SemaChecking.cpp revision 91f54ce93bec136fb9e18740b895cf1c1339524b
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 PP.getTargetInfo()); 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 // Find out if any arguments are required to be integer constant expressions. 135 unsigned ICEArguments = 0; 136 ASTContext::GetBuiltinTypeError Error; 137 Context.GetBuiltinType(BuiltinID, Error, &ICEArguments); 138 if (Error != ASTContext::GE_None) 139 ICEArguments = 0; // Don't diagnose previously diagnosed errors. 140 141 // If any arguments are required to be ICE's, check and diagnose. 142 for (unsigned ArgNo = 0; ICEArguments != 0; ++ArgNo) { 143 // Skip arguments not required to be ICE's. 144 if ((ICEArguments & (1 << ArgNo)) == 0) continue; 145 146 llvm::APSInt Result; 147 if (SemaBuiltinConstantArg(TheCall, ArgNo, Result)) 148 return true; 149 ICEArguments &= ~(1 << ArgNo); 150 } 151 152 switch (BuiltinID) { 153 case Builtin::BI__builtin___CFStringMakeConstantString: 154 assert(TheCall->getNumArgs() == 1 && 155 "Wrong # arguments to builtin CFStringMakeConstantString"); 156 if (CheckObjCString(TheCall->getArg(0))) 157 return ExprError(); 158 break; 159 case Builtin::BI__builtin_stdarg_start: 160 case Builtin::BI__builtin_va_start: 161 if (SemaBuiltinVAStart(TheCall)) 162 return ExprError(); 163 break; 164 case Builtin::BI__builtin_isgreater: 165 case Builtin::BI__builtin_isgreaterequal: 166 case Builtin::BI__builtin_isless: 167 case Builtin::BI__builtin_islessequal: 168 case Builtin::BI__builtin_islessgreater: 169 case Builtin::BI__builtin_isunordered: 170 if (SemaBuiltinUnorderedCompare(TheCall)) 171 return ExprError(); 172 break; 173 case Builtin::BI__builtin_fpclassify: 174 if (SemaBuiltinFPClassification(TheCall, 6)) 175 return ExprError(); 176 break; 177 case Builtin::BI__builtin_isfinite: 178 case Builtin::BI__builtin_isinf: 179 case Builtin::BI__builtin_isinf_sign: 180 case Builtin::BI__builtin_isnan: 181 case Builtin::BI__builtin_isnormal: 182 if (SemaBuiltinFPClassification(TheCall, 1)) 183 return ExprError(); 184 break; 185 case Builtin::BI__builtin_shufflevector: 186 return SemaBuiltinShuffleVector(TheCall); 187 // TheCall will be freed by the smart pointer here, but that's fine, since 188 // SemaBuiltinShuffleVector guts it, but then doesn't release it. 189 case Builtin::BI__builtin_prefetch: 190 if (SemaBuiltinPrefetch(TheCall)) 191 return ExprError(); 192 break; 193 case Builtin::BI__builtin_object_size: 194 if (SemaBuiltinObjectSize(TheCall)) 195 return ExprError(); 196 break; 197 case Builtin::BI__builtin_longjmp: 198 if (SemaBuiltinLongjmp(TheCall)) 199 return ExprError(); 200 break; 201 case Builtin::BI__builtin_constant_p: 202 if (TheCall->getNumArgs() == 0) 203 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 204 << 0 /*function call*/ << 1 << 0 << TheCall->getSourceRange(); 205 if (TheCall->getNumArgs() > 1) 206 return Diag(TheCall->getArg(1)->getLocStart(), 207 diag::err_typecheck_call_too_many_args) 208 << 0 /*function call*/ << 1 << TheCall->getNumArgs() 209 << TheCall->getArg(1)->getSourceRange(); 210 break; 211 case Builtin::BI__sync_fetch_and_add: 212 case Builtin::BI__sync_fetch_and_sub: 213 case Builtin::BI__sync_fetch_and_or: 214 case Builtin::BI__sync_fetch_and_and: 215 case Builtin::BI__sync_fetch_and_xor: 216 case Builtin::BI__sync_add_and_fetch: 217 case Builtin::BI__sync_sub_and_fetch: 218 case Builtin::BI__sync_and_and_fetch: 219 case Builtin::BI__sync_or_and_fetch: 220 case Builtin::BI__sync_xor_and_fetch: 221 case Builtin::BI__sync_val_compare_and_swap: 222 case Builtin::BI__sync_bool_compare_and_swap: 223 case Builtin::BI__sync_lock_test_and_set: 224 case Builtin::BI__sync_lock_release: 225 return SemaBuiltinAtomicOverloaded(move(TheCallResult)); 226 } 227 228 // Since the target specific builtins for each arch overlap, only check those 229 // of the arch we are compiling for. 230 if (BuiltinID >= Builtin::FirstTSBuiltin) { 231 switch (Context.Target.getTriple().getArch()) { 232 case llvm::Triple::arm: 233 case llvm::Triple::thumb: 234 if (CheckARMBuiltinFunctionCall(BuiltinID, TheCall)) 235 return ExprError(); 236 break; 237 default: 238 break; 239 } 240 } 241 242 return move(TheCallResult); 243} 244 245// Get the valid immediate range for the specified NEON type code. 246static unsigned RFT(unsigned t, bool shift = false) { 247 bool quad = t & 0x10; 248 249 switch (t & 0x7) { 250 case 0: // i8 251 return shift ? 7 : (8 << (int)quad) - 1; 252 case 1: // i16 253 return shift ? 15 : (4 << (int)quad) - 1; 254 case 2: // i32 255 return shift ? 31 : (2 << (int)quad) - 1; 256 case 3: // i64 257 return shift ? 63 : (1 << (int)quad) - 1; 258 case 4: // f32 259 assert(!shift && "cannot shift float types!"); 260 return (2 << (int)quad) - 1; 261 case 5: // poly8 262 assert(!shift && "cannot shift polynomial types!"); 263 return (8 << (int)quad) - 1; 264 case 6: // poly16 265 assert(!shift && "cannot shift polynomial types!"); 266 return (4 << (int)quad) - 1; 267 case 7: // float16 268 assert(!shift && "cannot shift float types!"); 269 return (4 << (int)quad) - 1; 270 } 271 return 0; 272} 273 274bool Sema::CheckARMBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) { 275 llvm::APSInt Result; 276 277 unsigned mask = 0; 278 unsigned TV = 0; 279 switch (BuiltinID) { 280#define GET_NEON_OVERLOAD_CHECK 281#include "clang/Basic/arm_neon.inc" 282#undef GET_NEON_OVERLOAD_CHECK 283 } 284 285 // For NEON intrinsics which are overloaded on vector element type, validate 286 // the immediate which specifies which variant to emit. 287 if (mask) { 288 unsigned ArgNo = TheCall->getNumArgs()-1; 289 if (SemaBuiltinConstantArg(TheCall, ArgNo, Result)) 290 return true; 291 292 TV = Result.getLimitedValue(32); 293 if ((TV > 31) || (mask & (1 << TV)) == 0) 294 return Diag(TheCall->getLocStart(), diag::err_invalid_neon_type_code) 295 << TheCall->getArg(ArgNo)->getSourceRange(); 296 } 297 298 // For NEON intrinsics which take an immediate value as part of the 299 // instruction, range check them here. 300 unsigned i = 0, l = 0, u = 0; 301 switch (BuiltinID) { 302 default: return false; 303 case ARM::BI__builtin_arm_ssat: i = 1; l = 1; u = 31; break; 304 case ARM::BI__builtin_arm_usat: i = 1; u = 31; break; 305 case ARM::BI__builtin_arm_vcvtr_f: 306 case ARM::BI__builtin_arm_vcvtr_d: i = 1; u = 1; break; 307#define GET_NEON_IMMEDIATE_CHECK 308#include "clang/Basic/arm_neon.inc" 309#undef GET_NEON_IMMEDIATE_CHECK 310 }; 311 312 // Check that the immediate argument is actually a constant. 313 if (SemaBuiltinConstantArg(TheCall, i, Result)) 314 return true; 315 316 // Range check against the upper/lower values for this isntruction. 317 unsigned Val = Result.getZExtValue(); 318 if (Val < l || Val > (u + l)) 319 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 320 << l << u+l << TheCall->getArg(i)->getSourceRange(); 321 322 // FIXME: VFP Intrinsics should error if VFP not present. 323 return false; 324} 325 326/// CheckFunctionCall - Check a direct function call for various correctness 327/// and safety properties not strictly enforced by the C type system. 328bool Sema::CheckFunctionCall(FunctionDecl *FDecl, CallExpr *TheCall) { 329 // Get the IdentifierInfo* for the called function. 330 IdentifierInfo *FnInfo = FDecl->getIdentifier(); 331 332 // None of the checks below are needed for functions that don't have 333 // simple names (e.g., C++ conversion functions). 334 if (!FnInfo) 335 return false; 336 337 // FIXME: This mechanism should be abstracted to be less fragile and 338 // more efficient. For example, just map function ids to custom 339 // handlers. 340 341 // Printf and scanf checking. 342 for (specific_attr_iterator<FormatAttr> 343 i = FDecl->specific_attr_begin<FormatAttr>(), 344 e = FDecl->specific_attr_end<FormatAttr>(); i != e ; ++i) { 345 346 const FormatAttr *Format = *i; 347 const bool b = Format->getType() == "scanf"; 348 if (b || CheckablePrintfAttr(Format, TheCall)) { 349 bool HasVAListArg = Format->getFirstArg() == 0; 350 CheckPrintfScanfArguments(TheCall, HasVAListArg, 351 Format->getFormatIdx() - 1, 352 HasVAListArg ? 0 : Format->getFirstArg() - 1, 353 !b); 354 } 355 } 356 357 for (specific_attr_iterator<NonNullAttr> 358 i = FDecl->specific_attr_begin<NonNullAttr>(), 359 e = FDecl->specific_attr_end<NonNullAttr>(); i != e; ++i) { 360 CheckNonNullArguments(*i, TheCall); 361 } 362 363 return false; 364} 365 366bool Sema::CheckBlockCall(NamedDecl *NDecl, CallExpr *TheCall) { 367 // Printf checking. 368 const FormatAttr *Format = NDecl->getAttr<FormatAttr>(); 369 if (!Format) 370 return false; 371 372 const VarDecl *V = dyn_cast<VarDecl>(NDecl); 373 if (!V) 374 return false; 375 376 QualType Ty = V->getType(); 377 if (!Ty->isBlockPointerType()) 378 return false; 379 380 const bool b = Format->getType() == "scanf"; 381 if (!b && !CheckablePrintfAttr(Format, TheCall)) 382 return false; 383 384 bool HasVAListArg = Format->getFirstArg() == 0; 385 CheckPrintfScanfArguments(TheCall, HasVAListArg, Format->getFormatIdx() - 1, 386 HasVAListArg ? 0 : Format->getFirstArg() - 1, !b); 387 388 return false; 389} 390 391/// SemaBuiltinAtomicOverloaded - We have a call to a function like 392/// __sync_fetch_and_add, which is an overloaded function based on the pointer 393/// type of its first argument. The main ActOnCallExpr routines have already 394/// promoted the types of arguments because all of these calls are prototyped as 395/// void(...). 396/// 397/// This function goes through and does final semantic checking for these 398/// builtins, 399ExprResult 400Sema::SemaBuiltinAtomicOverloaded(ExprResult TheCallResult) { 401 CallExpr *TheCall = (CallExpr *)TheCallResult.get(); 402 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 403 FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl()); 404 405 // Ensure that we have at least one argument to do type inference from. 406 if (TheCall->getNumArgs() < 1) { 407 Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args_at_least) 408 << 0 << 1 << TheCall->getNumArgs() 409 << TheCall->getCallee()->getSourceRange(); 410 return ExprError(); 411 } 412 413 // Inspect the first argument of the atomic builtin. This should always be 414 // a pointer type, whose element is an integral scalar or pointer type. 415 // Because it is a pointer type, we don't have to worry about any implicit 416 // casts here. 417 // FIXME: We don't allow floating point scalars as input. 