)->getLocStart(), 631 diag::err_typecheck_call_too_many_args) 632 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 633 << Fn->getSourceRange() 634 << SourceRange(TheCall->getArg(2)->getLocStart(), 635 (*(TheCall->arg_end()-1))->getLocEnd()); 636 return true; 637 } 638 639 if (TheCall->getNumArgs() < 2) { 640 return Diag(TheCall->getLocEnd(), 641 diag::err_typecheck_call_too_few_args_at_least) 642 << 0 /*function call*/ << 2 << TheCall->getNumArgs(); 643 } 644 645 // Determine whether the current function is variadic or not. 646 BlockScopeInfo *CurBlock = getCurBlock(); 647 bool isVariadic; 648 if (CurBlock) 649 isVariadic = CurBlock->TheDecl->isVariadic(); 650 else if (FunctionDecl *FD = getCurFunctionDecl()) 651 isVariadic = FD->isVariadic(); 652 else 653 isVariadic = getCurMethodDecl()->isVariadic(); 654 655 if (!isVariadic) { 656 Diag(Fn->getLocStart(), diag::err_va_start_used_in_non_variadic_function); 657 return true; 658 } 659 660 // Verify that the second argument to the builtin is the last argument of the 661 // current function or method. 662 bool SecondArgIsLastNamedArgument = false; 663 const Expr *Arg = TheCall->getArg(1)->IgnoreParenCasts(); 664 665 if (const DeclRefExpr *DR = dyn_cast<DeclRefExpr>(Arg)) { 666 if (const ParmVarDecl *PV = dyn_cast<ParmVarDecl>(DR->getDecl())) { 667 // FIXME: This isn't correct for methods (results in bogus warning). 668 // Get the last formal in the current function. 669 const ParmVarDecl *LastArg; 670 if (CurBlock) 671 LastArg = *(CurBlock->TheDecl->param_end()-1); 672 else if (FunctionDecl *FD = getCurFunctionDecl()) 673 LastArg = *(FD->param_end()-1); 674 else 675 LastArg = *(getCurMethodDecl()->param_end()-1); 676 SecondArgIsLastNamedArgument = PV == LastArg; 677 } 678 } 679 680 if (!SecondArgIsLastNamedArgument) 681 Diag(TheCall->getArg(1)->getLocStart(), 682 diag::warn_second_parameter_of_va_start_not_last_named_argument); 683 return false; 684} 685 686/// SemaBuiltinUnorderedCompare - Handle functions like __builtin_isgreater and 687/// friends. This is declared to take (...), so we have to check everything. 688bool Sema::SemaBuiltinUnorderedCompare(CallExpr *TheCall) { 689 if (TheCall->getNumArgs() < 2) 690 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 691 << 0 << 2 << TheCall->getNumArgs()/*function call*/; 692 if (TheCall->getNumArgs() > 2) 693 return Diag(TheCall->getArg(2)->getLocStart(), 694 diag::err_typecheck_call_too_many_args) 695 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 696 << SourceRange(TheCall->getArg(2)->getLocStart(), 697 (*(TheCall->arg_end()-1))->getLocEnd()); 698 699 Expr *OrigArg0 = TheCall->getArg(0); 700 Expr *OrigArg1 = TheCall->getArg(1); 701 702 // Do standard promotions between the two arguments, returning their common 703 // type. 704 QualType Res = UsualArithmeticConversions(OrigArg0, OrigArg1, false); 705 706 // Make sure any conversions are pushed back into the call; this is 707 // type safe since unordered compare builtins are declared as "_Bool 708 // foo(...)". 709 TheCall->setArg(0, OrigArg0); 710 TheCall->setArg(1, OrigArg1); 711 712 if (OrigArg0->isTypeDependent() || OrigArg1->isTypeDependent()) 713 return false; 714 715 // If the common type isn't a real floating type, then the arguments were 716 // invalid for this operation. 717 if (!Res->isRealFloatingType()) 718 return Diag(OrigArg0->getLocStart(), 719 diag::err_typecheck_call_invalid_ordered_compare) 720 << OrigArg0->getType() << OrigArg1->getType() 721 << SourceRange(OrigArg0->getLocStart(), OrigArg1->getLocEnd()); 722 723 return false; 724} 725 726/// SemaBuiltinSemaBuiltinFPClassification - Handle functions like 727/// __builtin_isnan and friends. This is declared to take (...), so we have 728/// to check everything. We expect the last argument to be a floating point 729/// value. 730bool Sema::SemaBuiltinFPClassification(CallExpr *TheCall, unsigned NumArgs) { 731 if (TheCall->getNumArgs() < NumArgs) 732 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 733 << 0 << NumArgs << TheCall->getNumArgs()/*function call*/; 734 if (TheCall->getNumArgs() > NumArgs) 735 return Diag(TheCall->getArg(NumArgs)->getLocStart(), 736 diag::err_typecheck_call_too_many_args) 737 << 0 /*function call*/ << NumArgs << TheCall->getNumArgs() 738 << SourceRange(TheCall->getArg(NumArgs)->getLocStart(), 739 (*(TheCall->arg_end()-1))->getLocEnd()); 740 741 Expr *OrigArg = TheCall->getArg(NumArgs-1); 742 743 if (OrigArg->isTypeDependent()) 744 return false; 745 746 // This operation requires a non-_Complex floating-point number. 747 if (!OrigArg->getType()->isRealFloatingType()) 748 return Diag(OrigArg->getLocStart(), 749 diag::err_typecheck_call_invalid_unary_fp) 750 << OrigArg->getType() << OrigArg->getSourceRange(); 751 752 // If this is an implicit conversion from float -> double, remove it. 753 if (ImplicitCastExpr *Cast = dyn_cast<ImplicitCastExpr>(OrigArg)) { 754 Expr *CastArg = Cast->getSubExpr(); 755 if (CastArg->getType()->isSpecificBuiltinType(BuiltinType::Float)) { 756 assert(Cast->getType()->isSpecificBuiltinType(BuiltinType::Double) && 757 "promotion from float to double is the only expected cast here"); 758 Cast->setSubExpr(0); 759 TheCall->setArg(NumArgs-1, CastArg); 760 OrigArg = CastArg; 761 } 762 } 763 764 return false; 765} 766 767/// SemaBuiltinShuffleVector - Handle __builtin_shufflevector. 768// This is declared to take (...), so we have to check everything. 769ExprResult Sema::SemaBuiltinShuffleVector(CallExpr *TheCall) { 770 if (TheCall->getNumArgs() < 2) 771 return ExprError(Diag(TheCall->getLocEnd(), 772 diag::err_typecheck_call_too_few_args_at_least) 773 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 774 << TheCall->getSourceRange()); 775 776 // Determine which of the following types of shufflevector we're checking: 777 // 1) unary, vector mask: (lhs, mask) 778 // 2) binary, vector mask: (lhs, rhs, mask) 779 // 3) binary, scalar mask: (lhs, rhs, index, ..., index) 780 QualType resType = TheCall->getArg(0)->getType(); 781 unsigned numElements = 0; 782 783 if (!TheCall->getArg(0)->isTypeDependent() && 784 !TheCall->getArg(1)->isTypeDependent()) { 785 QualType LHSType = TheCall->getArg(0)->getType(); 786 QualType RHSType = TheCall->getArg(1)->getType(); 787 788 if (!LHSType->isVectorType() || !RHSType->isVectorType()) { 789 Diag(TheCall->getLocStart(), diag::err_shufflevector_non_vector) 790 << SourceRange(TheCall->getArg(0)->getLocStart(), 791 TheCall->getArg(1)->getLocEnd()); 792 return ExprError(); 793 } 794 795 numElements = LHSType->getAs<VectorType>()->getNumElements(); 796 unsigned numResElements = TheCall->getNumArgs() - 2; 797 798 // Check to see if we have a call with 2 vector arguments, the unary shuffle 799 // with mask. If so, verify that RHS is an integer vector type with the 800 // same number of elts as lhs. 801 if (TheCall->getNumArgs() == 2) { 802 if (!RHSType->hasIntegerRepresentation() || 803 RHSType->getAs<VectorType>()->getNumElements() != numElements) 804 Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector) 805 << SourceRange(TheCall->getArg(1)->getLocStart(), 806 TheCall->getArg(1)->getLocEnd()); 807 numResElements = numElements; 808 } 809 else if (!Context.hasSameUnqualifiedType(LHSType, RHSType)) { 810 Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector) 811 << SourceRange(TheCall->getArg(0)->getLocStart(), 812 TheCall->getArg(1)->getLocEnd()); 813 return ExprError(); 814 } else if (numElements != numResElements) { 815 QualType eltType = LHSType->getAs<VectorType>()->getElementType(); 816 resType = Context.getVectorType(eltType, numResElements, 817 VectorType::NotAltiVec); 818 } 819 } 820 821 for (unsigned i = 2; i < TheCall->getNumArgs(); i++) { 822 if (TheCall->getArg(i)->isTypeDependent() || 823 TheCall->getArg(i)->isValueDependent()) 824 continue; 825 826 llvm::APSInt Result(32); 827 if (!TheCall->getArg(i)->isIntegerConstantExpr(Result, Context)) 828 return ExprError(Diag(TheCall->getLocStart(), 829 diag::err_shufflevector_nonconstant_argument) 830 << TheCall->getArg(i)->getSourceRange()); 831 832 if (Result.getActiveBits() > 64 || Result.getZExtValue() >= numElements*2) 833 return ExprError(Diag(TheCall->getLocStart(), 834 diag::err_shufflevector_argument_too_large) 835 << TheCall->getArg(i)->getSourceRange()); 836 } 837 838 llvm::SmallVector<Expr*, 32> exprs; 839 840 for (unsigned i = 0, e = TheCall->getNumArgs(); i != e; i++) { 841 exprs.push_back(TheCall->getArg(i)); 842 TheCall->setArg(i, 0); 843 } 844 845 return Owned(new (Context) ShuffleVectorExpr(Context, exprs.begin(), 846 exprs.size(), resType, 847 TheCall->getCallee()->getLocStart(), 848 TheCall->getRParenLoc())); 849} 850 851/// SemaBuiltinPrefetch - Handle __builtin_prefetch. 852// This is declared to take (const void*, ...) and can take two 853// optional constant int args. 854bool Sema::SemaBuiltinPrefetch(CallExpr *TheCall) { 855 unsigned NumArgs = TheCall->getNumArgs(); 856 857 if (NumArgs > 3) 858 return Diag(TheCall->getLocEnd(), 859 diag::err_typecheck_call_too_many_args_at_most) 860 << 0 /*function call*/ << 3 << NumArgs 861 << TheCall->getSourceRange(); 862 863 // Argument 0 is checked for us and the remaining arguments must be 864 // constant integers. 865 for (unsigned i = 1; i != NumArgs; ++i) { 866 Expr *Arg = TheCall->getArg(i); 867 868 llvm::APSInt Result; 869 if (SemaBuiltinConstantArg(TheCall, i, Result)) 870 return true; 871 872 // FIXME: gcc issues a warning and rewrites these to 0. These 873 // seems especially odd for the third argument since the default 874 // is 3. 875 if (i == 1) { 876 if (Result.getLimitedValue() > 1) 877 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 878 << "0" << "1" << Arg->getSourceRange(); 879 } else { 880 if (Result.getLimitedValue() > 3) 881 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 882 << "0" << "3" << Arg->getSourceRange(); 883 } 884 } 885 886 return false; 887} 888 889/// SemaBuiltinConstantArg - Handle a check if argument ArgNum of CallExpr 890/// TheCall is a constant expression. 891bool Sema::SemaBuiltinConstantArg(CallExpr *TheCall, int ArgNum, 892 llvm::APSInt &Result) { 893 Expr *Arg = TheCall->getArg(ArgNum); 894 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 895 FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl()); 896 897 if (Arg->isTypeDependent() || Arg->isValueDependent()) return false; 898 899 if (!Arg->isIntegerConstantExpr(Result, Context)) 900 return Diag(TheCall->getLocStart(), diag::err_constant_integer_arg_type) 901 << FDecl->getDeclName() << Arg->getSourceRange(); 902 903 return false; 904} 905 906/// SemaBuiltinObjectSize - Handle __builtin_object_size(void *ptr, 907/// int type). This simply type checks that type is one of the defined 908/// constants (0-3). 909// For compatability check 0-3, llvm only handles 0 and 2. 910bool Sema::SemaBuiltinObjectSize(CallExpr *TheCall) { 911 llvm::APSInt Result; 912 913 // Check constant-ness first. 914 if (SemaBuiltinConstantArg(TheCall, 1, Result)) 915 return true; 916 917 Expr *Arg = TheCall->getArg(1); 918 if (Result.getSExtValue() < 0 || Result.getSExtValue() > 3) { 919 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 920 << "0" << "3" << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 921 } 922 923 return false; 924} 925 926/// SemaBuiltinLongjmp - Handle __builtin_longjmp(void *env[5], int val). 927/// This checks that val is a constant 1. 928bool Sema::SemaBuiltinLongjmp(CallExpr *TheCall) { 929 Expr *Arg = TheCall->getArg(1); 930 llvm::APSInt Result; 931 932 // TODO: This is less than ideal. Overload this to take a value. 933 if (SemaBuiltinConstantArg(TheCall, 1, Result)) 934 return true; 935 936 if (Result != 1) 937 return Diag(TheCall->getLocStart(), diag::err_builtin_longjmp_invalid_val) 938 << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 939 940 return false; 941} 942 943// Handle i > 1 ? "x" : "y", recursivelly 944bool Sema::SemaCheckStringLiteral(const Expr *E, const CallExpr *TheCall, 945 bool HasVAListArg, 946 unsigned format_idx, unsigned firstDataArg, 947 bool isPrintf) { 948 tryAgain: 949 if (E->isTypeDependent() || E->isValueDependent()) 950 return false; 951 952 switch (E->getStmtClass()) { 953 case Stmt::ConditionalOperatorClass: { 954 const ConditionalOperator *C = cast<ConditionalOperator>(E); 955 return SemaCheckStringLiteral(C->getTrueExpr(), TheCall, HasVAListArg, 956 format_idx, firstDataArg, isPrintf) 957 && SemaCheckStringLiteral(C->getRHS(), TheCall, HasVAListArg, 958 format_idx, firstDataArg, isPrintf); 959 } 960 961 case Stmt::IntegerLiteralClass: 962 // Technically -Wformat-nonliteral does not warn about this case. 963 // The behavior of printf and friends in this case is implementation 964 // dependent. Ideally if the format string cannot be null then 965 // it should have a 'nonnull' attribute in the function prototype. 966 return true; 967 968 case Stmt::ImplicitCastExprClass: { 969 E = cast<ImplicitCastExpr>(E)->getSubExpr(); 970 goto tryAgain; 971 } 972 973 case Stmt::ParenExprClass: { 974 E = cast<ParenExpr>(E)->getSubExpr(); 975 goto tryAgain; 976 } 977 978 case Stmt::DeclRefExprClass: { 979 const DeclRefExpr *DR = cast<DeclRefExpr>(E); 980 981 // As an exception, do not flag errors for variables binding to 982 // const string literals. 983 if (const VarDecl *VD = dyn_cast<VarDecl>(DR->getDecl())) { 984 bool isConstant = false; 985 QualType T = DR->getType(); 986 987 if (const ArrayType *AT = Context.getAsArrayType(T)) { 988 isConstant = AT->getElementType().isConstant(Context); 989 } else if (const PointerType *PT = T->getAs<PointerType>()) { 990 isConstant = T.isConstant(Context) && 991 PT->getPointeeType().isConstant(Context); 992 } 993 994 if (isConstant) { 995 if (const Expr *Init = VD->getAnyInitializer()) 996 return SemaCheckStringLiteral(Init, TheCall, 997 HasVAListArg, format_idx, firstDataArg, 998 isPrintf); 999 } 1000 1001 // For vprintf* functions (i.e., HasVAListArg==true), we add a 1002 // special check to see if the format string is a function parameter 1003 // of the function calling the printf function. If the function 1004 // has an attribute indicating it is a printf-like function, then we 1005 // should suppress warnings concerning non-literals being used in a call 1006 // to a vprintf function. For example: 1007 // 1008 // void 1009 // logmessage(char const *fmt __attribute__ (format (printf, 1, 2)), ...){ 1010 // va_list ap; 1011 // va_start(ap, fmt); 1012 // vprintf(fmt, ap); // Do NOT emit a warning about "fmt". 1013 // ... 1014 // 1015 // 1016 // FIXME: We don't have full attribute support yet, so just check to see 1017 // if the argument is a DeclRefExpr that references a parameter. We'll 1018 // add proper support for checking the attribute later. 1019 if (HasVAListArg) 1020 if (isa<ParmVarDecl>(VD)) 1021 return true; 1022 } 1023 1024 return false; 1025 } 1026 1027 case Stmt::CallExprClass: { 1028 const CallExpr *CE = cast<CallExpr>(E); 1029 if (const ImplicitCastExpr *ICE 1030 = dyn_cast<ImplicitCastExpr>(CE->getCallee())) { 1031 if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(ICE->getSubExpr())) { 1032 if (const FunctionDecl *FD = dyn_cast<FunctionDecl>(DRE->getDecl())) { 1033 if (const FormatArgAttr *FA = FD->getAttr<FormatArgAttr>()) { 1034 unsigned ArgIndex = FA->getFormatIdx(); 1035 const Expr *Arg = CE->getArg(ArgIndex - 1); 1036 1037 return SemaCheckStringLiteral(Arg, TheCall, HasVAListArg, 1038 format_idx, firstDataArg, isPrintf); 1039 } 1040 } 1041 } 1042 } 1043 1044 return false; 1045 } 1046 case Stmt::ObjCStringLiteralClass: 1047 case Stmt::StringLiteralClass: { 1048 const StringLiteral *StrE = NULL; 1049 1050 if (const ObjCStringLiteral *ObjCFExpr = dyn_cast<ObjCStringLiteral>(E)) 1051 StrE = ObjCFExpr->getString(); 1052 else 1053 StrE = cast<StringLiteral>(E); 1054 1055 if (StrE) { 1056 CheckFormatString(StrE, E, TheCall, HasVAListArg, format_idx, 1057 firstDataArg, isPrintf); 1058 return true; 1059 } 1060 1061 return false; 1062 } 1063 1064 default: 1065 return false; 1066 } 1067} 1068 1069void 1070Sema::CheckNonNullArguments(const NonNullAttr *NonNull, 1071 const CallExpr *TheCall) { 1072 for (NonNullAttr::args_iterator i = NonNull->args_begin(), 1073 e = NonNull->args_end(); 1074 i != e; ++i) { 1075 const Expr *ArgExpr = TheCall->getArg(*i); 1076 if (ArgExpr->isNullPointerConstant(Context, 1077 Expr::NPC_ValueDependentIsNotNull)) 1078 Diag(TheCall->getCallee()->getLocStart(), diag::warn_null_arg) 1079 << ArgExpr->getSourceRange(); 1080 } 1081} 1082 1083/// CheckPrintfScanfArguments - Check calls to printf and scanf (and similar 1084/// functions) for correct use of format strings. 1085void 1086Sema::CheckPrintfScanfArguments(const CallExpr *TheCall, bool HasVAListArg, 1087 unsigned format_idx, unsigned firstDataArg, 1088 bool isPrintf) { 1089 1090 const Expr *Fn = TheCall->getCallee(); 1091 1092 // The way the format attribute works in GCC, the implicit this argument 1093 // of member functions is counted. However, it doesn't appear in our own 1094 // lists, so decrement format_idx in that case. 1095 if (isa<CXXMemberCallExpr>(TheCall)) { 1096 // Catch a format attribute mistakenly referring to the object argument. 1097 if (format_idx == 0) 1098 return; 1099 --format_idx; 1100 if(firstDataArg != 0) 1101 --firstDataArg; 1102 } 1103 1104 // CHECK: printf/scanf-like function is called with no format string. 1105 if (format_idx >= TheCall->getNumArgs()) { 1106 Diag(TheCall->getRParenLoc(), diag::warn_missing_format_string) 1107 << Fn->getSourceRange(); 1108 return; 1109 } 1110 1111 const Expr *OrigFormatExpr = TheCall->getArg(format_idx)->IgnoreParenCasts(); 1112 1113 // CHECK: format string is not a string literal. 1114 // 1115 // Dynamically generated format strings are difficult to 1116 // automatically vet at compile time. Requiring that format strings 1117 // are string literals: (1) permits the checking of format strings by 1118 // the compiler and thereby (2) can practically remove the source of 1119 // many format string exploits. 1120 1121 // Format string can be either ObjC string (e.g. @"%d") or 1122 // C string (e.g. "%d") 1123 // ObjC string uses the same format specifiers as C string, so we can use 1124 // the same format string checking logic for both ObjC and C strings. 1125 if (SemaCheckStringLiteral(OrigFormatExpr, TheCall, HasVAListArg, format_idx, 1126 firstDataArg, isPrintf)) 1127 return; // Literal format string found, check done! 1128 1129 // If there are no arguments specified, warn with -Wformat-security, otherwise 1130 // warn only with -Wformat-nonliteral. 1131 if (TheCall->getNumArgs() == format_idx+1) 1132 Diag(TheCall->getArg(format_idx)->getLocStart(), 1133 diag::warn_format_nonliteral_noargs) 1134 << OrigFormatExpr->getSourceRange(); 1135 else 1136 Diag(TheCall->getArg(format_idx)->getLocStart(), 1137 diag::warn_format_nonliteral) 1138 << OrigFormatExpr->getSourceRange(); 1139} 1140 1141namespace { 1142class CheckFormatHandler : public analyze_format_string::FormatStringHandler { 1143protected: 1144 Sema &S; 1145 const StringLiteral *FExpr; 1146 const Expr *OrigFormatExpr; 1147 const unsigned FirstDataArg; 1148 const unsigned NumDataArgs; 1149 const bool IsObjCLiteral; 1150 const char *Beg; // Start of format string. 1151 const bool HasVAListArg; 1152 const CallExpr *TheCall; 1153 unsigned FormatIdx; 1154 llvm::BitVector CoveredArgs; 1155 bool usesPositionalArgs; 1156 bool atFirstArg; 1157public: 1158 CheckFormatHandler(Sema &s, const StringLiteral *fexpr, 1159 const Expr *origFormatExpr, unsigned firstDataArg, 1160 unsigned numDataArgs, bool isObjCLiteral, 1161 const char *beg, bool hasVAListArg, 1162 const CallExpr *theCall, unsigned formatIdx) 1163 : S(s), FExpr(fexpr), OrigFormatExpr(origFormatExpr), 1164 FirstDataArg(firstDataArg), 1165 NumDataArgs(numDataArgs), 1166 IsObjCLiteral(isObjCLiteral), Beg(beg), 1167 HasVAListArg(hasVAListArg), 1168 TheCall(theCall), FormatIdx(formatIdx), 1169 usesPositionalArgs(false), atFirstArg(true) { 1170 CoveredArgs.resize(numDataArgs); 1171 CoveredArgs.reset(); 1172 } 1173 1174 void DoneProcessing(); 1175 1176 void HandleIncompleteSpecifier(const char *startSpecifier, 1177 unsigned specifierLen); 1178 1179 virtual void HandleInvalidPosition(const char *startSpecifier, 1180 unsigned specifierLen, 1181 analyze_format_string::PositionContext p); 1182 1183 virtual void HandleZeroPosition(const char *startPos, unsigned posLen); 1184 1185 void HandleNullChar(const char *nullCharacter); 1186 1187protected: 1188 bool HandleInvalidConversionSpecifier(unsigned argIndex, SourceLocation Loc, 1189 const char *startSpec, 1190 unsigned specifierLen, 1191 const char *csStart, unsigned csLen); 1192 1193 SourceRange getFormatStringRange(); 1194 CharSourceRange getSpecifierRange(const char *startSpecifier, 1195 unsigned specifierLen); 1196 SourceLocation getLocationOfByte(const char *x); 1197 1198 const Expr *getDataArg(unsigned i) const; 1199 1200 bool CheckNumArgs(const analyze_format_string::FormatSpecifier &FS, 1201 const analyze_format_string::ConversionSpecifier &CS, 1202 const char *startSpecifier, unsigned specifierLen, 1203 unsigned argIndex); 1204}; 1205} 1206 1207SourceRange CheckFormatHandler::getFormatStringRange() { 1208 return OrigFormatExpr->getSourceRange(); 1209} 1210 1211CharSourceRange CheckFormatHandler:: 1212getSpecifierRange(const char *startSpecifier, unsigned specifierLen) { 1213 SourceLocation Start = getLocationOfByte(startSpecifier); 1214 SourceLocation End = getLocationOfByte(startSpecifier + specifierLen - 1); 1215 1216 // Advance the end SourceLocation by one due to half-open ranges. 