418 Expr *FirstArg = TheCall->getArg(0); 419 if (!FirstArg->getType()->isPointerType()) { 420 Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer) 421 << FirstArg->getType() << FirstArg->getSourceRange(); 422 return ExprError(); 423 } 424 425 QualType ValType = 426 FirstArg->getType()->getAs<PointerType>()->getPointeeType(); 427 if (!ValType->isIntegerType() && !ValType->isAnyPointerType() && 428 !ValType->isBlockPointerType()) { 429 Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer_intptr) 430 << FirstArg->getType() << FirstArg->getSourceRange(); 431 return ExprError(); 432 } 433 434 // The majority of builtins return a value, but a few have special return 435 // types, so allow them to override appropriately below. 436 QualType ResultType = ValType; 437 438 // We need to figure out which concrete builtin this maps onto. For example, 439 // __sync_fetch_and_add with a 2 byte object turns into 440 // __sync_fetch_and_add_2. 441#define BUILTIN_ROW(x) \ 442 { Builtin::BI##x##_1, Builtin::BI##x##_2, Builtin::BI##x##_4, \ 443 Builtin::BI##x##_8, Builtin::BI##x##_16 } 444 445 static const unsigned BuiltinIndices[][5] = { 446 BUILTIN_ROW(__sync_fetch_and_add), 447 BUILTIN_ROW(__sync_fetch_and_sub), 448 BUILTIN_ROW(__sync_fetch_and_or), 449 BUILTIN_ROW(__sync_fetch_and_and), 450 BUILTIN_ROW(__sync_fetch_and_xor), 451 452 BUILTIN_ROW(__sync_add_and_fetch), 453 BUILTIN_ROW(__sync_sub_and_fetch), 454 BUILTIN_ROW(__sync_and_and_fetch), 455 BUILTIN_ROW(__sync_or_and_fetch), 456 BUILTIN_ROW(__sync_xor_and_fetch), 457 458 BUILTIN_ROW(__sync_val_compare_and_swap), 459 BUILTIN_ROW(__sync_bool_compare_and_swap), 460 BUILTIN_ROW(__sync_lock_test_and_set), 461 BUILTIN_ROW(__sync_lock_release) 462 }; 463#undef BUILTIN_ROW 464 465 // Determine the index of the size. 466 unsigned SizeIndex; 467 switch (Context.getTypeSizeInChars(ValType).getQuantity()) { 468 case 1: SizeIndex = 0; break; 469 case 2: SizeIndex = 1; break; 470 case 4: SizeIndex = 2; break; 471 case 8: SizeIndex = 3; break; 472 case 16: SizeIndex = 4; break; 473 default: 474 Diag(DRE->getLocStart(), diag::err_atomic_builtin_pointer_size) 475 << FirstArg->getType() << FirstArg->getSourceRange(); 476 return ExprError(); 477 } 478 479 // Each of these builtins has one pointer argument, followed by some number of 480 // values (0, 1 or 2) followed by a potentially empty varags list of stuff 481 // that we ignore. Find out which row of BuiltinIndices to read from as well 482 // as the number of fixed args. 483 unsigned BuiltinID = FDecl->getBuiltinID(); 484 unsigned BuiltinIndex, NumFixed = 1; 485 switch (BuiltinID) { 486 default: assert(0 && "Unknown overloaded atomic builtin!"); 487 case Builtin::BI__sync_fetch_and_add: BuiltinIndex = 0; break; 488 case Builtin::BI__sync_fetch_and_sub: BuiltinIndex = 1; break; 489 case Builtin::BI__sync_fetch_and_or: BuiltinIndex = 2; break; 490 case Builtin::BI__sync_fetch_and_and: BuiltinIndex = 3; break; 491 case Builtin::BI__sync_fetch_and_xor: BuiltinIndex = 4; break; 492 493 case Builtin::BI__sync_add_and_fetch: BuiltinIndex = 5; break; 494 case Builtin::BI__sync_sub_and_fetch: BuiltinIndex = 6; break; 495 case Builtin::BI__sync_and_and_fetch: BuiltinIndex = 7; break; 496 case Builtin::BI__sync_or_and_fetch: BuiltinIndex = 8; break; 497 case Builtin::BI__sync_xor_and_fetch: BuiltinIndex = 9; break; 498 499 case Builtin::BI__sync_val_compare_and_swap: 500 BuiltinIndex = 10; 501 NumFixed = 2; 502 break; 503 case Builtin::BI__sync_bool_compare_and_swap: 504 BuiltinIndex = 11; 505 NumFixed = 2; 506 ResultType = Context.BoolTy; 507 break; 508 case Builtin::BI__sync_lock_test_and_set: BuiltinIndex = 12; break; 509 case Builtin::BI__sync_lock_release: 510 BuiltinIndex = 13; 511 NumFixed = 0; 512 ResultType = Context.VoidTy; 513 break; 514 } 515 516 // Now that we know how many fixed arguments we expect, first check that we 517 // have at least that many. 518 if (TheCall->getNumArgs() < 1+NumFixed) { 519 Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args_at_least) 520 << 0 << 1+NumFixed << TheCall->getNumArgs() 521 << TheCall->getCallee()->getSourceRange(); 522 return ExprError(); 523 } 524 525 // Get the decl for the concrete builtin from this, we can tell what the 526 // concrete integer type we should convert to is. 527 unsigned NewBuiltinID = BuiltinIndices[BuiltinIndex][SizeIndex]; 528 const char *NewBuiltinName = Context.BuiltinInfo.GetName(NewBuiltinID); 529 IdentifierInfo *NewBuiltinII = PP.getIdentifierInfo(NewBuiltinName); 530 FunctionDecl *NewBuiltinDecl = 531 cast<FunctionDecl>(LazilyCreateBuiltin(NewBuiltinII, NewBuiltinID, 532 TUScope, false, DRE->getLocStart())); 533 534 // The first argument --- the pointer --- has a fixed type; we 535 // deduce the types of the rest of the arguments accordingly. Walk 536 // the remaining arguments, converting them to the deduced value type. 537 for (unsigned i = 0; i != NumFixed; ++i) { 538 Expr *Arg = TheCall->getArg(i+1); 539 540 // If the argument is an implicit cast, then there was a promotion due to 541 // "...", just remove it now. 542 if (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(Arg)) { 543 Arg = ICE->getSubExpr(); 544 ICE->setSubExpr(0); 545 TheCall->setArg(i+1, Arg); 546 } 547 548 // GCC does an implicit conversion to the pointer or integer ValType. This 549 // can fail in some cases (1i -> int**), check for this error case now. 550 CastKind Kind = CK_Invalid; 551 CXXCastPath BasePath; 552 if (CheckCastTypes(Arg->getSourceRange(), ValType, Arg, Kind, BasePath)) 553 return ExprError(); 554 555 // Okay, we have something that *can* be converted to the right type. Check 556 // to see if there is a potentially weird extension going on here. This can 557 // happen when you do an atomic operation on something like an char* and 558 // pass in 42. The 42 gets converted to char. This is even more strange 559 // for things like 45.123 -> char, etc. 560 // FIXME: Do this check. 561 ImpCastExprToType(Arg, ValType, Kind, VK_RValue, &BasePath); 562 TheCall->setArg(i+1, Arg); 563 } 564 565 // Switch the DeclRefExpr to refer to the new decl. 566 DRE->setDecl(NewBuiltinDecl); 567 DRE->setType(NewBuiltinDecl->getType()); 568 569 // Set the callee in the CallExpr. 570 // FIXME: This leaks the original parens and implicit casts. 571 Expr *PromotedCall = DRE; 572 UsualUnaryConversions(PromotedCall); 573 TheCall->setCallee(PromotedCall); 574 575 // Change the result type of the call to match the original value type. This 576 // is arbitrary, but the codegen for these builtins ins design to handle it 577 // gracefully. 578 TheCall->setType(ResultType); 579 580 return move(TheCallResult); 581} 582 583 584/// CheckObjCString - Checks that the argument to the builtin 585/// CFString constructor is correct 586/// Note: It might also make sense to do the UTF-16 conversion here (would 587/// simplify the backend). 588bool Sema::CheckObjCString(Expr *Arg) { 589 Arg = Arg->IgnoreParenCasts(); 590 StringLiteral *Literal = dyn_cast<StringLiteral>(Arg); 591 592 if (!Literal || Literal->isWide()) { 593 Diag(Arg->getLocStart(), diag::err_cfstring_literal_not_string_constant) 594 << Arg->getSourceRange(); 595 return true; 596 } 597 598 size_t NulPos = Literal->getString().find('\0'); 599 if (NulPos != llvm::StringRef::npos) { 600 Diag(getLocationOfStringLiteralByte(Literal, NulPos), 601 diag::warn_cfstring_literal_contains_nul_character) 602 << Arg->getSourceRange(); 603 } 604 if (Literal->containsNonAsciiOrNull()) { 605 llvm::StringRef String = Literal->getString(); 606 unsigned NumBytes = String.size(); 607 llvm::SmallVector<UTF16, 128> ToBuf(NumBytes); 608 const UTF8 *FromPtr = (UTF8 *)String.data(); 609 UTF16 *ToPtr = &ToBuf[0]; 610 611 ConversionResult Result = ConvertUTF8toUTF16(&FromPtr, FromPtr + NumBytes, 612 &ToPtr, ToPtr + NumBytes, 613 strictConversion); 614 // Check for conversion failure. 615 if (Result != conversionOK) 616 Diag(Arg->getLocStart(), 617 diag::warn_cfstring_truncated) << Arg->getSourceRange(); 618 } 619 return false; 620} 621 622/// SemaBuiltinVAStart - Check the arguments to __builtin_va_start for validity. 623/// Emit an error and return true on failure, return false on success. 624bool Sema::SemaBuiltinVAStart(CallExpr *TheCall) { 625 Expr *Fn = TheCall->getCallee(); 626 if (TheCall->getNumArgs() > 2) { 627 Diag(TheCall->getArg(2)->getLocStart(), 628 diag::err_typecheck_call_too_many_args) 629 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 630 << Fn->getSourceRange() 631 << SourceRange(TheCall->getArg(2)->getLocStart(), 632 (*(TheCall->arg_end()-1))->getLocEnd()); 633 return true; 634 } 635 636 if (TheCall->getNumArgs() < 2) { 637 return Diag(TheCall->getLocEnd(), 638 diag::err_typecheck_call_too_few_args_at_least) 639 << 0 /*function call*/ << 2 << TheCall->getNumArgs(); 640 } 641 642 // Determine whether the current function is variadic or not. 643 BlockScopeInfo *CurBlock = getCurBlock(); 644 bool isVariadic; 645 if (CurBlock) 646 isVariadic = CurBlock->TheDecl->isVariadic(); 647 else if (FunctionDecl *FD = getCurFunctionDecl()) 648 isVariadic = FD->isVariadic(); 649 else 650 isVariadic = getCurMethodDecl()->isVariadic(); 651 652 if (!isVariadic) { 653 Diag(Fn->getLocStart(), diag::err_va_start_used_in_non_variadic_function); 654 return true; 655 } 656 657 // Verify that the second argument to the builtin is the last argument of the 658 // current function or method. 659 bool SecondArgIsLastNamedArgument = false; 660 const Expr *Arg = TheCall->getArg(1)->IgnoreParenCasts(); 661 662 if (const DeclRefExpr *DR = dyn_cast<DeclRefExpr>(Arg)) { 663 if (const ParmVarDecl *PV = dyn_cast<ParmVarDecl>(DR->getDecl())) { 664 // FIXME: This isn't correct for methods (results in bogus warning). 665 // Get the last formal in the current function. 