1217 End = End.getFileLocWithOffset(1); 1218 1219 return CharSourceRange::getCharRange(Start, End); 1220} 1221 1222SourceLocation CheckFormatHandler::getLocationOfByte(const char *x) { 1223 return S.getLocationOfStringLiteralByte(FExpr, x - Beg); 1224} 1225 1226void CheckFormatHandler::HandleIncompleteSpecifier(const char *startSpecifier, 1227 unsigned specifierLen){ 1228 SourceLocation Loc = getLocationOfByte(startSpecifier); 1229 S.Diag(Loc, diag::warn_printf_incomplete_specifier) 1230 << getSpecifierRange(startSpecifier, specifierLen); 1231} 1232 1233void 1234CheckFormatHandler::HandleInvalidPosition(const char *startPos, unsigned posLen, 1235 analyze_format_string::PositionContext p) { 1236 SourceLocation Loc = getLocationOfByte(startPos); 1237 S.Diag(Loc, diag::warn_format_invalid_positional_specifier) 1238 << (unsigned) p << getSpecifierRange(startPos, posLen); 1239} 1240 1241void CheckFormatHandler::HandleZeroPosition(const char *startPos, 1242 unsigned posLen) { 1243 SourceLocation Loc = getLocationOfByte(startPos); 1244 S.Diag(Loc, diag::warn_format_zero_positional_specifier) 1245 << getSpecifierRange(startPos, posLen); 1246} 1247 1248void CheckFormatHandler::HandleNullChar(const char *nullCharacter) { 1249 // The presence of a null character is likely an error. 1250 S.Diag(getLocationOfByte(nullCharacter), 1251 diag::warn_printf_format_string_contains_null_char) 1252 << getFormatStringRange(); 1253} 1254 1255const Expr *CheckFormatHandler::getDataArg(unsigned i) const { 1256 return TheCall->getArg(FirstDataArg + i); 1257} 1258 1259void CheckFormatHandler::DoneProcessing() { 1260 // Does the number of data arguments exceed the number of 1261 // format conversions in the format string? 1262 if (!HasVAListArg) { 1263 // Find any arguments that weren't covered. 1264 CoveredArgs.flip(); 1265 signed notCoveredArg = CoveredArgs.find_first(); 1266 if (notCoveredArg >= 0) { 1267 assert((unsigned)notCoveredArg < NumDataArgs); 1268 S.Diag(getDataArg((unsigned) notCoveredArg)->getLocStart(), 1269 diag::warn_printf_data_arg_not_used) 1270 << getFormatStringRange(); 1271 } 1272 } 1273} 1274 1275bool 1276CheckFormatHandler::HandleInvalidConversionSpecifier(unsigned argIndex, 1277 SourceLocation Loc, 1278 const char *startSpec, 1279 unsigned specifierLen, 1280 const char *csStart, 1281 unsigned csLen) { 1282 1283 bool keepGoing = true; 1284 if (argIndex < NumDataArgs) { 1285 // Consider the argument coverered, even though the specifier doesn't 1286 // make sense. 1287 CoveredArgs.set(argIndex); 1288 } 1289 else { 1290 // If argIndex exceeds the number of data arguments we 1291 // don't issue a warning because that is just a cascade of warnings (and 1292 // they may have intended '%%' anyway). We don't want to continue processing 1293 // the format string after this point, however, as we will like just get 1294 // gibberish when trying to match arguments. 1295 keepGoing = false; 1296 } 1297 1298 S.Diag(Loc, diag::warn_format_invalid_conversion) 1299 << llvm::StringRef(csStart, csLen) 1300 << getSpecifierRange(startSpec, specifierLen); 1301 1302 return keepGoing; 1303} 1304 1305bool 1306CheckFormatHandler::CheckNumArgs( 1307 const analyze_format_string::FormatSpecifier &FS, 1308 const analyze_format_string::ConversionSpecifier &CS, 1309 const char *startSpecifier, unsigned specifierLen, unsigned argIndex) { 1310 1311 if (argIndex >= NumDataArgs) { 1312 if (FS.usesPositionalArg()) { 1313 S.Diag(getLocationOfByte(CS.getStart()), 1314 diag::warn_printf_positional_arg_exceeds_data_args) 1315 << (argIndex+1) << NumDataArgs 1316 << getSpecifierRange(startSpecifier, specifierLen); 1317 } 1318 else { 1319 S.Diag(getLocationOfByte(CS.getStart()), 1320 diag::warn_printf_insufficient_data_args) 1321 << getSpecifierRange(startSpecifier, specifierLen); 1322 } 1323 1324 return false; 1325 } 1326 return true; 1327} 1328 1329//===--- CHECK: Printf format string checking ------------------------------===// 1330 1331namespace { 1332class CheckPrintfHandler : public CheckFormatHandler { 1333public: 1334 CheckPrintfHandler(Sema &s, const StringLiteral *fexpr, 1335 const Expr *origFormatExpr, unsigned firstDataArg, 1336 unsigned numDataArgs, bool isObjCLiteral, 1337 const char *beg, bool hasVAListArg, 1338 const CallExpr *theCall, unsigned formatIdx) 1339 : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg, 1340 numDataArgs, isObjCLiteral, beg, hasVAListArg, 1341 theCall, formatIdx) {} 1342 1343 1344 bool HandleInvalidPrintfConversionSpecifier( 1345 const analyze_printf::PrintfSpecifier &FS, 1346 const char *startSpecifier, 1347 unsigned specifierLen); 1348 1349 bool HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier &FS, 1350 const char *startSpecifier, 1351 unsigned specifierLen); 1352 1353 bool HandleAmount(const analyze_format_string::OptionalAmount &Amt, unsigned k, 1354 const char *startSpecifier, unsigned specifierLen); 1355 void HandleInvalidAmount(const analyze_printf::PrintfSpecifier &FS, 1356 const analyze_printf::OptionalAmount &Amt, 1357 unsigned type, 1358 const char *startSpecifier, unsigned specifierLen); 1359 void HandleFlag(const analyze_printf::PrintfSpecifier &FS, 1360 const analyze_printf::OptionalFlag &flag, 1361 const char *startSpecifier, unsigned specifierLen); 1362 void HandleIgnoredFlag(const analyze_printf::PrintfSpecifier &FS, 1363 const analyze_printf::OptionalFlag &ignoredFlag, 1364 const analyze_printf::OptionalFlag &flag, 1365 const char *startSpecifier, unsigned specifierLen); 1366}; 1367} 1368 1369bool CheckPrintfHandler::HandleInvalidPrintfConversionSpecifier( 1370 const analyze_printf::PrintfSpecifier &FS, 1371 const char *startSpecifier, 1372 unsigned specifierLen) { 1373 const analyze_printf::PrintfConversionSpecifier &CS = 1374 FS.getConversionSpecifier(); 1375 1376 return HandleInvalidConversionSpecifier(FS.getArgIndex(), 1377 getLocationOfByte(CS.getStart()), 1378 startSpecifier, specifierLen, 1379 CS.getStart(), CS.getLength()); 1380} 1381 1382bool CheckPrintfHandler::HandleAmount( 1383 const analyze_format_string::OptionalAmount &Amt, 1384 unsigned k, const char *startSpecifier, 1385 unsigned specifierLen) { 1386 1387 if (Amt.hasDataArgument()) { 1388 if (!HasVAListArg) { 1389 unsigned argIndex = Amt.getArgIndex(); 1390 if (argIndex >= NumDataArgs) { 1391 S.Diag(getLocationOfByte(Amt.getStart()), 1392 diag::warn_printf_asterisk_missing_arg) 1393 << k << getSpecifierRange(startSpecifier, specifierLen); 1394 // Don't do any more checking. We will just emit 1395 // spurious errors. 1396 return false; 1397 } 1398 1399 // Type check the data argument. It should be an 'int'. 1400 // Although not in conformance with C99, we also allow the argument to be 1401 // an 'unsigned int' as that is a reasonably safe case. GCC also 1402 // doesn't emit a warning for that case. 1403 CoveredArgs.set(argIndex); 1404 const Expr *Arg = getDataArg(argIndex); 1405 QualType T = Arg->getType(); 1406 1407 const analyze_printf::ArgTypeResult &ATR = Amt.getArgType(S.Context); 1408 assert(ATR.isValid()); 1409 1410 if (!ATR.matchesType(S.Context, T)) { 1411 S.Diag(getLocationOfByte(Amt.getStart()), 1412 diag::warn_printf_asterisk_wrong_type) 1413 << k 1414 << ATR.getRepresentativeType(S.Context) << T 1415 << getSpecifierRange(startSpecifier, specifierLen) 1416 << Arg->getSourceRange(); 1417 // Don't do any more checking. We will just emit 1418 // spurious errors. 1419 return false; 1420 } 1421 } 1422 } 1423 return true; 1424} 1425 1426void CheckPrintfHandler::HandleInvalidAmount( 1427 const analyze_printf::PrintfSpecifier &FS, 1428 const analyze_printf::OptionalAmount &Amt, 1429 unsigned type, 1430 const char *startSpecifier, 1431 unsigned specifierLen) { 1432 const analyze_printf::PrintfConversionSpecifier &CS = 1433 FS.getConversionSpecifier(); 1434 switch (Amt.getHowSpecified()) { 1435 case analyze_printf::OptionalAmount::Constant: 1436 S.Diag(getLocationOfByte(Amt.getStart()), 1437 diag::warn_printf_nonsensical_optional_amount) 1438 << type 1439 << CS.toString() 1440 << getSpecifierRange(startSpecifier, specifierLen) 1441 << FixItHint::CreateRemoval(getSpecifierRange(Amt.getStart(), 1442 Amt.getConstantLength())); 1443 break; 1444 1445 default: 1446 S.Diag(getLocationOfByte(Amt.getStart()), 1447 diag::warn_printf_nonsensical_optional_amount) 1448 << type 1449 << CS.toString() 1450 << getSpecifierRange(startSpecifier, specifierLen); 1451 break; 1452 } 1453} 1454 1455void CheckPrintfHandler::HandleFlag(const analyze_printf::PrintfSpecifier &FS, 1456 const analyze_printf::OptionalFlag &flag, 1457 const char *startSpecifier, 1458 unsigned specifierLen) { 1459 // Warn about pointless flag with a fixit removal. 1460 const analyze_printf::PrintfConversionSpecifier &CS = 1461 FS.getConversionSpecifier(); 1462 S.Diag(getLocationOfByte(flag.getPosition()), 1463 diag::warn_printf_nonsensical_flag) 1464 << flag.toString() << CS.toString() 1465 << getSpecifierRange(startSpecifier, specifierLen) 1466 << FixItHint::CreateRemoval(getSpecifierRange(flag.getPosition(), 1)); 1467} 1468 1469void CheckPrintfHandler::HandleIgnoredFlag( 1470 const analyze_printf::PrintfSpecifier &FS, 1471 const analyze_printf::OptionalFlag &ignoredFlag, 1472 const analyze_printf::OptionalFlag &flag, 1473 const char *startSpecifier, 1474 unsigned specifierLen) { 1475 // Warn about ignored flag with a fixit removal. 1476 S.Diag(getLocationOfByte(ignoredFlag.getPosition()), 1477 diag::warn_printf_ignored_flag) 1478 << ignoredFlag.toString() << flag.toString() 1479 << getSpecifierRange(startSpecifier, specifierLen) 1480 << FixItHint::CreateRemoval(getSpecifierRange( 1481 ignoredFlag.getPosition(), 1)); 1482} 1483 1484bool 1485CheckPrintfHandler::HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier 1486 &FS, 1487 const char *startSpecifier, 1488 unsigned specifierLen) { 1489 1490 using namespace analyze_format_string; 1491 using namespace analyze_printf; 1492 const PrintfConversionSpecifier &CS = FS.getConversionSpecifier(); 1493 1494 if (FS.consumesDataArgument()) { 1495 if (atFirstArg) { 1496 atFirstArg = false; 1497 usesPositionalArgs = FS.usesPositionalArg(); 1498 } 1499 else if (usesPositionalArgs != FS.usesPositionalArg()) { 1500 // Cannot mix-and-match positional and non-positional arguments. 1501 S.Diag(getLocationOfByte(CS.getStart()), 1502 diag::warn_format_mix_positional_nonpositional_args) 1503 << getSpecifierRange(startSpecifier, specifierLen); 1504 return false; 1505 } 1506 } 1507 1508 // First check if the field width, precision, and conversion specifier 1509 // have matching data arguments. 1510 if (!HandleAmount(FS.getFieldWidth(), /* field width */ 0, 1511 startSpecifier, specifierLen)) { 1512 return false; 1513 } 1514 1515 if (!HandleAmount(FS.getPrecision(), /* precision */ 1, 1516 startSpecifier, specifierLen)) { 1517 return false; 1518 } 1519 1520 if (!CS.consumesDataArgument()) { 1521 // FIXME: Technically specifying a precision or field width here 1522 // makes no sense. Worth issuing a warning at some point. 1523 return true; 1524 } 1525 1526 // Consume the argument. 1527 unsigned argIndex = FS.getArgIndex(); 1528 if (argIndex < NumDataArgs) { 1529 // The check to see if the argIndex is valid will come later. 1530 // We set the bit here because we may exit early from this 1531 // function if we encounter some other error. 1532 CoveredArgs.set(argIndex); 1533 } 1534 1535 // Check for using an Objective-C specific conversion specifier 1536 // in a non-ObjC literal. 1537 if (!IsObjCLiteral && CS.isObjCArg()) { 1538 return HandleInvalidPrintfConversionSpecifier(FS, startSpecifier, 1539 specifierLen); 1540 } 1541 1542 // Check for invalid use of field width 1543 if (!FS.hasValidFieldWidth()) { 1544 HandleInvalidAmount(FS, FS.getFieldWidth(), /* field width */ 0, 1545 startSpecifier, specifierLen); 1546 } 1547 1548 // Check for invalid use of precision 1549 if (!FS.hasValidPrecision()) { 1550 HandleInvalidAmount(FS, FS.getPrecision(), /* precision */ 1, 1551 startSpecifier, specifierLen); 1552 } 1553 1554 // Check each flag does not conflict with any other component. 1555 if (!FS.hasValidLeadingZeros()) 1556 HandleFlag(FS, FS.hasLeadingZeros(), startSpecifier, specifierLen); 1557 if (!FS.hasValidPlusPrefix()) 1558 HandleFlag(FS, FS.hasPlusPrefix(), startSpecifier, specifierLen); 1559 if (!FS.hasValidSpacePrefix()) 1560 HandleFlag(FS, FS.hasSpacePrefix(), startSpecifier, specifierLen); 1561 if (!FS.hasValidAlternativeForm()) 1562 HandleFlag(FS, FS.hasAlternativeForm(), startSpecifier, specifierLen); 1563 if (!FS.hasValidLeftJustified()) 1564 HandleFlag(FS, FS.isLeftJustified(), startSpecifier, specifierLen); 1565 1566 // Check that flags are not ignored by another flag 1567 if (FS.hasSpacePrefix() && FS.hasPlusPrefix()) // ' ' ignored by '+' 1568 HandleIgnoredFlag(FS, FS.hasSpacePrefix(), FS.hasPlusPrefix(), 1569 startSpecifier, specifierLen); 1570 if (FS.hasLeadingZeros() && FS.isLeftJustified()) // '0' ignored by '-' 1571 HandleIgnoredFlag(FS, FS.hasLeadingZeros(), FS.isLeftJustified(), 1572 startSpecifier, specifierLen); 1573 1574 // Check the length modifier is valid with the given conversion specifier. 1575 const LengthModifier &LM = FS.getLengthModifier(); 1576 if (!FS.hasValidLengthModifier()) 1577 S.Diag(getLocationOfByte(LM.getStart()), 1578 diag::warn_format_nonsensical_length) 1579 << LM.toString() << CS.toString() 1580 << getSpecifierRange(startSpecifier, specifierLen) 1581 << FixItHint::CreateRemoval(getSpecifierRange(LM.getStart(), 1582 LM.getLength())); 1583 1584 // Are we using '%n'? 1585 if (CS.getKind() == ConversionSpecifier::nArg) { 1586 // Issue a warning about this being a possible security issue. 1587 S.Diag(getLocationOfByte(CS.getStart()), diag::warn_printf_write_back) 1588 << getSpecifierRange(startSpecifier, specifierLen); 1589 // Continue checking the other format specifiers. 1590 return true; 1591 } 1592 1593 // The remaining checks depend on the data arguments. 1594 if (HasVAListArg) 1595 return true; 1596 1597 if (!CheckNumArgs(FS, CS, startSpecifier, specifierLen, argIndex)) 1598 return false; 1599 1600 // Now type check the data expression that matches the 1601 // format specifier. 1602 const Expr *Ex = getDataArg(argIndex); 1603 const analyze_printf::ArgTypeResult &ATR = FS.getArgType(S.Context); 1604 if (ATR.isValid() && !ATR.matchesType(S.Context, Ex->getType())) { 1605 // Check if we didn't match because of an implicit cast from a 'char' 1606 // or 'short' to an 'int'. This is done because printf is a varargs 1607 // function. 1608 if (const ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(Ex)) 1609 if (ICE->getType() == S.Context.IntTy) 1610 if (ATR.matchesType(S.Context, ICE->getSubExpr()->getType())) 1611 return true; 1612 1613 // We may be able to offer a FixItHint if it is a supported type. 1614 PrintfSpecifier fixedFS = FS; 1615 bool success = fixedFS.fixType(Ex->getType()); 1616 1617 if (success) { 1618 // Get the fix string from the fixed format specifier 1619 llvm::SmallString<128> buf; 1620 llvm::raw_svector_ostream os(buf); 1621 fixedFS.toString(os); 1622 1623 // FIXME: getRepresentativeType() perhaps should return a string 1624 // instead of a QualType to better handle when the representative 1625 // type is 'wint_t' (which is defined in the system headers). 1626 S.Diag(getLocationOfByte(CS.getStart()), 1627 diag::warn_printf_conversion_argument_type_mismatch) 1628 << ATR.getRepresentativeType(S.Context) << Ex->getType() 1629 << getSpecifierRange(startSpecifier, specifierLen) 1630 << Ex->getSourceRange() 1631 << FixItHint::CreateReplacement( 1632 getSpecifierRange(startSpecifier, specifierLen), 1633 os.str()); 1634 } 1635 else { 1636 S.Diag(getLocationOfByte(CS.getStart()), 1637 diag::warn_printf_conversion_argument_type_mismatch) 1638 << ATR.getRepresentativeType(S.Context) << Ex->getType() 1639 << getSpecifierRange(startSpecifier, specifierLen) 1640 << Ex->getSourceRange(); 1641 } 1642 } 1643 1644 return true; 1645} 1646 1647//===--- CHECK: Scanf format string checking ------------------------------===// 1648 1649namespace { 1650class CheckScanfHandler : public CheckFormatHandler { 1651public: 1652 CheckScanfHandler(Sema &s, const StringLiteral *fexpr, 1653 const Expr *origFormatExpr, unsigned firstDataArg, 1654 unsigned numDataArgs, bool isObjCLiteral, 1655 const char *beg, bool hasVAListArg, 1656 const CallExpr *theCall, unsigned formatIdx) 1657 : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg, 1658 numDataArgs, isObjCLiteral, beg, hasVAListArg, 1659 theCall, formatIdx) {} 1660 1661 bool HandleScanfSpecifier(const analyze_scanf::ScanfSpecifier &FS, 1662 const char *startSpecifier, 1663 unsigned specifierLen); 1664 1665 bool HandleInvalidScanfConversionSpecifier( 1666 const analyze_scanf::ScanfSpecifier &FS, 1667 const char *startSpecifier, 1668 unsigned specifierLen); 1669 1670 void HandleIncompleteScanList(const char *start, const char *end); 1671}; 1672} 1673 1674void CheckScanfHandler::HandleIncompleteScanList(const char *start, 1675 const char *end) { 1676 S.Diag(getLocationOfByte(end), diag::warn_scanf_scanlist_incomplete) 1677 << getSpecifierRange(start, end - start); 1678} 1679 1680bool CheckScanfHandler::HandleInvalidScanfConversionSpecifier( 1681 const analyze_scanf::ScanfSpecifier &FS, 1682 const char *startSpecifier, 1683 unsigned specifierLen) { 1684 1685 const analyze_scanf::ScanfConversionSpecifier &CS = 1686 FS.getConversionSpecifier(); 1687 1688 return HandleInvalidConversionSpecifier(FS.getArgIndex(), 1689 getLocationOfByte(CS.getStart()), 1690 startSpecifier, specifierLen, 1691 CS.getStart(), CS.getLength()); 1692} 1693 1694bool CheckScanfHandler::HandleScanfSpecifier( 1695 const analyze_scanf::ScanfSpecifier &FS, 1696 const char *startSpecifier, 1697 unsigned specifierLen) { 1698 1699 using namespace analyze_scanf; 1700 using namespace analyze_format_string; 1701 1702 const ScanfConversionSpecifier &CS = FS.getConversionSpecifier(); 1703 1704 // Handle case where '%' and '*' don't consume an argument. These shouldn't 1705 // be used to decide if we are using positional arguments consistently. 1706 if (FS.consumesDataArgument()) { 1707 if (atFirstArg) { 1708 atFirstArg = false; 1709 usesPositionalArgs = FS.usesPositionalArg(); 1710 } 1711 else if (usesPositionalArgs != FS.usesPositionalArg()) { 1712 // Cannot mix-and-match positional and non-positional arguments. 1713 S.Diag(getLocationOfByte(CS.getStart()), 1714 diag::warn_format_mix_positional_nonpositional_args) 1715 << getSpecifierRange(startSpecifier, specifierLen); 1716 return false; 1717 } 1718 } 1719 1720 // Check if the field with is non-zero. 1721 const OptionalAmount &Amt = FS.getFieldWidth(); 1722 if (Amt.getHowSpecified() == OptionalAmount::Constant) { 1723 if (Amt.getConstantAmount() == 0) { 1724 const CharSourceRange &R = getSpecifierRange(Amt.getStart(), 1725 Amt.getConstantLength()); 1726 S.Diag(getLocationOfByte(Amt.getStart()), 1727 diag::warn_scanf_nonzero_width) 1728 << R << FixItHint::CreateRemoval(R); 1729 } 1730 } 1731 1732 if (!FS.consumesDataArgument()) { 1733 // FIXME: Technically specifying a precision or field width here 1734 // makes no sense. Worth issuing a warning at some point. 1735 return true; 1736 } 1737 1738 // Consume the argument. 1739 unsigned argIndex = FS.getArgIndex(); 1740 if (argIndex < NumDataArgs) { 1741 // The check to see if the argIndex is valid will come later. 1742 // We set the bit here because we may exit early from this 1743 // function if we encounter some other error. 1744 CoveredArgs.set(argIndex); 1745 } 1746 1747 // Check the length modifier is valid with the given conversion specifier. 1748 const LengthModifier &LM = FS.getLengthModifier(); 1749 if (!FS.hasValidLengthModifier()) { 1750 S.Diag(getLocationOfByte(LM.getStart()), 1751 diag::warn_format_nonsensical_length) 1752 << LM.toString() << CS.toString() 1753 << getSpecifierRange(startSpecifier, specifierLen) 1754 << FixItHint::CreateRemoval(getSpecifierRange(LM.getStart(), 1755 LM.getLength())); 1756 } 1757 1758 // The remaining checks depend on the data arguments. 