666 const ParmVarDecl *LastArg; 667 if (CurBlock) 668 LastArg = *(CurBlock->TheDecl->param_end()-1); 669 else if (FunctionDecl *FD = getCurFunctionDecl()) 670 LastArg = *(FD->param_end()-1); 671 else 672 LastArg = *(getCurMethodDecl()->param_end()-1); 673 SecondArgIsLastNamedArgument = PV == LastArg; 674 } 675 } 676 677 if (!SecondArgIsLastNamedArgument) 678 Diag(TheCall->getArg(1)->getLocStart(), 679 diag::warn_second_parameter_of_va_start_not_last_named_argument); 680 return false; 681} 682 683/// SemaBuiltinUnorderedCompare - Handle functions like __builtin_isgreater and 684/// friends. This is declared to take (...), so we have to check everything. 685bool Sema::SemaBuiltinUnorderedCompare(CallExpr *TheCall) { 686 if (TheCall->getNumArgs() < 2) 687 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 688 << 0 << 2 << TheCall->getNumArgs()/*function call*/; 689 if (TheCall->getNumArgs() > 2) 690 return Diag(TheCall->getArg(2)->getLocStart(), 691 diag::err_typecheck_call_too_many_args) 692 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 693 << SourceRange(TheCall->getArg(2)->getLocStart(), 694 (*(TheCall->arg_end()-1))->getLocEnd()); 695 696 Expr *OrigArg0 = TheCall->getArg(0); 697 Expr *OrigArg1 = TheCall->getArg(1); 698 699 // Do standard promotions between the two arguments, returning their common 700 // type. 701 QualType Res = UsualArithmeticConversions(OrigArg0, OrigArg1, false); 702 703 // Make sure any conversions are pushed back into the call; this is 704 // type safe since unordered compare builtins are declared as "_Bool 705 // foo(...)". 706 TheCall->setArg(0, OrigArg0); 707 TheCall->setArg(1, OrigArg1); 708 709 if (OrigArg0->isTypeDependent() || OrigArg1->isTypeDependent()) 710 return false; 711 712 // If the common type isn't a real floating type, then the arguments were 713 // invalid for this operation. 714 if (!Res->isRealFloatingType()) 715 return Diag(OrigArg0->getLocStart(), 716 diag::err_typecheck_call_invalid_ordered_compare) 717 << OrigArg0->getType() << OrigArg1->getType() 718 << SourceRange(OrigArg0->getLocStart(), OrigArg1->getLocEnd()); 719 720 return false; 721} 722 723/// SemaBuiltinSemaBuiltinFPClassification - Handle functions like 724/// __builtin_isnan and friends. This is declared to take (...), so we have 725/// to check everything. We expect the last argument to be a floating point 726/// value. 727bool Sema::SemaBuiltinFPClassification(CallExpr *TheCall, unsigned NumArgs) { 728 if (TheCall->getNumArgs() < NumArgs) 729 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 730 << 0 << NumArgs << TheCall->getNumArgs()/*function call*/; 731 if (TheCall->getNumArgs() > NumArgs) 732 return Diag(TheCall->getArg(NumArgs)->getLocStart(), 733 diag::err_typecheck_call_too_many_args) 734 << 0 /*function call*/ << NumArgs << TheCall->getNumArgs() 735 << SourceRange(TheCall->getArg(NumArgs)->getLocStart(), 736 (*(TheCall->arg_end()-1))->getLocEnd()); 737 738 Expr *OrigArg = TheCall->getArg(NumArgs-1); 739 740 if (OrigArg->isTypeDependent()) 741 return false; 742 743 // This operation requires a non-_Complex floating-point number. 744 if (!OrigArg->getType()->isRealFloatingType()) 745 return Diag(OrigArg->getLocStart(), 746 diag::err_typecheck_call_invalid_unary_fp) 747 << OrigArg->getType() << OrigArg->getSourceRange(); 748 749 // If this is an implicit conversion from float -> double, remove it. 750 if (ImplicitCastExpr *Cast = dyn_cast<ImplicitCastExpr>(OrigArg)) { 751 Expr *CastArg = Cast->getSubExpr(); 752 if (CastArg->getType()->isSpecificBuiltinType(BuiltinType::Float)) { 753 assert(Cast->getType()->isSpecificBuiltinType(BuiltinType::Double) && 754 "promotion from float to double is the only expected cast here"); 755 Cast->setSubExpr(0); 756 TheCall->setArg(NumArgs-1, CastArg); 757 OrigArg = CastArg; 758 } 759 } 760 761 return false; 762} 763 764/// SemaBuiltinShuffleVector - Handle __builtin_shufflevector. 765// This is declared to take (...), so we have to check everything. 766ExprResult Sema::SemaBuiltinShuffleVector(CallExpr *TheCall) { 767 if (TheCall->getNumArgs() < 2) 768 return ExprError(Diag(TheCall->getLocEnd(), 769 diag::err_typecheck_call_too_few_args_at_least) 770 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 771 << TheCall->getSourceRange()); 772 773 // Determine which of the following types of shufflevector we're checking: 774 // 1) unary, vector mask: (lhs, mask) 775 // 2) binary, vector mask: (lhs, rhs, mask) 776 // 3) binary, scalar mask: (lhs, rhs, index, ..., index) 777 QualType resType = TheCall->getArg(0)->getType(); 778 unsigned numElements = 0; 779 780 if (!TheCall->getArg(0)->isTypeDependent() && 781 !TheCall->getArg(1)->isTypeDependent()) { 782 QualType LHSType = TheCall->getArg(0)->getType(); 783 QualType RHSType = TheCall->getArg(1)->getType(); 784 785 if (!LHSType->isVectorType() || !RHSType->isVectorType()) { 786 Diag(TheCall->getLocStart(), diag::err_shufflevector_non_vector) 787 << SourceRange(TheCall->getArg(0)->getLocStart(), 788 TheCall->getArg(1)->getLocEnd()); 789 return ExprError(); 790 } 791 792 numElements = LHSType->getAs<VectorType>()->getNumElements(); 793 unsigned numResElements = TheCall->getNumArgs() - 2; 794 795 // Check to see if we have a call with 2 vector arguments, the unary shuffle 796 // with mask. If so, verify that RHS is an integer vector type with the 797 // same number of elts as lhs. 798 if (TheCall->getNumArgs() == 2) { 799 if (!RHSType->hasIntegerRepresentation() || 800 RHSType->getAs<VectorType>()->getNumElements() != numElements) 801 Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector) 802 << SourceRange(TheCall->getArg(1)->getLocStart(), 803 TheCall->getArg(1)->getLocEnd()); 804 numResElements = numElements; 805 } 806 else if (!Context.hasSameUnqualifiedType(LHSType, RHSType)) { 807 Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector) 808 << SourceRange(TheCall->getArg(0)->getLocStart(), 809 TheCall->getArg(1)->getLocEnd()); 810 return ExprError(); 811 } else if (numElements != numResElements) { 812 QualType eltType = LHSType->getAs<VectorType>()->getElementType(); 813 resType = Context.getVectorType(eltType, numResElements, 814 VectorType::GenericVector); 815 } 816 } 817 818 for (unsigned i = 2; i < TheCall->getNumArgs(); i++) { 819 if (TheCall->getArg(i)->isTypeDependent() || 820 TheCall->getArg(i)->isValueDependent()) 821 continue; 822 823 llvm::APSInt Result(32); 824 if (!TheCall->getArg(i)->isIntegerConstantExpr(Result, Context)) 825 return ExprError(Diag(TheCall->getLocStart(), 826 diag::err_shufflevector_nonconstant_argument) 827 << TheCall->getArg(i)->getSourceRange()); 828 829 if (Result.getActiveBits() > 64 || Result.getZExtValue() >= numElements*2) 830 return ExprError(Diag(TheCall->getLocStart(), 831 diag::err_shufflevector_argument_too_large) 832 << TheCall->getArg(i)->getSourceRange()); 833 } 834 835 llvm::SmallVector<Expr*, 32> exprs; 836 837 for (unsigned i = 0, e = TheCall->getNumArgs(); i != e; i++) { 838 exprs.push_back(TheCall->getArg(i)); 839 TheCall->setArg(i, 0); 840 } 841 842 return Owned(new (Context) ShuffleVectorExpr(Context, exprs.begin(), 843 exprs.size(), resType, 844 TheCall->getCallee()->getLocStart(), 845 TheCall->getRParenLoc())); 846} 847 848/// SemaBuiltinPrefetch - Handle __builtin_prefetch. 849// This is declared to take (const void*, ...) and can take two 850// optional constant int args. 851bool Sema::SemaBuiltinPrefetch(CallExpr *TheCall) { 852 unsigned NumArgs = TheCall->getNumArgs(); 853 854 if (NumArgs > 3) 855 return Diag(TheCall->getLocEnd(), 856 diag::err_typecheck_call_too_many_args_at_most) 857 << 0 /*function call*/ << 3 << NumArgs 858 << TheCall->getSourceRange(); 859 860 // Argument 0 is checked for us and the remaining arguments must be 861 // constant integers. 862 for (unsigned i = 1; i != NumArgs; ++i) { 863 Expr *Arg = TheCall->getArg(i); 864 865 llvm::APSInt Result; 866 if (SemaBuiltinConstantArg(TheCall, i, Result)) 867 return true; 868 869 // FIXME: gcc issues a warning and rewrites these to 0. These 870 // seems especially odd for the third argument since the default 871 // is 3. 872 if (i == 1) { 873 if (Result.getLimitedValue() > 1) 874 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 875 << "0" << "1" << Arg->getSourceRange(); 876 } else { 877 if (Result.getLimitedValue() > 3) 878 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 879 << "0" << "3" << Arg->getSourceRange(); 880 } 881 } 882 883 return false; 884} 885 886/// SemaBuiltinConstantArg - Handle a check if argument ArgNum of CallExpr 887/// TheCall is a constant expression. 888bool Sema::SemaBuiltinConstantArg(CallExpr *TheCall, int ArgNum, 889 llvm::APSInt &Result) { 890 Expr *Arg = TheCall->getArg(ArgNum); 891 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 892 FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl()); 893 894 if (Arg->isTypeDependent() || Arg->isValueDependent()) return false; 895 896 if (!Arg->isIntegerConstantExpr(Result, Context)) 897 return Diag(TheCall->getLocStart(), diag::err_constant_integer_arg_type) 898 << FDecl->getDeclName() << Arg->getSourceRange(); 899 900 return false; 901} 902 903/// SemaBuiltinObjectSize - Handle __builtin_object_size(void *ptr, 904/// int type). This simply type checks that type is one of the defined 905/// constants (0-3). 906// For compatability check 0-3, llvm only handles 0 and 2. 907bool Sema::SemaBuiltinObjectSize(CallExpr *TheCall) { 908 llvm::APSInt Result; 909 910 // Check constant-ness first. 911 if (SemaBuiltinConstantArg(TheCall, 1, Result)) 912 return true; 913 914 Expr *Arg = TheCall->getArg(1); 915 if (Result.getSExtValue() < 0 || Result.getSExtValue() > 3) { 916 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 917 << "0" << "3" << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 918 } 919 920 return false; 921} 922 923/// SemaBuiltinLongjmp - Handle __builtin_longjmp(void *env[5], int val). 924/// This checks that val is a constant 1. 925bool Sema::SemaBuiltinLongjmp(CallExpr *TheCall) { 926 Expr *Arg = TheCall->getArg(1); 927 llvm::APSInt Result; 928 929 // TODO: This is less than ideal. Overload this to take a value. 