1759 if (HasVAListArg) 1760 return true; 1761 1762 if (!CheckNumArgs(FS, CS, startSpecifier, specifierLen, argIndex)) 1763 return false; 1764 1765 // FIXME: Check that the argument type matches the format specifier. 1766 1767 return true; 1768} 1769 1770void Sema::CheckFormatString(const StringLiteral *FExpr, 1771 const Expr *OrigFormatExpr, 1772 const CallExpr *TheCall, bool HasVAListArg, 1773 unsigned format_idx, unsigned firstDataArg, 1774 bool isPrintf) { 1775 1776 // CHECK: is the format string a wide literal? 1777 if (FExpr->isWide()) { 1778 Diag(FExpr->getLocStart(), 1779 diag::warn_format_string_is_wide_literal) 1780 << OrigFormatExpr->getSourceRange(); 1781 return; 1782 } 1783 1784 // Str - The format string. NOTE: this is NOT null-terminated! 1785 llvm::StringRef StrRef = FExpr->getString(); 1786 const char *Str = StrRef.data(); 1787 unsigned StrLen = StrRef.size(); 1788 1789 // CHECK: empty format string? 1790 if (StrLen == 0) { 1791 Diag(FExpr->getLocStart(), diag::warn_empty_format_string) 1792 << OrigFormatExpr->getSourceRange(); 1793 return; 1794 } 1795 1796 if (isPrintf) { 1797 CheckPrintfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg, 1798 TheCall->getNumArgs() - firstDataArg, 1799 isa<ObjCStringLiteral>(OrigFormatExpr), Str, 1800 HasVAListArg, TheCall, format_idx); 1801 1802 if (!analyze_format_string::ParsePrintfString(H, Str, Str + StrLen)) 1803 H.DoneProcessing(); 1804 } 1805 else { 1806 CheckScanfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg, 1807 TheCall->getNumArgs() - firstDataArg, 1808 isa<ObjCStringLiteral>(OrigFormatExpr), Str, 1809 HasVAListArg, TheCall, format_idx); 1810 1811 if (!analyze_format_string::ParseScanfString(H, Str, Str + StrLen)) 1812 H.DoneProcessing(); 1813 } 1814} 1815 1816//===--- CHECK: Return Address of Stack Variable --------------------------===// 1817 1818static DeclRefExpr* EvalVal(Expr *E); 1819static DeclRefExpr* EvalAddr(Expr* E); 1820 1821/// CheckReturnStackAddr - Check if a return statement returns the address 1822/// of a stack variable. 1823void 1824Sema::CheckReturnStackAddr(Expr *RetValExp, QualType lhsType, 1825 SourceLocation ReturnLoc) { 1826 1827 // Perform checking for returned stack addresses. 1828 if (lhsType->isPointerType() || lhsType->isBlockPointerType()) { 1829 if (DeclRefExpr *DR = EvalAddr(RetValExp)) 1830 Diag(DR->getLocStart(), diag::warn_ret_stack_addr) 1831 << DR->getDecl()->getDeclName() << RetValExp->getSourceRange(); 1832 1833 // Skip over implicit cast expressions when checking for block expressions. 1834 RetValExp = RetValExp->IgnoreParenCasts(); 1835 1836 if (BlockExpr *C = dyn_cast<BlockExpr>(RetValExp)) 1837 if (C->hasBlockDeclRefExprs()) 1838 Diag(C->getLocStart(), diag::err_ret_local_block) 1839 << C->getSourceRange(); 1840 1841 if (AddrLabelExpr *ALE = dyn_cast<AddrLabelExpr>(RetValExp)) 1842 Diag(ALE->getLocStart(), diag::warn_ret_addr_label) 1843 << ALE->getSourceRange(); 1844 1845 } else if (lhsType->isReferenceType()) { 1846 // Perform checking for stack values returned by reference. 1847 // Check for a reference to the stack 1848 if (DeclRefExpr *DR = EvalVal(RetValExp)) 1849 Diag(DR->getLocStart(), diag::warn_ret_stack_ref) 1850 << DR->getDecl()->getDeclName() << RetValExp->getSourceRange(); 1851 } 1852} 1853 1854/// EvalAddr - EvalAddr and EvalVal are mutually recursive functions that 1855/// check if the expression in a return statement evaluates to an address 1856/// to a location on the stack. The recursion is used to traverse the 1857/// AST of the return expression, with recursion backtracking when we 1858/// encounter a subexpression that (1) clearly does not lead to the address 1859/// of a stack variable or (2) is something we cannot determine leads to 1860/// the address of a stack variable based on such local checking. 1861/// 1862/// EvalAddr processes expressions that are pointers that are used as 1863/// references (and not L-values). EvalVal handles all other values. 1864/// At the base case of the recursion is a check for a DeclRefExpr* in 1865/// the refers to a stack variable. 1866/// 1867/// This implementation handles: 1868/// 1869/// * pointer-to-pointer casts 1870/// * implicit conversions from array references to pointers 1871/// * taking the address of fields 1872/// * arbitrary interplay between "&" and "*" operators 1873/// * pointer arithmetic from an address of a stack variable 1874/// * taking the address of an array element where the array is on the stack 1875static DeclRefExpr* EvalAddr(Expr *E) { 1876 // We should only be called for evaluating pointer expressions. 1877 assert((E->getType()->isAnyPointerType() || 1878 E->getType()->isBlockPointerType() || 1879 E->getType()->isObjCQualifiedIdType()) && 1880 "EvalAddr only works on pointers"); 1881 1882 // Our "symbolic interpreter" is just a dispatch off the currently 1883 // viewed AST node. We then recursively traverse the AST by calling 1884 // EvalAddr and EvalVal appropriately. 1885 switch (E->getStmtClass()) { 1886 case Stmt::ParenExprClass: 1887 // Ignore parentheses. 1888 return EvalAddr(cast<ParenExpr>(E)->getSubExpr()); 1889 1890 case Stmt::UnaryOperatorClass: { 1891 // The only unary operator that make sense to handle here 1892 // is AddrOf. All others don't make sense as pointers. 1893 UnaryOperator *U = cast<UnaryOperator>(E); 1894 1895 if (U->getOpcode() == UO_AddrOf) 1896 return EvalVal(U->getSubExpr()); 1897 else 1898 return NULL; 1899 } 1900 1901 case Stmt::BinaryOperatorClass: { 1902 // Handle pointer arithmetic. All other binary operators are not valid 1903 // in this context. 1904 BinaryOperator *B = cast<BinaryOperator>(E); 1905 BinaryOperatorKind op = B->getOpcode(); 1906 1907 if (op != BO_Add && op != BO_Sub) 1908 return NULL; 1909 1910 Expr *Base = B->getLHS(); 1911 1912 // Determine which argument is the real pointer base. It could be 1913 // the RHS argument instead of the LHS. 1914 if (!Base->getType()->isPointerType()) Base = B->getRHS(); 1915 1916 assert (Base->getType()->isPointerType()); 1917 return EvalAddr(Base); 1918 } 1919 1920 // For conditional operators we need to see if either the LHS or RHS are 1921 // valid DeclRefExpr*s. If one of them is valid, we return it. 1922 case Stmt::ConditionalOperatorClass: { 1923 ConditionalOperator *C = cast<ConditionalOperator>(E); 1924 1925 // Handle the GNU extension for missing LHS. 1926 if (Expr *lhsExpr = C->getLHS()) 1927 if (DeclRefExpr* LHS = EvalAddr(lhsExpr)) 1928 return LHS; 1929 1930 return EvalAddr(C->getRHS()); 1931 } 1932 1933 // For casts, we need to handle conversions from arrays to 1934 // pointer values, and pointer-to-pointer conversions. 1935 case Stmt::ImplicitCastExprClass: 1936 case Stmt::CStyleCastExprClass: 1937 case Stmt::CXXFunctionalCastExprClass: { 1938 Expr* SubExpr = cast<CastExpr>(E)->getSubExpr(); 1939 QualType T = SubExpr->getType(); 1940 1941 if (SubExpr->getType()->isPointerType() || 1942 SubExpr->getType()->isBlockPointerType() || 1943 SubExpr->getType()->isObjCQualifiedIdType()) 1944 return EvalAddr(SubExpr); 1945 else if (T->isArrayType()) 1946 return EvalVal(SubExpr); 1947 else 1948 return 0; 1949 } 1950 1951 // C++ casts. For dynamic casts, static casts, and const casts, we 1952 // are always converting from a pointer-to-pointer, so we just blow 1953 // through the cast. In the case the dynamic cast doesn't fail (and 1954 // return NULL), we take the conservative route and report cases 1955 // where we return the address of a stack variable. For Reinterpre 1956 // FIXME: The comment about is wrong; we're not always converting 1957 // from pointer to pointer. I'm guessing that this code should also 1958 // handle references to objects. 1959 case Stmt::CXXStaticCastExprClass: 1960 case Stmt::CXXDynamicCastExprClass: 1961 case Stmt::CXXConstCastExprClass: 1962 case Stmt::CXXReinterpretCastExprClass: { 1963 Expr *S = cast<CXXNamedCastExpr>(E)->getSubExpr(); 1964 if (S->getType()->isPointerType() || S->getType()->isBlockPointerType()) 1965 return EvalAddr(S); 1966 else 1967 return NULL; 1968 } 1969 1970 // Everything else: we simply don't reason about them. 1971 default: 1972 return NULL; 1973 } 1974} 1975 1976 1977/// EvalVal - This function is complements EvalAddr in the mutual recursion. 1978/// See the comments for EvalAddr for more details. 1979static DeclRefExpr* EvalVal(Expr *E) { 1980do { 1981 // We should only be called for evaluating non-pointer expressions, or 1982 // expressions with a pointer type that are not used as references but instead 1983 // are l-values (e.g., DeclRefExpr with a pointer type). 1984 1985 // Our "symbolic interpreter" is just a dispatch off the currently 1986 // viewed AST node. We then recursively traverse the AST by calling 1987 // EvalAddr and EvalVal appropriately. 1988 switch (E->getStmtClass()) { 1989 case Stmt::ImplicitCastExprClass: { 1990 ImplicitCastExpr *IE = cast<ImplicitCastExpr>(E); 1991 if (IE->getValueKind() == VK_LValue) { 1992 E = IE->getSubExpr(); 1993 continue; 1994 } 1995 return NULL; 1996 } 1997 1998 case Stmt::DeclRefExprClass: { 1999 // DeclRefExpr: the base case. When we hit a DeclRefExpr we are looking 2000 // at code that refers to a variable's name. We check if it has local 2001 // storage within the function, and if so, return the expression. 2002 DeclRefExpr *DR = cast<DeclRefExpr>(E); 2003 2004 if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl())) 2005 if (V->hasLocalStorage() && !V->getType()->isReferenceType()) return DR; 2006 2007 return NULL; 2008 } 2009 2010 case Stmt::ParenExprClass: { 2011 // Ignore parentheses. 2012 E = cast<ParenExpr>(E)->getSubExpr(); 2013 continue; 2014 } 2015 2016 case Stmt::UnaryOperatorClass: { 2017 // The only unary operator that make sense to handle here 2018 // is Deref. All others don't resolve to a "name." This includes 2019 // handling all sorts of rvalues passed to a unary operator. 2020 UnaryOperator *U = cast<UnaryOperator>(E); 2021 2022 if (U->getOpcode() == UO_Deref) 2023 return EvalAddr(U->getSubExpr()); 2024 2025 return NULL; 2026 } 2027 2028 case Stmt::ArraySubscriptExprClass: { 2029 // Array subscripts are potential references to data on the stack. We 2030 // retrieve the DeclRefExpr* for the array variable if it indeed 2031 // has local storage. 2032 return EvalAddr(cast<ArraySubscriptExpr>(E)->getBase()); 2033 } 2034 2035 case Stmt::ConditionalOperatorClass: { 2036 // For conditional operators we need to see if either the LHS or RHS are 2037 // non-NULL DeclRefExpr's. If one is non-NULL, we return it. 2038 ConditionalOperator *C = cast<ConditionalOperator>(E); 2039 2040 // Handle the GNU extension for missing LHS. 2041 if (Expr *lhsExpr = C->getLHS()) 2042 if (DeclRefExpr *LHS = EvalVal(lhsExpr)) 2043 return LHS; 2044 2045 return EvalVal(C->getRHS()); 2046 } 2047 2048 // Accesses to members are potential references to data on the stack. 2049 case Stmt::MemberExprClass: { 2050 MemberExpr *M = cast<MemberExpr>(E); 2051 2052 // Check for indirect access. We only want direct field accesses. 2053 if (M->isArrow()) 2054 return NULL; 2055 2056 // Check whether the member type is itself a reference, in which case 2057 // we're not going to refer to the member, but to what the member refers to. 2058 if (M->getMemberDecl()->getType()->isReferenceType()) 2059 return NULL; 2060 2061 return EvalVal(M->getBase()); 2062 } 2063 2064 // Everything else: we simply don't reason about them. 2065 default: 2066 return NULL; 2067 } 2068} while (true); 2069} 2070 2071//===--- CHECK: Floating-Point comparisons (-Wfloat-equal) ---------------===// 2072 2073/// Check for comparisons of floating point operands using != and ==. 2074/// Issue a warning if these are no self-comparisons, as they are not likely 2075/// to do what the programmer intended. 2076void Sema::CheckFloatComparison(SourceLocation loc, Expr* lex, Expr *rex) { 2077 bool EmitWarning = true; 2078 2079 Expr* LeftExprSansParen = lex->IgnoreParens(); 2080 Expr* RightExprSansParen = rex->IgnoreParens(); 2081 2082 // Special case: check for x == x (which is OK). 2083 // Do not emit warnings for such cases. 2084 if (DeclRefExpr* DRL = dyn_cast<DeclRefExpr>(LeftExprSansParen)) 2085 if (DeclRefExpr* DRR = dyn_cast<DeclRefExpr>(RightExprSansParen)) 2086 if (DRL->getDecl() == DRR->getDecl()) 2087 EmitWarning = false; 2088 2089 2090 // Special case: check for comparisons against literals that can be exactly 2091 // represented by APFloat. In such cases, do not emit a warning. This 2092 // is a heuristic: often comparison against such literals are used to 2093 // detect if a value in a variable has not changed. This clearly can 2094 // lead to false negatives. 2095 if (EmitWarning) { 2096 if (FloatingLiteral* FLL = dyn_cast<FloatingLiteral>(LeftExprSansParen)) { 2097 if (FLL->isExact()) 2098 EmitWarning = false; 2099 } else 2100 if (FloatingLiteral* FLR = dyn_cast<FloatingLiteral>(RightExprSansParen)){ 2101 if (FLR->isExact()) 2102 EmitWarning = false; 2103 } 2104 } 2105 2106 // Check for comparisons with builtin types. 2107 if (EmitWarning) 2108 if (CallExpr* CL = dyn_cast<CallExpr>(LeftExprSansParen)) 2109 if (CL->isBuiltinCall(Context)) 2110 EmitWarning = false; 2111 2112 if (EmitWarning) 2113 if (CallExpr* CR = dyn_cast<CallExpr>(RightExprSansParen)) 2114 if (CR->isBuiltinCall(Context)) 2115 EmitWarning = false; 2116 2117 // Emit the diagnostic. 2118 if (EmitWarning) 2119 Diag(loc, diag::warn_floatingpoint_eq) 2120 << lex->getSourceRange() << rex->getSourceRange(); 2121} 2122 2123//===--- CHECK: Integer mixed-sign comparisons (-Wsign-compare) --------===// 2124//===--- CHECK: Lossy implicit conversions (-Wconversion) --------------===// 2125 2126namespace { 2127 2128/// Structure recording the 'active' range of an integer-valued 2129/// expression. 2130struct IntRange { 2131 /// The number of bits active in the int. 2132 unsigned Width; 2133 2134 /// True if the int is known not to have negative values. 2135 bool NonNegative; 2136 2137 IntRange(unsigned Width, bool NonNegative) 2138 : Width(Width), NonNegative(NonNegative) 2139 {} 2140 2141 // Returns the range of the bool type. 2142 static IntRange forBoolType() { 2143 return IntRange(1, true); 2144 } 2145 2146 // Returns the range of an integral type. 2147 static IntRange forType(ASTContext &C, QualType T) { 2148 return forCanonicalType(C, T->getCanonicalTypeInternal().getTypePtr()); 2149 } 2150 2151 // Returns the range of an integeral type based on its canonical 2152 // representation. 2153 static IntRange forCanonicalType(ASTContext &C, const Type *T) { 2154 assert(T->isCanonicalUnqualified()); 2155 2156 if (const VectorType *VT = dyn_cast<VectorType>(T)) 2157 T = VT->getElementType().getTypePtr(); 2158 if (const ComplexType *CT = dyn_cast<ComplexType>(T)) 2159 T = CT->getElementType().getTypePtr(); 2160 2161 if (const EnumType *ET = dyn_cast<EnumType>(T)) { 2162 EnumDecl *Enum = ET->getDecl(); 2163 unsigned NumPositive = Enum->getNumPositiveBits(); 2164 unsigned NumNegative = Enum->getNumNegativeBits(); 2165 2166 return IntRange(std::max(NumPositive, NumNegative), NumNegative == 0); 2167 } 2168 2169 const BuiltinType *BT = cast<BuiltinType>(T); 2170 assert(BT->isInteger()); 2171 2172 return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger()); 2173 } 2174 2175 // Returns the supremum of two ranges: i.e. their conservative merge. 2176 static IntRange join(IntRange L, IntRange R) { 2177 return IntRange(std::max(L.Width, R.Width), 2178 L.NonNegative && R.NonNegative); 2179 } 2180 2181 // Returns the infinum of two ranges: i.e. their aggressive merge. 2182 static IntRange meet(IntRange L, IntRange R) { 2183 return IntRange(std::min(L.Width, R.Width), 2184 L.NonNegative || R.NonNegative); 2185 } 2186}; 2187 2188IntRange GetValueRange(ASTContext &C, llvm::APSInt &value, unsigned MaxWidth) { 2189 if (value.isSigned() && value.isNegative()) 2190 return IntRange(value.getMinSignedBits(), false); 2191 2192 if (value.getBitWidth() > MaxWidth) 2193 value.trunc(MaxWidth); 2194 2195 // isNonNegative() just checks the sign bit without considering 2196 // signedness. 2197 return IntRange(value.getActiveBits(), true); 2198} 2199 2200IntRange GetValueRange(ASTContext &C, APValue &result, QualType Ty, 2201 unsigned MaxWidth) { 2202 if (result.isInt()) 2203 return GetValueRange(C, result.getInt(), MaxWidth); 2204 2205 if (result.isVector()) { 2206 IntRange R = GetValueRange(C, result.getVectorElt(0), Ty, MaxWidth); 2207 for (unsigned i = 1, e = result.getVectorLength(); i != e; ++i) { 2208 IntRange El = GetValueRange(C, result.getVectorElt(i), Ty, MaxWidth); 2209 R = IntRange::join(R, El); 2210 } 2211 return R; 2212 } 2213 2214 if (result.isComplexInt()) { 2215 IntRange R = GetValueRange(C, result.getComplexIntReal(), MaxWidth); 2216 IntRange I = GetValueRange(C, result.getComplexIntImag(), MaxWidth); 2217 return IntRange::join(R, I); 2218 } 2219 2220 // This can happen with lossless casts to intptr_t of "based" lvalues. 2221 // Assume it might use arbitrary bits. 2222 // FIXME: The only reason we need to pass the type in here is to get 2223 // the sign right on this one case. It would be nice if APValue 2224 // preserved this. 2225 assert(result.isLValue()); 2226 return IntRange(MaxWidth, Ty->isUnsignedIntegerType()); 2227} 2228 2229/// Pseudo-evaluate the given integer expression, estimating the 2230/// range of values it might take. 2231/// 2232/// \param MaxWidth - the width to which the value will be truncated 2233IntRange GetExprRange(ASTContext &C, Expr *E, unsigned MaxWidth) { 2234 E = E->IgnoreParens(); 2235 2236 // Try a full evaluation first. 2237 Expr::EvalResult result; 2238 if (E->Evaluate(result, C)) 2239 return GetValueRange(C, result.Val, E->getType(), MaxWidth); 2240 2241 // I think we only want to look through implicit casts here; if the 2242 // user has an explicit widening cast, we should treat the value as 2243 // being of the new, wider type. 2244 if (ImplicitCastExpr *CE = dyn_cast<ImplicitCastExpr>(E)) { 2245 if (CE->getCastKind() == CK_NoOp) 2246 return GetExprRange(C, CE->getSubExpr(), MaxWidth); 2247 2248 IntRange OutputTypeRange = IntRange::forType(C, CE->getType()); 2249 2250 bool isIntegerCast = (CE->getCastKind() == CK_IntegralCast); 2251 if (!isIntegerCast && CE->getCastKind() == CK_Unknown) 2252 isIntegerCast = CE->getSubExpr()->getType()->isIntegerType(); 2253 2254 // Assume that non-integer casts can span the full range of the type. 2255 if (!isIntegerCast) 2256 return OutputTypeRange; 2257 2258 IntRange SubRange 2259 = GetExprRange(C, CE->getSubExpr(), 2260 std::min(MaxWidth, OutputTypeRange.Width)); 2261 2262 // Bail out if the subexpr's range is as wide as the cast type. 2263 if (SubRange.Width >= OutputTypeRange.Width) 2264 return OutputTypeRange; 2265 2266 // Otherwise, we take the smaller width, and we're non-negative if 2267 // either the output type or the subexpr is. 2268 return IntRange(SubRange.Width, 2269 SubRange.NonNegative || OutputTypeRange.NonNegative); 2270 } 2271 2272 if (ConditionalOperator *CO = dyn_cast<ConditionalOperator>(E)) { 2273 // If we can fold the condition, just take that operand. 2274 bool CondResult; 2275 if (CO->getCond()->EvaluateAsBooleanCondition(CondResult, C)) 2276 return GetExprRange(C, CondResult ? CO->getTrueExpr() 2277 : CO->getFalseExpr(), 2278 MaxWidth); 2279 2280 // Otherwise, conservatively merge. 2281 IntRange L = GetExprRange(C, CO->getTrueExpr(), MaxWidth); 2282 IntRange R = GetExprRange(C, CO->getFalseExpr(), MaxWidth); 2283 return IntRange::join(L, R); 2284 } 2285 2286 if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) { 2287 switch (BO->getOpcode()) { 2288 2289 // Boolean-valued operations are single-bit and positive. 2290 case BO_LAnd: 2291 case BO_LOr: 2292 case BO_LT: 2293 case BO_GT: 2294 case BO_LE: 2295 case BO_GE: 2296 case BO_EQ: 2297 case BO_NE: 2298 return IntRange::forBoolType(); 2299 2300 // The type of these compound assignments is the type of the LHS, 2301 // so the RHS is not necessarily an integer. 2302 case BO_MulAssign: 2303 case BO_DivAssign: 2304 case BO_RemAssign: 2305 case BO_AddAssign: 2306 case BO_SubAssign: 2307 return IntRange::forType(C, E->getType()); 2308 2309 // Operations with opaque sources are black-listed. 2310 case BO_PtrMemD: 2311 case BO_PtrMemI: 2312 return IntRange::forType(C, E->getType()); 2313 2314 // Bitwise-and uses the *infinum* of the two source ranges. 2315 case BO_And: 2316 case BO_AndAssign: 2317 return IntRange::meet(GetExprRange(C, BO->getLHS(), MaxWidth), 2318 GetExprRange(C, BO->getRHS(), MaxWidth)); 2319 2320 // Left shift gets black-listed based on a judgement call. 2321 case BO_Shl: 2322 // ...except that we want to treat '1 << (blah)' as logically 2323 // positive. It's an important idiom. 