930 if (SemaBuiltinConstantArg(TheCall, 1, Result)) 931 return true; 932 933 if (Result != 1) 934 return Diag(TheCall->getLocStart(), diag::err_builtin_longjmp_invalid_val) 935 << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 936 937 return false; 938} 939 940// Handle i > 1 ? "x" : "y", recursivelly 941bool Sema::SemaCheckStringLiteral(const Expr *E, const CallExpr *TheCall, 942 bool HasVAListArg, 943 unsigned format_idx, unsigned firstDataArg, 944 bool isPrintf) { 945 tryAgain: 946 if (E->isTypeDependent() || E->isValueDependent()) 947 return false; 948 949 switch (E->getStmtClass()) { 950 case Stmt::ConditionalOperatorClass: { 951 const ConditionalOperator *C = cast<ConditionalOperator>(E); 952 return SemaCheckStringLiteral(C->getTrueExpr(), TheCall, HasVAListArg, 953 format_idx, firstDataArg, isPrintf) 954 && SemaCheckStringLiteral(C->getRHS(), TheCall, HasVAListArg, 955 format_idx, firstDataArg, isPrintf); 956 } 957 958 case Stmt::IntegerLiteralClass: 959 // Technically -Wformat-nonliteral does not warn about this case. 960 // The behavior of printf and friends in this case is implementation 961 // dependent. Ideally if the format string cannot be null then 962 // it should have a 'nonnull' attribute in the function prototype. 963 return true; 964 965 case Stmt::ImplicitCastExprClass: { 966 E = cast<ImplicitCastExpr>(E)->getSubExpr(); 967 goto tryAgain; 968 } 969 970 case Stmt::ParenExprClass: { 971 E = cast<ParenExpr>(E)->getSubExpr(); 972 goto tryAgain; 973 } 974 975 case Stmt::DeclRefExprClass: { 976 const DeclRefExpr *DR = cast<DeclRefExpr>(E); 977 978 // As an exception, do not flag errors for variables binding to 979 // const string literals. 980 if (const VarDecl *VD = dyn_cast<VarDecl>(DR->getDecl())) { 981 bool isConstant = false; 982 QualType T = DR->getType(); 983 984 if (const ArrayType *AT = Context.getAsArrayType(T)) { 985 isConstant = AT->getElementType().isConstant(Context); 986 } else if (const PointerType *PT = T->getAs<PointerType>()) { 987 isConstant = T.isConstant(Context) && 988 PT->getPointeeType().isConstant(Context); 989 } 990 991 if (isConstant) { 992 if (const Expr *Init = VD->getAnyInitializer()) 993 return SemaCheckStringLiteral(Init, TheCall, 994 HasVAListArg, format_idx, firstDataArg, 995 isPrintf); 996 } 997 998 // For vprintf* functions (i.e., HasVAListArg==true), we add a 999 // special check to see if the format string is a function parameter 1000 // of the function calling the printf function. If the function 1001 // has an attribute indicating it is a printf-like function, then we 1002 // should suppress warnings concerning non-literals being used in a call 1003 // to a vprintf function. For example: 1004 // 1005 // void 1006 // logmessage(char const *fmt __attribute__ (format (printf, 1, 2)), ...){ 1007 // va_list ap; 1008 // va_start(ap, fmt); 1009 // vprintf(fmt, ap); // Do NOT emit a warning about "fmt". 1010 // ... 1011 // 1012 // 1013 // FIXME: We don't have full attribute support yet, so just check to see 1014 // if the argument is a DeclRefExpr that references a parameter. We'll 1015 // add proper support for checking the attribute later. 1016 if (HasVAListArg) 1017 if (isa<ParmVarDecl>(VD)) 1018 return true; 1019 } 1020 1021 return false; 1022 } 1023 1024 case Stmt::CallExprClass: { 1025 const CallExpr *CE = cast<CallExpr>(E); 1026 if (const ImplicitCastExpr *ICE 1027 = dyn_cast<ImplicitCastExpr>(CE->getCallee())) { 1028 if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(ICE->getSubExpr())) { 1029 if (const FunctionDecl *FD = dyn_cast<FunctionDecl>(DRE->getDecl())) { 1030 if (const FormatArgAttr *FA = FD->getAttr<FormatArgAttr>()) { 1031 unsigned ArgIndex = FA->getFormatIdx(); 1032 const Expr *Arg = CE->getArg(ArgIndex - 1); 1033 1034 return SemaCheckStringLiteral(Arg, TheCall, HasVAListArg, 1035 format_idx, firstDataArg, isPrintf); 1036 } 1037 } 1038 } 1039 } 1040 1041 return false; 1042 } 1043 case Stmt::ObjCStringLiteralClass: 1044 case Stmt::StringLiteralClass: { 1045 const StringLiteral *StrE = NULL; 1046 1047 if (const ObjCStringLiteral *ObjCFExpr = dyn_cast<ObjCStringLiteral>(E)) 1048 StrE = ObjCFExpr->getString(); 1049 else 1050 StrE = cast<StringLiteral>(E); 1051 1052 if (StrE) { 1053 CheckFormatString(StrE, E, TheCall, HasVAListArg, format_idx, 1054 firstDataArg, isPrintf); 1055 return true; 1056 } 1057 1058 return false; 1059 } 1060 1061 default: 1062 return false; 1063 } 1064} 1065 1066void 1067Sema::CheckNonNullArguments(const NonNullAttr *NonNull, 1068 const CallExpr *TheCall) { 1069 for (NonNullAttr::args_iterator i = NonNull->args_begin(), 1070 e = NonNull->args_end(); 1071 i != e; ++i) { 1072 const Expr *ArgExpr = TheCall->getArg(*i); 1073 if (ArgExpr->isNullPointerConstant(Context, 1074 Expr::NPC_ValueDependentIsNotNull)) 1075 Diag(TheCall->getCallee()->getLocStart(), diag::warn_null_arg) 1076 << ArgExpr->getSourceRange(); 1077 } 1078} 1079 1080/// CheckPrintfScanfArguments - Check calls to printf and scanf (and similar 1081/// functions) for correct use of format strings. 1082void 1083Sema::CheckPrintfScanfArguments(const CallExpr *TheCall, bool HasVAListArg, 1084 unsigned format_idx, unsigned firstDataArg, 1085 bool isPrintf) { 1086 1087 const Expr *Fn = TheCall->getCallee(); 1088 1089 // The way the format attribute works in GCC, the implicit this argument 1090 // of member functions is counted. However, it doesn't appear in our own 1091 // lists, so decrement format_idx in that case. 1092 if (isa<CXXMemberCallExpr>(TheCall)) { 1093 const CXXMethodDecl *method_decl = 1094 dyn_cast<CXXMethodDecl>(TheCall->getCalleeDecl()); 1095 if (method_decl && method_decl->isInstance()) { 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 1105 // CHECK: printf/scanf-like function is called with no format string. 1106 if (format_idx >= TheCall->getNumArgs()) { 1107 Diag(TheCall->getRParenLoc(), diag::warn_missing_format_string) 1108 << Fn->getSourceRange(); 1109 return; 1110 } 1111 1112 const Expr *OrigFormatExpr = TheCall->getArg(format_idx)->IgnoreParenCasts(); 1113 1114 // CHECK: format string is not a string literal. 1115 // 1116 // Dynamically generated format strings are difficult to 1117 // automatically vet at compile time. Requiring that format strings 1118 // are string literals: (1) permits the checking of format strings by 1119 // the compiler and thereby (2) can practically remove the source of 1120 // many format string exploits. 1121 1122 // Format string can be either ObjC string (e.g. @"%d") or 1123 // C string (e.g. "%d") 1124 // ObjC string uses the same format specifiers as C string, so we can use 1125 // the same format string checking logic for both ObjC and C strings. 1126 if (SemaCheckStringLiteral(OrigFormatExpr, TheCall, HasVAListArg, format_idx, 1127 firstDataArg, isPrintf)) 1128 return; // Literal format string found, check done! 1129 1130 // If there are no arguments specified, warn with -Wformat-security, otherwise 1131 // warn only with -Wformat-nonliteral. 1132 if (TheCall->getNumArgs() == format_idx+1) 1133 Diag(TheCall->getArg(format_idx)->getLocStart(), 1134 diag::warn_format_nonliteral_noargs) 1135 << OrigFormatExpr->getSourceRange(); 1136 else 1137 Diag(TheCall->getArg(format_idx)->getLocStart(), 1138 diag::warn_format_nonliteral) 1139 << OrigFormatExpr->getSourceRange(); 1140} 1141 1142namespace { 1143class CheckFormatHandler : public analyze_format_string::FormatStringHandler { 1144protected: 1145 Sema &S; 1146 const StringLiteral *FExpr; 1147 const Expr *OrigFormatExpr; 1148 const unsigned FirstDataArg; 1149 const unsigned NumDataArgs; 1150 const bool IsObjCLiteral; 1151 const char *Beg; // Start of format string. 1152 const bool HasVAListArg; 1153 const CallExpr *TheCall; 1154 unsigned FormatIdx; 1155 llvm::BitVector CoveredArgs; 1156 bool usesPositionalArgs; 1157 bool atFirstArg; 1158public: 1159 CheckFormatHandler(Sema &s, const StringLiteral *fexpr, 1160 const Expr *origFormatExpr, unsigned firstDataArg, 1161 unsigned numDataArgs, bool isObjCLiteral, 1162 const char *beg, bool hasVAListArg, 1163 const CallExpr *theCall, unsigned formatIdx) 1164 : S(s), FExpr(fexpr), OrigFormatExpr(origFormatExpr), 1165 FirstDataArg(firstDataArg), 1166 NumDataArgs(numDataArgs), 1167 IsObjCLiteral(isObjCLiteral), Beg(beg), 1168 HasVAListArg(hasVAListArg), 1169 TheCall(theCall), FormatIdx(formatIdx), 1170 usesPositionalArgs(false), atFirstArg(true) { 1171 CoveredArgs.resize(numDataArgs); 1172 CoveredArgs.reset(); 1173 } 1174 1175 void DoneProcessing(); 1176 1177 void HandleIncompleteSpecifier(const char *startSpecifier, 1178 unsigned specifierLen); 1179 1180 virtual void HandleInvalidPosition(const char *startSpecifier, 1181 unsigned specifierLen, 1182 analyze_format_string::PositionContext p); 1183 1184 virtual void HandleZeroPosition(const char *startPos, unsigned posLen); 1185 1186 void HandleNullChar(const char *nullCharacter); 1187 1188protected: 1189 bool HandleInvalidConversionSpecifier(unsigned argIndex, SourceLocation Loc, 1190 const char *startSpec, 1191 unsigned specifierLen, 1192 const char *csStart, unsigned csLen); 1193 1194 SourceRange getFormatStringRange(); 1195 CharSourceRange getSpecifierRange(const char *startSpecifier, 1196 unsigned specifierLen); 1197 SourceLocation getLocationOfByte(const char *x); 1198 1199 const Expr *getDataArg(unsigned i) const; 1200 1201 bool CheckNumArgs(const analyze_format_string::FormatSpecifier &FS, 1202 const analyze_format_string::ConversionSpecifier &CS, 1203 const char *startSpecifier, unsigned specifierLen, 1204 unsigned argIndex); 1205}; 1206} 1207 1208SourceRange CheckFormatHandler::getFormatStringRange() { 1209 return OrigFormatExpr->getSourceRange(); 1210} 1211 1212CharSourceRange CheckFormatHandler:: 1213getSpecifierRange(const char *startSpecifier, unsigned specifierLen) { 1214 SourceLocation Start = getLocationOfByte(startSpecifier); 1215 SourceLocation End = getLocationOfByte(startSpecifier + specifierLen - 1); 1216 1217 // Advance the end SourceLocation by one due to half-open ranges. 1218 End = End.getFileLocWithOffset(1); 1219 1220 return CharSourceRange::getCharRange(Start, End); 1221} 1222 1223SourceLocation CheckFormatHandler::getLocationOfByte(const char *x) { 1224 return S.getLocationOfStringLiteralByte(FExpr, x - Beg); 1225} 1226 1227void CheckFormatHandler::HandleIncompleteSpecifier(const char *startSpecifier, 1228 unsigned specifierLen){ 1229 SourceLocation Loc = getLocationOfByte(startSpecifier); 1230 S.Diag(Loc, diag::warn_printf_incomplete_specifier) 1231 << getSpecifierRange(startSpecifier, specifierLen); 1232} 1233 1234void 1235CheckFormatHandler::HandleInvalidPosition(const char *startPos, unsigned posLen, 1236 analyze_format_string::PositionContext p) { 1237 SourceLocation Loc = getLocationOfByte(startPos); 1238 S.Diag(Loc, diag::warn_format_invalid_positional_specifier) 1239 << (unsigned) p << getSpecifierRange(startPos, posLen); 1240} 1241 1242void CheckFormatHandler::HandleZeroPosition(const char *startPos, 1243 unsigned posLen) { 1244 SourceLocation Loc = getLocationOfByte(startPos); 1245 S.Diag(Loc, diag::warn_format_zero_positional_specifier) 1246 << getSpecifierRange(startPos, posLen); 1247} 1248 1249void CheckFormatHandler::HandleNullChar(const char *nullCharacter) { 1250 // The presence of a null character is likely an error. 1251 S.Diag(getLocationOfByte(nullCharacter), 1252 diag::warn_printf_format_string_contains_null_char) 1253 << getFormatStringRange(); 1254} 1255 1256const Expr *CheckFormatHandler::getDataArg(unsigned i) const { 1257 return TheCall->getArg(FirstDataArg + i); 1258} 1259 1260void CheckFormatHandler::DoneProcessing() { 1261 // Does the number of data arguments exceed the number of 1262 // format conversions in the format string? 1263 if (!HasVAListArg) { 1264 // Find any arguments that weren't covered. 1265 CoveredArgs.flip(); 1266 signed notCoveredArg = CoveredArgs.find_first(); 1267 if (notCoveredArg >= 0) { 1268 assert((unsigned)notCoveredArg < NumDataArgs); 1269 S.Diag(getDataArg((unsigned) notCoveredArg)->getLocStart(), 1270 diag::warn_printf_data_arg_not_used) 1271 << getFormatStringRange(); 1272 } 1273 } 1274} 1275 1276bool 1277CheckFormatHandler::HandleInvalidConversionSpecifier(unsigned argIndex, 1278 SourceLocation Loc, 1279 const char *startSpec, 1280 unsigned specifierLen, 1281 const char *csStart, 1282 unsigned csLen) { 1283 1284 bool keepGoing = true; 1285 if (argIndex < NumDataArgs) { 1286 // Consider the argument coverered, even though the specifier doesn't 1287 // make sense. 1288 CoveredArgs.set(argIndex); 1289 } 1290 else { 1291 // If argIndex exceeds the number of data arguments we 1292 // don't issue a warning because that is just a cascade of warnings (and 1293 // they may have intended '%%' anyway). We don't want to continue processing 1294 // the format string after this point, however, as we will like just get 1295 // gibberish when trying to match arguments. 1296 keepGoing = false; 1297 } 1298 1299 S.Diag(Loc, diag::warn_format_invalid_conversion) 1300 << llvm::StringRef(csStart, csLen) 1301 << getSpecifierRange(startSpec, specifierLen); 1302 1303 return keepGoing; 1304} 1305 1306bool 1307CheckFormatHandler::CheckNumArgs( 1308 const analyze_format_string::FormatSpecifier &FS, 1309 const analyze_format_string::ConversionSpecifier &CS, 1310 const char *startSpecifier, unsigned specifierLen, unsigned argIndex) { 1311 1312 if (argIndex >= NumDataArgs) { 1313 if (FS.usesPositionalArg()) { 1314 S.Diag(getLocationOfByte(CS.getStart()), 1315 diag::warn_printf_positional_arg_exceeds_data_args) 1316 << (argIndex+1) << NumDataArgs 1317 << getSpecifierRange(startSpecifier, specifierLen); 1318 } 1319 else { 1320 S.Diag(getLocationOfByte(CS.getStart()), 1321 diag::warn_printf_insufficient_data_args) 1322 << getSpecifierRange(startSpecifier, specifierLen); 1323 } 1324 1325 return false; 1326 } 1327 return true; 1328} 1329 1330//===--- CHECK: Printf format string checking ------------------------------===// 1331 1332namespace { 1333class CheckPrintfHandler : public CheckFormatHandler { 1334public: 1335 CheckPrintfHandler(Sema &s, const StringLiteral *fexpr, 1336 const Expr *origFormatExpr, unsigned firstDataArg, 1337 unsigned numDataArgs, bool isObjCLiteral, 1338 const char *beg, bool hasVAListArg, 1339 const CallExpr *theCall, unsigned formatIdx) 1340 : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg, 1341 numDataArgs, isObjCLiteral, beg, hasVAListArg, 1342 theCall, formatIdx) {} 1343 1344 1345 bool HandleInvalidPrintfConversionSpecifier( 1346 const analyze_printf::PrintfSpecifier &FS, 1347 const char *startSpecifier, 1348 unsigned specifierLen); 1349 1350 bool HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier &FS, 1351 const char *startSpecifier, 1352 unsigned specifierLen); 1353 1354 bool HandleAmount(const analyze_format_string::OptionalAmount &Amt, unsigned k, 1355 const char *startSpecifier, unsigned specifierLen); 1356 void HandleInvalidAmount(const analyze_printf::PrintfSpecifier &FS, 1357 const analyze_printf::OptionalAmount &Amt, 1358 unsigned type, 1359 const char *startSpecifier, unsigned specifierLen); 1360 void HandleFlag(const analyze_printf::PrintfSpecifier &FS, 1361 const analyze_printf::OptionalFlag &flag, 1362 const char *startSpecifier, unsigned specifierLen); 1363 void HandleIgnoredFlag(const analyze_printf::PrintfSpecifier &FS, 1364 const analyze_printf::OptionalFlag &ignoredFlag, 1365 const analyze_printf::OptionalFlag &flag, 1366 const char *startSpecifier, unsigned specifierLen); 1367}; 1368} 1369 1370bool CheckPrintfHandler::HandleInvalidPrintfConversionSpecifier( 1371 const analyze_printf::PrintfSpecifier &FS, 1372 const char *startSpecifier, 1373 unsigned specifierLen) { 1374 const analyze_printf::PrintfConversionSpecifier &CS = 1375 FS.getConversionSpecifier(); 1376 1377 return HandleInvalidConversionSpecifier(FS.getArgIndex(), 1378 getLocationOfByte(CS.getStart()), 1379 startSpecifier, specifierLen, 1380 CS.getStart(), CS.getLength()); 1381} 1382 1383bool CheckPrintfHandler::HandleAmount( 1384 const analyze_format_string::OptionalAmount &Amt, 1385 unsigned k, const char *startSpecifier, 1386 unsigned specifierLen) { 1387 1388 if (Amt.hasDataArgument()) { 1389 if (!HasVAListArg) { 1390 unsigned argIndex = Amt.getArgIndex(); 1391 if (argIndex >= NumDataArgs) { 1392 S.Diag(getLocationOfByte(Amt.getStart()), 1393 diag::warn_printf_asterisk_missing_arg) 1394 << k << getSpecifierRange(startSpecifier, specifierLen); 1395 // Don't do any more checking. We will just emit 1396 // spurious errors. 1397 return false; 1398 } 1399 1400 // Type check the data argument. It should be an 'int'. 1401 // Although not in conformance with C99, we also allow the argument to be 1402 // an 'unsigned int' as that is a reasonably safe case. GCC also 1403 // doesn't emit a warning for that case. 1404 CoveredArgs.set(argIndex); 1405 const Expr *Arg = getDataArg(argIndex); 1406 QualType T = Arg->getType(); 1407 1408 const analyze_printf::ArgTypeResult &ATR = Amt.getArgType(S.Context); 1409 assert(ATR.isValid()); 1410 1411 if (!ATR.matchesType(S.Context, T)) { 1412 S.Diag(getLocationOfByte(Amt.getStart()), 1413 diag::warn_printf_asterisk_wrong_type) 1414 << k 1415 << ATR.getRepresentativeType(S.Context) << T 1416 << getSpecifierRange(startSpecifier, specifierLen) 1417 << Arg->getSourceRange(); 1418 // Don't do any more checking. We will just emit 1419 // spurious errors. 1420 return false; 1421 } 1422 } 1423 } 1424 return true; 1425} 1426 1427void CheckPrintfHandler::HandleInvalidAmount( 1428 const analyze_printf::PrintfSpecifier &FS, 1429 const analyze_printf::OptionalAmount &Amt, 1430 unsigned type, 1431 const char *startSpecifier, 1432 unsigned specifierLen) { 1433 const analyze_printf::PrintfConversionSpecifier &CS = 1434 FS.getConversionSpecifier(); 1435 switch (Amt.getHowSpecified()) { 1436 case analyze_printf::OptionalAmount::Constant: 1437 S.Diag(getLocationOfByte(Amt.getStart()), 1438 diag::warn_printf_nonsensical_optional_amount) 1439 << type 1440 << CS.toString() 1441 << getSpecifierRange(startSpecifier, specifierLen) 1442 << FixItHint::CreateRemoval(getSpecifierRange(Amt.getStart(), 1443 Amt.getConstantLength())); 1444 break; 1445 1446 default: 1447 S.Diag(getLocationOfByte(Amt.getStart()), 1448 diag::warn_printf_nonsensical_optional_amount) 1449 << type 1450 << CS.toString() 1451 << getSpecifierRange(startSpecifier, specifierLen); 1452 break; 1453 } 1454} 1455 1456void CheckPrintfHandler::HandleFlag(const analyze_printf::PrintfSpecifier &FS, 1457 const analyze_printf::OptionalFlag &flag, 1458 const char *startSpecifier, 1459 unsigned specifierLen) { 1460 // Warn about pointless flag with a fixit removal. 1461 const analyze_printf::PrintfConversionSpecifier &CS = 1462 FS.getConversionSpecifier(); 1463 S.Diag(getLocationOfByte(flag.getPosition()), 1464 diag::warn_printf_nonsensical_flag) 1465 << flag.toString() << CS.toString() 1466 << getSpecifierRange(startSpecifier, specifierLen) 1467 << FixItHint::CreateRemoval(getSpecifierRange(flag.getPosition(), 1)); 1468} 1469 1470void CheckPrintfHandler::HandleIgnoredFlag( 1471 const analyze_printf::PrintfSpecifier &FS, 1472 const analyze_printf::OptionalFlag &ignoredFlag, 1473 const analyze_printf::OptionalFlag &flag, 1474 const char *startSpecifier, 1475 unsigned specifierLen) { 1476 // Warn about ignored flag with a fixit removal. 1477 S.Diag(getLocationOfByte(ignoredFlag.getPosition()), 1478 diag::warn_printf_ignored_flag) 1479 << ignoredFlag.toString() << flag.toString() 1480 << getSpecifierRange(startSpecifier, specifierLen) 1481 << FixItHint::CreateRemoval(getSpecifierRange( 1482 ignoredFlag.getPosition(), 1)); 1483} 1484 1485bool 1486CheckPrintfHandler::HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier 1487 &FS, 1488 const char *startSpecifier, 1489 unsigned specifierLen) { 1490 1491 using namespace analyze_format_string; 1492 using namespace analyze_printf; 1493 const PrintfConversionSpecifier &CS = FS.getConversionSpecifier(); 1494 1495 if (FS.consumesDataArgument()) { 1496 if (atFirstArg) { 1497 atFirstArg = false; 1498 usesPositionalArgs = FS.usesPositionalArg(); 1499 } 1500 else if (usesPositionalArgs != FS.usesPositionalArg()) { 1501 // Cannot mix-and-match positional and non-positional arguments. 1502 S.Diag(getLocationOfByte(CS.getStart()), 1503 diag::warn_format_mix_positional_nonpositional_args) 1504 << getSpecifierRange(startSpecifier, specifierLen); 1505 return false; 1506 } 1507 } 1508 1509 // First check if the field width, precision, and conversion specifier 1510 // have matching data arguments. 