2324 if (IntegerLiteral *I 2325 = dyn_cast<IntegerLiteral>(BO->getLHS()->IgnoreParenCasts())) { 2326 if (I->getValue() == 1) { 2327 IntRange R = IntRange::forType(C, E->getType()); 2328 return IntRange(R.Width, /*NonNegative*/ true); 2329 } 2330 } 2331 // fallthrough 2332 2333 case BO_ShlAssign: 2334 return IntRange::forType(C, E->getType()); 2335 2336 // Right shift by a constant can narrow its left argument. 2337 case BO_Shr: 2338 case BO_ShrAssign: { 2339 IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth); 2340 2341 // If the shift amount is a positive constant, drop the width by 2342 // that much. 2343 llvm::APSInt shift; 2344 if (BO->getRHS()->isIntegerConstantExpr(shift, C) && 2345 shift.isNonNegative()) { 2346 unsigned zext = shift.getZExtValue(); 2347 if (zext >= L.Width) 2348 L.Width = (L.NonNegative ? 0 : 1); 2349 else 2350 L.Width -= zext; 2351 } 2352 2353 return L; 2354 } 2355 2356 // Comma acts as its right operand. 2357 case BO_Comma: 2358 return GetExprRange(C, BO->getRHS(), MaxWidth); 2359 2360 // Black-list pointer subtractions. 2361 case BO_Sub: 2362 if (BO->getLHS()->getType()->isPointerType()) 2363 return IntRange::forType(C, E->getType()); 2364 // fallthrough 2365 2366 default: 2367 break; 2368 } 2369 2370 // Treat every other operator as if it were closed on the 2371 // narrowest type that encompasses both operands. 2372 IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth); 2373 IntRange R = GetExprRange(C, BO->getRHS(), MaxWidth); 2374 return IntRange::join(L, R); 2375 } 2376 2377 if (UnaryOperator *UO = dyn_cast<UnaryOperator>(E)) { 2378 switch (UO->getOpcode()) { 2379 // Boolean-valued operations are white-listed. 2380 case UO_LNot: 2381 return IntRange::forBoolType(); 2382 2383 // Operations with opaque sources are black-listed. 2384 case UO_Deref: 2385 case UO_AddrOf: // should be impossible 2386 return IntRange::forType(C, E->getType()); 2387 2388 default: 2389 return GetExprRange(C, UO->getSubExpr(), MaxWidth); 2390 } 2391 } 2392 2393 if (dyn_cast<OffsetOfExpr>(E)) { 2394 IntRange::forType(C, E->getType()); 2395 } 2396 2397 FieldDecl *BitField = E->getBitField(); 2398 if (BitField) { 2399 llvm::APSInt BitWidthAP = BitField->getBitWidth()->EvaluateAsInt(C); 2400 unsigned BitWidth = BitWidthAP.getZExtValue(); 2401 2402 return IntRange(BitWidth, BitField->getType()->isUnsignedIntegerType()); 2403 } 2404 2405 return IntRange::forType(C, E->getType()); 2406} 2407 2408IntRange GetExprRange(ASTContext &C, Expr *E) { 2409 return GetExprRange(C, E, C.getIntWidth(E->getType())); 2410} 2411 2412/// Checks whether the given value, which currently has the given 2413/// source semantics, has the same value when coerced through the 2414/// target semantics. 2415bool IsSameFloatAfterCast(const llvm::APFloat &value, 2416 const llvm::fltSemantics &Src, 2417 const llvm::fltSemantics &Tgt) { 2418 llvm::APFloat truncated = value; 2419 2420 bool ignored; 2421 truncated.convert(Src, llvm::APFloat::rmNearestTiesToEven, &ignored); 2422 truncated.convert(Tgt, llvm::APFloat::rmNearestTiesToEven, &ignored); 2423 2424 return truncated.bitwiseIsEqual(value); 2425} 2426 2427/// Checks whether the given value, which currently has the given 2428/// source semantics, has the same value when coerced through the 2429/// target semantics. 2430/// 2431/// The value might be a vector of floats (or a complex number). 2432bool IsSameFloatAfterCast(const APValue &value, 2433 const llvm::fltSemantics &Src, 2434 const llvm::fltSemantics &Tgt) { 2435 if (value.isFloat()) 2436 return IsSameFloatAfterCast(value.getFloat(), Src, Tgt); 2437 2438 if (value.isVector()) { 2439 for (unsigned i = 0, e = value.getVectorLength(); i != e; ++i) 2440 if (!IsSameFloatAfterCast(value.getVectorElt(i), Src, Tgt)) 2441 return false; 2442 return true; 2443 } 2444 2445 assert(value.isComplexFloat()); 2446 return (IsSameFloatAfterCast(value.getComplexFloatReal(), Src, Tgt) && 2447 IsSameFloatAfterCast(value.getComplexFloatImag(), Src, Tgt)); 2448} 2449 2450void AnalyzeImplicitConversions(Sema &S, Expr *E); 2451 2452bool IsZero(Sema &S, Expr *E) { 2453 llvm::APSInt Value; 2454 return E->isIntegerConstantExpr(Value, S.Context) && Value == 0; 2455} 2456 2457void CheckTrivialUnsignedComparison(Sema &S, BinaryOperator *E) { 2458 BinaryOperatorKind op = E->getOpcode(); 2459 if (op == BO_LT && IsZero(S, E->getRHS())) { 2460 S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison) 2461 << "< 0" << "false" 2462 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 2463 } else if (op == BO_GE && IsZero(S, E->getRHS())) { 2464 S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison) 2465 << ">= 0" << "true" 2466 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 2467 } else if (op == BO_GT && IsZero(S, E->getLHS())) { 2468 S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison) 2469 << "0 >" << "false" 2470 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 2471 } else if (op == BO_LE && IsZero(S, E->getLHS())) { 2472 S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison) 2473 << "0 <=" << "true" 2474 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 2475 } 2476} 2477 2478/// Analyze the operands of the given comparison. Implements the 2479/// fallback case from AnalyzeComparison. 2480void AnalyzeImpConvsInComparison(Sema &S, BinaryOperator *E) { 2481 AnalyzeImplicitConversions(S, E->getLHS()); 2482 AnalyzeImplicitConversions(S, E->getRHS()); 2483} 2484 2485/// \brief Implements -Wsign-compare. 2486/// 2487/// \param lex the left-hand expression 2488/// \param rex the right-hand expression 2489/// \param OpLoc the location of the joining operator 2490/// \param BinOpc binary opcode or 0 2491void AnalyzeComparison(Sema &S, BinaryOperator *E) { 2492 // The type the comparison is being performed in. 2493 QualType T = E->getLHS()->getType(); 2494 assert(S.Context.hasSameUnqualifiedType(T, E->getRHS()->getType()) 2495 && "comparison with mismatched types"); 2496 2497 // We don't do anything special if this isn't an unsigned integral 2498 // comparison: we're only interested in integral comparisons, and 2499 // signed comparisons only happen in cases we don't care to warn about. 2500 if (!T->hasUnsignedIntegerRepresentation()) 2501 return AnalyzeImpConvsInComparison(S, E); 2502 2503 Expr *lex = E->getLHS()->IgnoreParenImpCasts(); 2504 Expr *rex = E->getRHS()->IgnoreParenImpCasts(); 2505 2506 // Check to see if one of the (unmodified) operands is of different 2507 // signedness. 2508 Expr *signedOperand, *unsignedOperand; 2509 if (lex->getType()->hasSignedIntegerRepresentation()) { 2510 assert(!rex->getType()->hasSignedIntegerRepresentation() && 2511 "unsigned comparison between two signed integer expressions?"); 2512 signedOperand = lex; 2513 unsignedOperand = rex; 2514 } else if (rex->getType()->hasSignedIntegerRepresentation()) { 2515 signedOperand = rex; 2516 unsignedOperand = lex; 2517 } else { 2518 CheckTrivialUnsignedComparison(S, E); 2519 return AnalyzeImpConvsInComparison(S, E); 2520 } 2521 2522 // Otherwise, calculate the effective range of the signed operand. 2523 IntRange signedRange = GetExprRange(S.Context, signedOperand); 2524 2525 // Go ahead and analyze implicit conversions in the operands. Note 2526 // that we skip the implicit conversions on both sides. 2527 AnalyzeImplicitConversions(S, lex); 2528 AnalyzeImplicitConversions(S, rex); 2529 2530 // If the signed range is non-negative, -Wsign-compare won't fire, 2531 // but we should still check for comparisons which are always true 2532 // or false. 2533 if (signedRange.NonNegative) 2534 return CheckTrivialUnsignedComparison(S, E); 2535 2536 // For (in)equality comparisons, if the unsigned operand is a 2537 // constant which cannot collide with a overflowed signed operand, 2538 // then reinterpreting the signed operand as unsigned will not 2539 // change the result of the comparison. 2540 if (E->isEqualityOp()) { 2541 unsigned comparisonWidth = S.Context.getIntWidth(T); 2542 IntRange unsignedRange = GetExprRange(S.Context, unsignedOperand); 2543 2544 // We should never be unable to prove that the unsigned operand is 2545 // non-negative. 2546 assert(unsignedRange.NonNegative && "unsigned range includes negative?"); 2547 2548 if (unsignedRange.Width < comparisonWidth) 2549 return; 2550 } 2551 2552 S.Diag(E->getOperatorLoc(), diag::warn_mixed_sign_comparison) 2553 << lex->getType() << rex->getType() 2554 << lex->getSourceRange() << rex->getSourceRange(); 2555} 2556 2557/// Diagnose an implicit cast; purely a helper for CheckImplicitConversion. 2558void DiagnoseImpCast(Sema &S, Expr *E, QualType T, unsigned diag) { 2559 S.Diag(E->getExprLoc(), diag) << E->getType() << T << E->getSourceRange(); 2560} 2561 2562void CheckImplicitConversion(Sema &S, Expr *E, QualType T, 2563 bool *ICContext = 0) { 2564 if (E->isTypeDependent() || E->isValueDependent()) return; 2565 2566 const Type *Source = S.Context.getCanonicalType(E->getType()).getTypePtr(); 2567 const Type *Target = S.Context.getCanonicalType(T).getTypePtr(); 2568 if (Source == Target) return; 2569 if (Target->isDependentType()) return; 2570 2571 // Never diagnose implicit casts to bool. 2572 if (Target->isSpecificBuiltinType(BuiltinType::Bool)) 2573 return; 2574 2575 // Strip vector types. 2576 if (isa<VectorType>(Source)) { 2577 if (!isa<VectorType>(Target)) 2578 return DiagnoseImpCast(S, E, T, diag::warn_impcast_vector_scalar); 2579 2580 Source = cast<VectorType>(Source)->getElementType().getTypePtr(); 2581 Target = cast<VectorType>(Target)->getElementType().getTypePtr(); 2582 } 2583 2584 // Strip complex types. 2585 if (isa<ComplexType>(Source)) { 2586 if (!isa<ComplexType>(Target)) 2587 return DiagnoseImpCast(S, E, T, diag::warn_impcast_complex_scalar); 2588 2589 Source = cast<ComplexType>(Source)->getElementType().getTypePtr(); 2590 Target = cast<ComplexType>(Target)->getElementType().getTypePtr(); 2591 } 2592 2593 const BuiltinType *SourceBT = dyn_cast<BuiltinType>(Source); 2594 const BuiltinType *TargetBT = dyn_cast<BuiltinType>(Target); 2595 2596 // If the source is floating point... 2597 if (SourceBT && SourceBT->isFloatingPoint()) { 2598 // ...and the target is floating point... 2599 if (TargetBT && TargetBT->isFloatingPoint()) { 2600 // ...then warn if we're dropping FP rank. 2601 2602 // Builtin FP kinds are ordered by increasing FP rank. 2603 if (SourceBT->getKind() > TargetBT->getKind()) { 2604 // Don't warn about float constants that are precisely 2605 // representable in the target type. 2606 Expr::EvalResult result; 2607 if (E->Evaluate(result, S.Context)) { 2608 // Value might be a float, a float vector, or a float complex. 