1511 if (!HandleAmount(FS.getFieldWidth(), /* field width */ 0, 1512 startSpecifier, specifierLen)) { 1513 return false; 1514 } 1515 1516 if (!HandleAmount(FS.getPrecision(), /* precision */ 1, 1517 startSpecifier, specifierLen)) { 1518 return false; 1519 } 1520 1521 if (!CS.consumesDataArgument()) { 1522 // FIXME: Technically specifying a precision or field width here 1523 // makes no sense. Worth issuing a warning at some point. 1524 return true; 1525 } 1526 1527 // Consume the argument. 1528 unsigned argIndex = FS.getArgIndex(); 1529 if (argIndex < NumDataArgs) { 1530 // The check to see if the argIndex is valid will come later. 1531 // We set the bit here because we may exit early from this 1532 // function if we encounter some other error. 1533 CoveredArgs.set(argIndex); 1534 } 1535 1536 // Check for using an Objective-C specific conversion specifier 1537 // in a non-ObjC literal. 1538 if (!IsObjCLiteral && CS.isObjCArg()) { 1539 return HandleInvalidPrintfConversionSpecifier(FS, startSpecifier, 1540 specifierLen); 1541 } 1542 1543 // Check for invalid use of field width 1544 if (!FS.hasValidFieldWidth()) { 1545 HandleInvalidAmount(FS, FS.getFieldWidth(), /* field width */ 0, 1546 startSpecifier, specifierLen); 1547 } 1548 1549 // Check for invalid use of precision 1550 if (!FS.hasValidPrecision()) { 1551 HandleInvalidAmount(FS, FS.getPrecision(), /* precision */ 1, 1552 startSpecifier, specifierLen); 1553 } 1554 1555 // Check each flag does not conflict with any other component. 1556 if (!FS.hasValidLeadingZeros()) 1557 HandleFlag(FS, FS.hasLeadingZeros(), startSpecifier, specifierLen); 1558 if (!FS.hasValidPlusPrefix()) 1559 HandleFlag(FS, FS.hasPlusPrefix(), startSpecifier, specifierLen); 1560 if (!FS.hasValidSpacePrefix()) 1561 HandleFlag(FS, FS.hasSpacePrefix(), startSpecifier, specifierLen); 1562 if (!FS.hasValidAlternativeForm()) 1563 HandleFlag(FS, FS.hasAlternativeForm(), startSpecifier, specifierLen); 1564 if (!FS.hasValidLeftJustified()) 1565 HandleFlag(FS, FS.isLeftJustified(), startSpecifier, specifierLen); 1566 1567 // Check that flags are not ignored by another flag 1568 if (FS.hasSpacePrefix() && FS.hasPlusPrefix()) // ' ' ignored by '+' 1569 HandleIgnoredFlag(FS, FS.hasSpacePrefix(), FS.hasPlusPrefix(), 1570 startSpecifier, specifierLen); 1571 if (FS.hasLeadingZeros() && FS.isLeftJustified()) // '0' ignored by '-' 1572 HandleIgnoredFlag(FS, FS.hasLeadingZeros(), FS.isLeftJustified(), 1573 startSpecifier, specifierLen); 1574 1575 // Check the length modifier is valid with the given conversion specifier. 1576 const LengthModifier &LM = FS.getLengthModifier(); 1577 if (!FS.hasValidLengthModifier()) 1578 S.Diag(getLocationOfByte(LM.getStart()), 1579 diag::warn_format_nonsensical_length) 1580 << LM.toString() << CS.toString() 1581 << getSpecifierRange(startSpecifier, specifierLen) 1582 << FixItHint::CreateRemoval(getSpecifierRange(LM.getStart(), 1583 LM.getLength())); 1584 1585 // Are we using '%n'? 1586 if (CS.getKind() == ConversionSpecifier::nArg) { 1587 // Issue a warning about this being a possible security issue. 1588 S.Diag(getLocationOfByte(CS.getStart()), diag::warn_printf_write_back) 1589 << getSpecifierRange(startSpecifier, specifierLen); 1590 // Continue checking the other format specifiers. 1591 return true; 1592 } 1593 1594 // The remaining checks depend on the data arguments. 1595 if (HasVAListArg) 1596 return true; 1597 1598 if (!CheckNumArgs(FS, CS, startSpecifier, specifierLen, argIndex)) 1599 return false; 1600 1601 // Now type check the data expression that matches the 1602 // format specifier. 1603 const Expr *Ex = getDataArg(argIndex); 1604 const analyze_printf::ArgTypeResult &ATR = FS.getArgType(S.Context); 1605 if (ATR.isValid() && !ATR.matchesType(S.Context, Ex->getType())) { 1606 // Check if we didn't match because of an implicit cast from a 'char' 1607 // or 'short' to an 'int'. This is done because printf is a varargs 1608 // function. 1609 if (const ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(Ex)) 1610 if (ICE->getType() == S.Context.IntTy) { 1611 // All further checking is done on the subexpression. 1612 Ex = ICE->getSubExpr(); 1613 if (ATR.matchesType(S.Context, Ex->getType())) 1614 return true; 1615 } 1616 1617 // We may be able to offer a FixItHint if it is a supported type. 1618 PrintfSpecifier fixedFS = FS; 1619 bool success = fixedFS.fixType(Ex->getType()); 1620 1621 if (success) { 1622 // Get the fix string from the fixed format specifier 1623 llvm::SmallString<128> buf; 1624 llvm::raw_svector_ostream os(buf); 1625 fixedFS.toString(os); 1626 1627 // FIXME: getRepresentativeType() perhaps should return a string 1628 // instead of a QualType to better handle when the representative 1629 // type is 'wint_t' (which is defined in the system headers). 1630 S.Diag(getLocationOfByte(CS.getStart()), 1631 diag::warn_printf_conversion_argument_type_mismatch) 1632 << ATR.getRepresentativeType(S.Context) << Ex->getType() 1633 << getSpecifierRange(startSpecifier, specifierLen) 1634 << Ex->getSourceRange() 1635 << FixItHint::CreateReplacement( 1636 getSpecifierRange(startSpecifier, specifierLen), 1637 os.str()); 1638 } 1639 else { 1640 S.Diag(getLocationOfByte(CS.getStart()), 1641 diag::warn_printf_conversion_argument_type_mismatch) 1642 << ATR.getRepresentativeType(S.Context) << Ex->getType() 1643 << getSpecifierRange(startSpecifier, specifierLen) 1644 << Ex->getSourceRange(); 1645 } 1646 } 1647 1648 return true; 1649} 1650 1651//===--- CHECK: Scanf format string checking ------------------------------===// 1652 1653namespace { 1654class CheckScanfHandler : public CheckFormatHandler { 1655public: 1656 CheckScanfHandler(Sema &s, const StringLiteral *fexpr, 1657 const Expr *origFormatExpr, unsigned firstDataArg, 1658 unsigned numDataArgs, bool isObjCLiteral, 1659 const char *beg, bool hasVAListArg, 1660 const CallExpr *theCall, unsigned formatIdx) 1661 : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg, 1662 numDataArgs, isObjCLiteral, beg, hasVAListArg, 1663 theCall, formatIdx) {} 1664 1665 bool HandleScanfSpecifier(const analyze_scanf::ScanfSpecifier &FS, 1666 const char *startSpecifier, 1667 unsigned specifierLen); 1668 1669 bool HandleInvalidScanfConversionSpecifier( 1670 const analyze_scanf::ScanfSpecifier &FS, 1671 const char *startSpecifier, 1672 unsigned specifierLen); 1673 1674 void HandleIncompleteScanList(const char *start, const char *end); 1675}; 1676} 1677 1678void CheckScanfHandler::HandleIncompleteScanList(const char *start, 1679 const char *end) { 1680 S.Diag(getLocationOfByte(end), diag::warn_scanf_scanlist_incomplete) 1681 << getSpecifierRange(start, end - start); 1682} 1683 1684bool CheckScanfHandler::HandleInvalidScanfConversionSpecifier( 1685 const analyze_scanf::ScanfSpecifier &FS, 1686 const char *startSpecifier, 1687 unsigned specifierLen) { 1688 1689 const analyze_scanf::ScanfConversionSpecifier &CS = 1690 FS.getConversionSpecifier(); 1691 1692 return HandleInvalidConversionSpecifier(FS.getArgIndex(), 1693 getLocationOfByte(CS.getStart()), 1694 startSpecifier, specifierLen, 1695 CS.getStart(), CS.getLength()); 1696} 1697 1698bool CheckScanfHandler::HandleScanfSpecifier( 1699 const analyze_scanf::ScanfSpecifier &FS, 1700 const char *startSpecifier, 1701 unsigned specifierLen) { 1702 1703 using namespace analyze_scanf; 1704 using namespace analyze_format_string; 1705 1706 const ScanfConversionSpecifier &CS = FS.getConversionSpecifier(); 1707 1708 // Handle case where '%' and '*' don't consume an argument. These shouldn't 1709 // be used to decide if we are using positional arguments consistently. 1710 if (FS.consumesDataArgument()) { 1711 if (atFirstArg) { 1712 atFirstArg = false; 1713 usesPositionalArgs = FS.usesPositionalArg(); 1714 } 1715 else if (usesPositionalArgs != FS.usesPositionalArg()) { 1716 // Cannot mix-and-match positional and non-positional arguments. 1717 S.Diag(getLocationOfByte(CS.getStart()), 1718 diag::warn_format_mix_positional_nonpositional_args) 1719 << getSpecifierRange(startSpecifier, specifierLen); 1720 return false; 1721 } 1722 } 1723 1724 // Check if the field with is non-zero. 1725 const OptionalAmount &Amt = FS.getFieldWidth(); 1726 if (Amt.getHowSpecified() == OptionalAmount::Constant) { 1727 if (Amt.getConstantAmount() == 0) { 1728 const CharSourceRange &R = getSpecifierRange(Amt.getStart(), 1729 Amt.getConstantLength()); 1730 S.Diag(getLocationOfByte(Amt.getStart()), 1731 diag::warn_scanf_nonzero_width) 1732 << R << FixItHint::CreateRemoval(R); 1733 } 1734 } 1735 1736 if (!FS.consumesDataArgument()) { 1737 // FIXME: Technically specifying a precision or field width here 1738 // makes no sense. Worth issuing a warning at some point. 1739 return true; 1740 } 1741 1742 // Consume the argument. 1743 unsigned argIndex = FS.getArgIndex(); 1744 if (argIndex < NumDataArgs) { 1745 // The check to see if the argIndex is valid will come later. 1746 // We set the bit here because we may exit early from this 1747 // function if we encounter some other error. 1748 CoveredArgs.set(argIndex); 1749 } 1750 1751 // Check the length modifier is valid with the given conversion specifier. 1752 const LengthModifier &LM = FS.getLengthModifier(); 1753 if (!FS.hasValidLengthModifier()) { 1754 S.Diag(getLocationOfByte(LM.getStart()), 1755 diag::warn_format_nonsensical_length) 1756 << LM.toString() << CS.toString() 1757 << getSpecifierRange(startSpecifier, specifierLen) 1758 << FixItHint::CreateRemoval(getSpecifierRange(LM.getStart(), 1759 LM.getLength())); 1760 } 1761 1762 // The remaining checks depend on the data arguments. 1763 if (HasVAListArg) 1764 return true; 1765 1766 if (!CheckNumArgs(FS, CS, startSpecifier, specifierLen, argIndex)) 1767 return false; 1768 1769 // FIXME: Check that the argument type matches the format specifier. 1770 1771 return true; 1772} 1773 1774void Sema::CheckFormatString(const StringLiteral *FExpr, 1775 const Expr *OrigFormatExpr, 1776 const CallExpr *TheCall, bool HasVAListArg, 1777 unsigned format_idx, unsigned firstDataArg, 1778 bool isPrintf) { 1779 1780 // CHECK: is the format string a wide literal? 1781 if (FExpr->isWide()) { 1782 Diag(FExpr->getLocStart(), 1783 diag::warn_format_string_is_wide_literal) 1784 << OrigFormatExpr->getSourceRange(); 1785 return; 1786 } 1787 1788 // Str - The format string. NOTE: this is NOT null-terminated! 1789 llvm::StringRef StrRef = FExpr->getString(); 1790 const char *Str = StrRef.data(); 1791 unsigned StrLen = StrRef.size(); 1792 1793 // CHECK: empty format string? 1794 if (StrLen == 0) { 1795 Diag(FExpr->getLocStart(), diag::warn_empty_format_string) 1796 << OrigFormatExpr->getSourceRange(); 1797 return; 1798 } 1799 1800 if (isPrintf) { 1801 CheckPrintfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg, 1802 TheCall->getNumArgs() - firstDataArg, 1803 isa<ObjCStringLiteral>(OrigFormatExpr), Str, 1804 HasVAListArg, TheCall, format_idx); 1805 1806 if (!analyze_format_string::ParsePrintfString(H, Str, Str + StrLen)) 1807 H.DoneProcessing(); 1808 } 1809 else { 1810 CheckScanfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg, 1811 TheCall->getNumArgs() - firstDataArg, 1812 isa<ObjCStringLiteral>(OrigFormatExpr), Str, 1813 HasVAListArg, TheCall, format_idx); 1814 1815 if (!analyze_format_string::ParseScanfString(H, Str, Str + StrLen)) 1816 H.DoneProcessing(); 1817 } 1818} 1819 1820//===--- CHECK: Return Address of Stack Variable --------------------------===// 1821 1822static DeclRefExpr* EvalVal(Expr *E); 1823static DeclRefExpr* EvalAddr(Expr* E); 1824 1825/// CheckReturnStackAddr - Check if a return statement returns the address 1826/// of a stack variable. 1827void 1828Sema::CheckReturnStackAddr(Expr *RetValExp, QualType lhsType, 1829 SourceLocation ReturnLoc) { 1830 1831 // Perform checking for returned stack addresses. 1832 if (lhsType->isPointerType() || lhsType->isBlockPointerType()) { 1833 if (DeclRefExpr *DR = EvalAddr(RetValExp)) 1834 Diag(DR->getLocStart(), diag::warn_ret_stack_addr) 1835 << DR->getDecl()->getDeclName() << RetValExp->getSourceRange(); 1836 1837 // Skip over implicit cast expressions when checking for block expressions. 1838 RetValExp = RetValExp->IgnoreParenCasts(); 1839 1840 if (BlockExpr *C = dyn_cast<BlockExpr>(RetValExp)) 1841 if (C->hasBlockDeclRefExprs()) 1842 Diag(C->getLocStart(), diag::err_ret_local_block) 1843 << C->getSourceRange(); 1844 1845 if (AddrLabelExpr *ALE = dyn_cast<AddrLabelExpr>(RetValExp)) 1846 Diag(ALE->getLocStart(), diag::warn_ret_addr_label) 1847 << ALE->getSourceRange(); 1848 1849 } else if (lhsType->isReferenceType()) { 1850 // Perform checking for stack values returned by reference. 1851 // Check for a reference to the stack 1852 if (DeclRefExpr *DR = EvalVal(RetValExp)) 1853 Diag(DR->getLocStart(), diag::warn_ret_stack_ref) 1854 << DR->getDecl()->getDeclName() << RetValExp->getSourceRange(); 1855 } 1856} 1857 1858/// EvalAddr - EvalAddr and EvalVal are mutually recursive functions that 1859/// check if the expression in a return statement evaluates to an address 1860/// to a location on the stack. The recursion is used to traverse the 1861/// AST of the return expression, with recursion backtracking when we 1862/// encounter a subexpression that (1) clearly does not lead to the address 1863/// of a stack variable or (2) is something we cannot determine leads to 1864/// the address of a stack variable based on such local checking. 1865/// 1866/// EvalAddr processes expressions that are pointers that are used as 1867/// references (and not L-values). EvalVal handles all other values. 1868/// At the base case of the recursion is a check for a DeclRefExpr* in 1869/// the refers to a stack variable. 1870/// 1871/// This implementation handles: 1872/// 1873/// * pointer-to-pointer casts 1874/// * implicit conversions from array references to pointers 1875/// * taking the address of fields 1876/// * arbitrary interplay between "&" and "*" operators 1877/// * pointer arithmetic from an address of a stack variable 1878/// * taking the address of an array element where the array is on the stack 1879static DeclRefExpr* EvalAddr(Expr *E) { 1880 // We should only be called for evaluating pointer expressions. 1881 assert((E->getType()->isAnyPointerType() || 1882 E->getType()->isBlockPointerType() || 1883 E->getType()->isObjCQualifiedIdType()) && 1884 "EvalAddr only works on pointers"); 1885 1886 // Our "symbolic interpreter" is just a dispatch off the currently 1887 // viewed AST node. We then recursively traverse the AST by calling 1888 // EvalAddr and EvalVal appropriately. 1889 switch (E->getStmtClass()) { 1890 case Stmt::ParenExprClass: 1891 // Ignore parentheses. 1892 return EvalAddr(cast<ParenExpr>(E)->getSubExpr()); 1893 1894 case Stmt::UnaryOperatorClass: { 1895 // The only unary operator that make sense to handle here 1896 // is AddrOf. All others don't make sense as pointers. 1897 UnaryOperator *U = cast<UnaryOperator>(E); 1898 1899 if (U->getOpcode() == UO_AddrOf) 1900 return EvalVal(U->getSubExpr()); 1901 else 1902 return NULL; 1903 } 1904 1905 case Stmt::BinaryOperatorClass: { 1906 // Handle pointer arithmetic. All other binary operators are not valid 1907 // in this context. 1908 BinaryOperator *B = cast<BinaryOperator>(E); 1909 BinaryOperatorKind op = B->getOpcode(); 1910 1911 if (op != BO_Add && op != BO_Sub) 1912 return NULL; 1913 1914 Expr *Base = B->getLHS(); 1915 1916 // Determine which argument is the real pointer base. It could be 1917 // the RHS argument instead of the LHS. 1918 if (!Base->getType()->isPointerType()) Base = B->getRHS(); 1919 1920 assert (Base->getType()->isPointerType()); 1921 return EvalAddr(Base); 1922 } 1923 1924 // For conditional operators we need to see if either the LHS or RHS are 1925 // valid DeclRefExpr*s. If one of them is valid, we return it. 1926 case Stmt::ConditionalOperatorClass: { 1927 ConditionalOperator *C = cast<ConditionalOperator>(E); 1928 1929 // Handle the GNU extension for missing LHS. 1930 if (Expr *lhsExpr = C->getLHS()) { 1931 // In C++, we can have a throw-expression, which has 'void' type. 1932 if (!lhsExpr->getType()->isVoidType()) 1933 if (DeclRefExpr* LHS = EvalAddr(lhsExpr)) 1934 return LHS; 1935 } 1936 1937 // In C++, we can have a throw-expression, which has 'void' type. 1938 if (C->getRHS()->getType()->isVoidType()) 1939 return NULL; 1940 1941 return EvalAddr(C->getRHS()); 1942 } 1943 1944 // For casts, we need to handle conversions from arrays to 1945 // pointer values, and pointer-to-pointer conversions. 1946 case Stmt::ImplicitCastExprClass: 1947 case Stmt::CStyleCastExprClass: 1948 case Stmt::CXXFunctionalCastExprClass: { 1949 Expr* SubExpr = cast<CastExpr>(E)->getSubExpr(); 1950 QualType T = SubExpr->getType(); 1951 1952 if (SubExpr->getType()->isPointerType() || 1953 SubExpr->getType()->isBlockPointerType() || 1954 SubExpr->getType()->isObjCQualifiedIdType()) 1955 return EvalAddr(SubExpr); 1956 else if (T->isArrayType()) 1957 return EvalVal(SubExpr); 1958 else 1959 return 0; 1960 } 1961 1962 // C++ casts. For dynamic casts, static casts, and const casts, we 1963 // are always converting from a pointer-to-pointer, so we just blow 1964 // through the cast. In the case the dynamic cast doesn't fail (and 1965 // return NULL), we take the conservative route and report cases 1966 // where we return the address of a stack variable. For Reinterpre 1967 // FIXME: The comment about is wrong; we're not always converting 1968 // from pointer to pointer. I'm guessing that this code should also 1969 // handle references to objects. 1970 case Stmt::CXXStaticCastExprClass: 1971 case Stmt::CXXDynamicCastExprClass: 1972 case Stmt::CXXConstCastExprClass: 1973 case Stmt::CXXReinterpretCastExprClass: { 1974 Expr *S = cast<CXXNamedCastExpr>(E)->getSubExpr(); 1975 if (S->getType()->isPointerType() || S->getType()->isBlockPointerType()) 1976 return EvalAddr(S); 1977 else 1978 return NULL; 1979 } 1980 1981 // Everything else: we simply don't reason about them. 1982 default: 1983 return NULL; 1984 } 1985} 1986 1987 1988/// EvalVal - This function is complements EvalAddr in the mutual recursion. 1989/// See the comments for EvalAddr for more details. 1990static DeclRefExpr* EvalVal(Expr *E) { 1991do { 1992 // We should only be called for evaluating non-pointer expressions, or 1993 // expressions with a pointer type that are not used as references but instead 1994 // are l-values (e.g., DeclRefExpr with a pointer type). 1995 1996 // Our "symbolic interpreter" is just a dispatch off the currently 1997 // viewed AST node. We then recursively traverse the AST by calling 1998 // EvalAddr and EvalVal appropriately. 1999 switch (E->getStmtClass()) { 2000 case Stmt::ImplicitCastExprClass: { 2001 ImplicitCastExpr *IE = cast<ImplicitCastExpr>(E); 2002 if (IE->getValueKind() == VK_LValue) { 2003 E = IE->getSubExpr(); 2004 continue; 2005 } 2006 return NULL; 2007 } 2008 2009 case Stmt::DeclRefExprClass: { 2010 // DeclRefExpr: the base case. When we hit a DeclRefExpr we are looking 2011 // at code that refers to a variable's name. We check if it has local 2012 // storage within the function, and if so, return the expression. 2013 DeclRefExpr *DR = cast<DeclRefExpr>(E); 2014 2015 if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl())) 2016 if (V->hasLocalStorage() && !V->getType()->isReferenceType()) return DR; 2017 2018 return NULL; 2019 } 2020 2021 case Stmt::ParenExprClass: { 2022 // Ignore parentheses. 2023 E = cast<ParenExpr>(E)->getSubExpr(); 2024 continue; 2025 } 2026 2027 case Stmt::UnaryOperatorClass: { 2028 // The only unary operator that make sense to handle here 2029 // is Deref. All others don't resolve to a "name." This includes 2030 // handling all sorts of rvalues passed to a unary operator. 2031 UnaryOperator *U = cast<UnaryOperator>(E); 2032 2033 if (U->getOpcode() == UO_Deref) 2034 return EvalAddr(U->getSubExpr()); 2035 2036 return NULL; 2037 } 2038 2039 case Stmt::ArraySubscriptExprClass: { 2040 // Array subscripts are potential references to data on the stack. We 2041 // retrieve the DeclRefExpr* for the array variable if it indeed 2042 // has local storage. 