2609 if (IsSameFloatAfterCast(result.Val, 2610 S.Context.getFloatTypeSemantics(QualType(TargetBT, 0)), 2611 S.Context.getFloatTypeSemantics(QualType(SourceBT, 0)))) 2612 return; 2613 } 2614 2615 DiagnoseImpCast(S, E, T, diag::warn_impcast_float_precision); 2616 } 2617 return; 2618 } 2619 2620 // If the target is integral, always warn. 2621 if ((TargetBT && TargetBT->isInteger())) 2622 // TODO: don't warn for integer values? 2623 DiagnoseImpCast(S, E, T, diag::warn_impcast_float_integer); 2624 2625 return; 2626 } 2627 2628 if (!Source->isIntegerType() || !Target->isIntegerType()) 2629 return; 2630 2631 IntRange SourceRange = GetExprRange(S.Context, E); 2632 IntRange TargetRange = IntRange::forCanonicalType(S.Context, Target); 2633 2634 if (SourceRange.Width > TargetRange.Width) { 2635 // People want to build with -Wshorten-64-to-32 and not -Wconversion 2636 // and by god we'll let them. 2637 if (SourceRange.Width == 64 && TargetRange.Width == 32) 2638 return DiagnoseImpCast(S, E, T, diag::warn_impcast_integer_64_32); 2639 return DiagnoseImpCast(S, E, T, diag::warn_impcast_integer_precision); 2640 } 2641 2642 if ((TargetRange.NonNegative && !SourceRange.NonNegative) || 2643 (!TargetRange.NonNegative && SourceRange.NonNegative && 2644 SourceRange.Width == TargetRange.Width)) { 2645 unsigned DiagID = diag::warn_impcast_integer_sign; 2646 2647 // Traditionally, gcc has warned about this under -Wsign-compare. 2648 // We also want to warn about it in -Wconversion. 2649 // So if -Wconversion is off, use a completely identical diagnostic 2650 // in the sign-compare group. 2651 // The conditional-checking code will 2652 if (ICContext) { 2653 DiagID = diag::warn_impcast_integer_sign_conditional; 2654 *ICContext = true; 2655 } 2656 2657 return DiagnoseImpCast(S, E, T, DiagID); 2658 } 2659 2660 return; 2661} 2662 2663void CheckConditionalOperator(Sema &S, ConditionalOperator *E, QualType T); 2664 2665void CheckConditionalOperand(Sema &S, Expr *E, QualType T, 2666 bool &ICContext) { 2667 E = E->IgnoreParenImpCasts(); 2668 2669 if (isa<ConditionalOperator>(E)) 2670 return CheckConditionalOperator(S, cast<ConditionalOperator>(E), T); 2671 2672 AnalyzeImplicitConversions(S, E); 2673 if (E->getType() != T) 2674 return CheckImplicitConversion(S, E, T, &ICContext); 2675 return; 2676} 2677 2678void CheckConditionalOperator(Sema &S, ConditionalOperator *E, QualType T) { 2679 AnalyzeImplicitConversions(S, E->getCond()); 2680 2681 bool Suspicious = false; 2682 CheckConditionalOperand(S, E->getTrueExpr(), T, Suspicious); 2683 CheckConditionalOperand(S, E->getFalseExpr(), T, Suspicious); 2684 2685 // If -Wconversion would have warned about either of the candidates 2686 // for a signedness conversion to the context type... 2687 if (!Suspicious) return; 2688 2689 // ...but it's currently ignored... 2690 if (S.Diags.getDiagnosticLevel(diag::warn_impcast_integer_sign_conditional)) 2691 return; 2692 2693 // ...and -Wsign-compare isn't... 2694 if (!S.Diags.getDiagnosticLevel(diag::warn_mixed_sign_conditional)) 2695 return; 2696 2697 // ...then check whether it would have warned about either of the 2698 // candidates for a signedness conversion to the condition type. 2699 if (E->getType() != T) { 2700 Suspicious = false; 2701 CheckImplicitConversion(S, E->getTrueExpr()->IgnoreParenImpCasts(), 2702 E->getType(), &Suspicious); 2703 if (!Suspicious) 2704 CheckImplicitConversion(S, E->getFalseExpr()->IgnoreParenImpCasts(), 2705 E->getType(), &Suspicious); 2706 if (!Suspicious) 2707 return; 2708 } 2709 2710 // If so, emit a diagnostic under -Wsign-compare. 2711 Expr *lex = E->getTrueExpr()->IgnoreParenImpCasts(); 2712 Expr *rex = E->getFalseExpr()->IgnoreParenImpCasts(); 2713 S.Diag(E->getQuestionLoc(), diag::warn_mixed_sign_conditional) 2714 << lex->getType() << rex->getType() 2715 << lex->getSourceRange() << rex->getSourceRange(); 2716} 2717 2718/// AnalyzeImplicitConversions - Find and report any interesting 2719/// implicit conversions in the given expression. There are a couple 2720/// of competing diagnostics here, -Wconversion and -Wsign-compare. 2721void AnalyzeImplicitConversions(Sema &S, Expr *OrigE) { 2722 QualType T = OrigE->getType(); 2723 Expr *E = OrigE->IgnoreParenImpCasts(); 2724 2725 // For conditional operators, we analyze the arguments as if they 2726 // were being fed directly into the output. 2727 if (isa<ConditionalOperator>(E)) { 2728 ConditionalOperator *CO = cast<ConditionalOperator>(E); 2729 CheckConditionalOperator(S, CO, T); 2730 return; 2731 } 2732 2733 // Go ahead and check any implicit conversions we might have skipped. 2734 // The non-canonical typecheck is just an optimization; 2735 // CheckImplicitConversion will filter out dead implicit conversions. 2736 if (E->getType() != T) 2737 CheckImplicitConversion(S, E, T); 2738 2739 // Now continue drilling into this expression. 2740 2741 // Skip past explicit casts. 2742 if (isa<ExplicitCastExpr>(E)) { 2743 E = cast<ExplicitCastExpr>(E)->getSubExpr()->IgnoreParenImpCasts(); 2744 return AnalyzeImplicitConversions(S, E); 2745 } 2746 2747 // Do a somewhat different check with comparison operators. 2748 if (isa<BinaryOperator>(E) && cast<BinaryOperator>(E)->isComparisonOp()) 2749 return AnalyzeComparison(S, cast<BinaryOperator>(E)); 2750 2751 // These break the otherwise-useful invariant below. Fortunately, 2752 // we don't really need to recurse into them, because any internal 2753 // expressions should have been analyzed already when they were 2754 // built into statements. 2755 if (isa<StmtExpr>(E)) return; 2756 2757 // Don't descend into unevaluated contexts. 2758 if (isa<SizeOfAlignOfExpr>(E)) return; 2759 2760 // Now just recurse over the expression's children. 2761 for (Stmt::child_iterator I = E->child_begin(), IE = E->child_end(); 2762 I != IE; ++I) 2763 AnalyzeImplicitConversions(S, cast<Expr>(*I)); 2764} 2765 2766} // end anonymous namespace 2767 2768/// Diagnoses "dangerous" implicit conversions within the given 2769/// expression (which is a full expression). Implements -Wconversion 2770/// and -Wsign-compare. 2771void Sema::CheckImplicitConversions(Expr *E) { 2772 // Don't diagnose in unevaluated contexts. 2773 if (ExprEvalContexts.back().Context == Sema::Unevaluated) 2774 return; 2775 2776 // Don't diagnose for value- or type-dependent expressions. 2777 if (E->isTypeDependent() || E->isValueDependent()) 2778 return; 2779 2780 AnalyzeImplicitConversions(*this, E); 2781} 2782 2783/// CheckParmsForFunctionDef - Check that the parameters of the given 2784/// function are appropriate for the definition of a function. This 2785/// takes care of any checks that cannot be performed on the 2786/// declaration itself, e.g., that the types of each of the function 2787/// parameters are complete. 2788bool Sema::CheckParmsForFunctionDef(FunctionDecl *FD) { 2789 bool HasInvalidParm = false; 2790 for (unsigned p = 0, NumParams = FD->getNumParams(); p < NumParams; ++p) { 2791 ParmVarDecl *Param = FD->getParamDecl(p); 2792 2793 // C99 6.7.5.3p4: the parameters in a parameter type list in a 2794 // function declarator that is part of a function definition of 2795 // that function shall not have incomplete type. 2796 // 2797 // This is also C++ [dcl.fct]p6. 2798 if (!Param->isInvalidDecl() && 2799 RequireCompleteType(Param->getLocation(), Param->getType(), 2800 diag::err_typecheck_decl_incomplete_type)) { 2801 Param->setInvalidDecl(); 2802 HasInvalidParm = true; 2803 } 2804 2805 // C99 6.9.1p5: If the declarator includes a parameter type list, the 2806 // declaration of each parameter shall include an identifier. 2807 if (Param->getIdentifier() == 0 && 2808 !Param->isImplicit() && 2809 !getLangOptions().CPlusPlus) 2810 Diag(Param->getLocation(), diag::err_parameter_name_omitted); 2811 2812 // C99 6.7.5.3p12: 2813 // If the function declarator is not part of a definition of that 2814 // function, parameters may have incomplete type and may use the [*] 2815 // notation in their sequences of declarator specifiers to specify 2816 // variable length array types. 2817 QualType PType = Param->getOriginalType(); 2818 if (const ArrayType *AT = Context.getAsArrayType(PType)) { 2819 if (AT->getSizeModifier() == ArrayType::Star) { 2820 // FIXME: This diagnosic should point the the '[*]' if source-location 2821 // information is added for it. 2822 Diag(Param->getLocation(), diag::err_array_star_in_function_definition); 2823 } 2824 } 2825 } 2826 2827 return HasInvalidParm; 2828} 2829 2830/// CheckCastAlign - Implements -Wcast-align, which warns when a 2831/// pointer cast increases the alignment requirements. 2832void Sema::CheckCastAlign(Expr *Op, QualType T, SourceRange TRange) { 2833 // This is actually a lot of work to potentially be doing on every 2834 // cast; don't do it if we're ignoring -Wcast_align (as is the default). 2835 if (getDiagnostics().getDiagnosticLevel(diag::warn_cast_align) 2836 == Diagnostic::Ignored) 2837 return; 2838 2839 // Ignore dependent types. 2840 if (T->isDependentType() || Op->getType()->isDependentType()) 2841 return; 2842 2843 // Require that the destination be a pointer type. 2844 const PointerType *DestPtr = T->getAs<PointerType>(); 2845 if (!DestPtr) return; 2846 2847 // If the destination has alignment 1, we're done. 2848 QualType DestPointee = DestPtr->getPointeeType(); 2849 if (DestPointee->isIncompleteType()) return; 2850 CharUnits DestAlign = Context.getTypeAlignInChars(DestPointee); 2851 if (DestAlign.isOne()) return; 2852 2853 // Require that the source be a pointer type. 2854 const PointerType *SrcPtr = Op->getType()->getAs<PointerType>(); 2855 if (!SrcPtr) return; 2856 QualType SrcPointee = SrcPtr->getPointeeType(); 2857 2858 // Whitelist casts from cv void*. We already implicitly 2859 // whitelisted casts to cv void*, since they have alignment 1. 2860 // Also whitelist casts involving incomplete types, which implicitly 2861 // includes 'void'. 2862 if (SrcPointee->isIncompleteType()) return; 2863 2864 CharUnits SrcAlign = Context.getTypeAlignInChars(SrcPointee); 2865 if (SrcAlign >= DestAlign) return; 2866 2867 Diag(TRange.getBegin(), diag::warn_cast_align) 2868 << Op->getType() << T 2869 << static_cast<unsigned>(SrcAlign.getQuantity()) 2870 << static_cast<unsigned>(DestAlign.getQuantity()) 2871 << TRange << Op->getSourceRange(); 2872} 2873 2874