2043 return EvalAddr(cast<ArraySubscriptExpr>(E)->getBase()); 2044 } 2045 2046 case Stmt::ConditionalOperatorClass: { 2047 // For conditional operators we need to see if either the LHS or RHS are 2048 // non-NULL DeclRefExpr's. If one is non-NULL, we return it. 2049 ConditionalOperator *C = cast<ConditionalOperator>(E); 2050 2051 // Handle the GNU extension for missing LHS. 2052 if (Expr *lhsExpr = C->getLHS()) 2053 if (DeclRefExpr *LHS = EvalVal(lhsExpr)) 2054 return LHS; 2055 2056 return EvalVal(C->getRHS()); 2057 } 2058 2059 // Accesses to members are potential references to data on the stack. 2060 case Stmt::MemberExprClass: { 2061 MemberExpr *M = cast<MemberExpr>(E); 2062 2063 // Check for indirect access. We only want direct field accesses. 2064 if (M->isArrow()) 2065 return NULL; 2066 2067 // Check whether the member type is itself a reference, in which case 2068 // we're not going to refer to the member, but to what the member refers to. 2069 if (M->getMemberDecl()->getType()->isReferenceType()) 2070 return NULL; 2071 2072 return EvalVal(M->getBase()); 2073 } 2074 2075 // Everything else: we simply don't reason about them. 2076 default: 2077 return NULL; 2078 } 2079} while (true); 2080} 2081 2082//===--- CHECK: Floating-Point comparisons (-Wfloat-equal) ---------------===// 2083 2084/// Check for comparisons of floating point operands using != and ==. 2085/// Issue a warning if these are no self-comparisons, as they are not likely 2086/// to do what the programmer intended. 2087void Sema::CheckFloatComparison(SourceLocation loc, Expr* lex, Expr *rex) { 2088 bool EmitWarning = true; 2089 2090 Expr* LeftExprSansParen = lex->IgnoreParens(); 2091 Expr* RightExprSansParen = rex->IgnoreParens(); 2092 2093 // Special case: check for x == x (which is OK). 2094 // Do not emit warnings for such cases. 2095 if (DeclRefExpr* DRL = dyn_cast<DeclRefExpr>(LeftExprSansParen)) 2096 if (DeclRefExpr* DRR = dyn_cast<DeclRefExpr>(RightExprSansParen)) 2097 if (DRL->getDecl() == DRR->getDecl()) 2098 EmitWarning = false; 2099 2100 2101 // Special case: check for comparisons against literals that can be exactly 2102 // represented by APFloat. In such cases, do not emit a warning. This 2103 // is a heuristic: often comparison against such literals are used to 2104 // detect if a value in a variable has not changed. This clearly can 2105 // lead to false negatives. 2106 if (EmitWarning) { 2107 if (FloatingLiteral* FLL = dyn_cast<FloatingLiteral>(LeftExprSansParen)) { 2108 if (FLL->isExact()) 2109 EmitWarning = false; 2110 } else 2111 if (FloatingLiteral* FLR = dyn_cast<FloatingLiteral>(RightExprSansParen)){ 2112 if (FLR->isExact()) 2113 EmitWarning = false; 2114 } 2115 } 2116 2117 // Check for comparisons with builtin types. 2118 if (EmitWarning) 2119 if (CallExpr* CL = dyn_cast<CallExpr>(LeftExprSansParen)) 2120 if (CL->isBuiltinCall(Context)) 2121 EmitWarning = false; 2122 2123 if (EmitWarning) 2124 if (CallExpr* CR = dyn_cast<CallExpr>(RightExprSansParen)) 2125 if (CR->isBuiltinCall(Context)) 2126 EmitWarning = false; 2127 2128 // Emit the diagnostic. 2129 if (EmitWarning) 2130 Diag(loc, diag::warn_floatingpoint_eq) 2131 << lex->getSourceRange() << rex->getSourceRange(); 2132} 2133 2134//===--- CHECK: Integer mixed-sign comparisons (-Wsign-compare) --------===// 2135//===--- CHECK: Lossy implicit conversions (-Wconversion) --------------===// 2136 2137namespace { 2138 2139/// Structure recording the 'active' range of an integer-valued 2140/// expression. 2141struct IntRange { 2142 /// The number of bits active in the int. 2143 unsigned Width; 2144 2145 /// True if the int is known not to have negative values. 2146 bool NonNegative; 2147 2148 IntRange(unsigned Width, bool NonNegative) 2149 : Width(Width), NonNegative(NonNegative) 2150 {} 2151 2152 /// Returns the range of the bool type. 2153 static IntRange forBoolType() { 2154 return IntRange(1, true); 2155 } 2156 2157 /// Returns the range of an opaque value of the given integral type. 2158 static IntRange forValueOfType(ASTContext &C, QualType T) { 2159 return forValueOfCanonicalType(C, 2160 T->getCanonicalTypeInternal().getTypePtr()); 2161 } 2162 2163 /// Returns the range of an opaque value of a canonical integral type. 2164 static IntRange forValueOfCanonicalType(ASTContext &C, const Type *T) { 2165 assert(T->isCanonicalUnqualified()); 2166 2167 if (const VectorType *VT = dyn_cast<VectorType>(T)) 2168 T = VT->getElementType().getTypePtr(); 2169 if (const ComplexType *CT = dyn_cast<ComplexType>(T)) 2170 T = CT->getElementType().getTypePtr(); 2171 2172 // For enum types, use the known bit width of the enumerators. 2173 if (const EnumType *ET = dyn_cast<EnumType>(T)) { 2174 EnumDecl *Enum = ET->getDecl(); 2175 if (!Enum->isDefinition()) 2176 return IntRange(C.getIntWidth(QualType(T, 0)), false); 2177 2178 unsigned NumPositive = Enum->getNumPositiveBits(); 2179 unsigned NumNegative = Enum->getNumNegativeBits(); 2180 2181 return IntRange(std::max(NumPositive, NumNegative), NumNegative == 0); 2182 } 2183 2184 const BuiltinType *BT = cast<BuiltinType>(T); 2185 assert(BT->isInteger()); 2186 2187 return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger()); 2188 } 2189 2190 /// Returns the "target" range of a canonical integral type, i.e. 2191 /// the range of values expressible in the type. 2192 /// 2193 /// This matches forValueOfCanonicalType except that enums have the 2194 /// full range of their type, not the range of their enumerators. 2195 static IntRange forTargetOfCanonicalType(ASTContext &C, const Type *T) { 2196 assert(T->isCanonicalUnqualified()); 2197 2198 if (const VectorType *VT = dyn_cast<VectorType>(T)) 2199 T = VT->getElementType().getTypePtr(); 2200 if (const ComplexType *CT = dyn_cast<ComplexType>(T)) 2201 T = CT->getElementType().getTypePtr(); 2202 if (const EnumType *ET = dyn_cast<EnumType>(T)) 2203 T = ET->getDecl()->getIntegerType().getTypePtr(); 2204 2205 const BuiltinType *BT = cast<BuiltinType>(T); 2206 assert(BT->isInteger()); 2207 2208 return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger()); 2209 } 2210 2211 /// Returns the supremum of two ranges: i.e. their conservative merge. 2212 static IntRange join(IntRange L, IntRange R) { 2213 return IntRange(std::max(L.Width, R.Width), 2214 L.NonNegative && R.NonNegative); 2215 } 2216 2217 /// Returns the infinum of two ranges: i.e. their aggressive merge. 2218 static IntRange meet(IntRange L, IntRange R) { 2219 return IntRange(std::min(L.Width, R.Width), 2220 L.NonNegative || R.NonNegative); 2221 } 2222}; 2223 2224IntRange GetValueRange(ASTContext &C, llvm::APSInt &value, unsigned MaxWidth) { 2225 if (value.isSigned() && value.isNegative()) 2226 return IntRange(value.getMinSignedBits(), false); 2227 2228 if (value.getBitWidth() > MaxWidth) 2229 value.trunc(MaxWidth); 2230 2231 // isNonNegative() just checks the sign bit without considering 2232 // signedness. 2233 return IntRange(value.getActiveBits(), true); 2234} 2235 2236IntRange GetValueRange(ASTContext &C, APValue &result, QualType Ty, 2237 unsigned MaxWidth) { 2238 if (result.isInt()) 2239 return GetValueRange(C, result.getInt(), MaxWidth); 2240 2241 if (result.isVector()) { 2242 IntRange R = GetValueRange(C, result.getVectorElt(0), Ty, MaxWidth); 2243 for (unsigned i = 1, e = result.getVectorLength(); i != e; ++i) { 2244 IntRange El = GetValueRange(C, result.getVectorElt(i), Ty, MaxWidth); 2245 R = IntRange::join(R, El); 2246 } 2247 return R; 2248 } 2249 2250 if (result.isComplexInt()) { 2251 IntRange R = GetValueRange(C, result.getComplexIntReal(), MaxWidth); 2252 IntRange I = GetValueRange(C, result.getComplexIntImag(), MaxWidth); 2253 return IntRange::join(R, I); 2254 } 2255 2256 // This can happen with lossless casts to intptr_t of "based" lvalues. 2257 // Assume it might use arbitrary bits. 2258 // FIXME: The only reason we need to pass the type in here is to get 2259 // the sign right on this one case. It would be nice if APValue 2260 // preserved this. 2261 assert(result.isLValue()); 2262 return IntRange(MaxWidth, Ty->isUnsignedIntegerType()); 2263} 2264 2265/// Pseudo-evaluate the given integer expression, estimating the 2266/// range of values it might take. 2267/// 2268/// \param MaxWidth - the width to which the value will be truncated 2269IntRange GetExprRange(ASTContext &C, Expr *E, unsigned MaxWidth) { 2270 E = E->IgnoreParens(); 2271 2272 // Try a full evaluation first. 2273 Expr::EvalResult result; 2274 if (E->Evaluate(result, C)) 2275 return GetValueRange(C, result.Val, E->getType(), MaxWidth); 2276 2277 // I think we only want to look through implicit casts here; if the 2278 // user has an explicit widening cast, we should treat the value as 2279 // being of the new, wider type. 2280 if (ImplicitCastExpr *CE = dyn_cast<ImplicitCastExpr>(E)) { 2281 if (CE->getCastKind() == CK_NoOp) 2282 return GetExprRange(C, CE->getSubExpr(), MaxWidth); 2283 2284 IntRange OutputTypeRange = IntRange::forValueOfType(C, CE->getType()); 2285 2286 bool isIntegerCast = (CE->getCastKind() == CK_IntegralCast); 2287 2288 // Assume that non-integer casts can span the full range of the type. 2289 if (!isIntegerCast) 2290 return OutputTypeRange; 2291 2292 IntRange SubRange 2293 = GetExprRange(C, CE->getSubExpr(), 2294 std::min(MaxWidth, OutputTypeRange.Width)); 2295 2296 // Bail out if the subexpr's range is as wide as the cast type. 2297 if (SubRange.Width >= OutputTypeRange.Width) 2298 return OutputTypeRange; 2299 2300 // Otherwise, we take the smaller width, and we're non-negative if 2301 // either the output type or the subexpr is. 2302 return IntRange(SubRange.Width, 2303 SubRange.NonNegative || OutputTypeRange.NonNegative); 2304 } 2305 2306 if (ConditionalOperator *CO = dyn_cast<ConditionalOperator>(E)) { 2307 // If we can fold the condition, just take that operand. 2308 bool CondResult; 2309 if (CO->getCond()->EvaluateAsBooleanCondition(CondResult, C)) 2310 return GetExprRange(C, CondResult ? CO->getTrueExpr() 2311 : CO->getFalseExpr(), 2312 MaxWidth); 2313 2314 // Otherwise, conservatively merge. 2315 IntRange L = GetExprRange(C, CO->getTrueExpr(), MaxWidth); 2316 IntRange R = GetExprRange(C, CO->getFalseExpr(), MaxWidth); 2317 return IntRange::join(L, R); 2318 } 2319 2320 if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) { 2321 switch (BO->getOpcode()) { 2322 2323 // Boolean-valued operations are single-bit and positive. 2324 case BO_LAnd: 2325 case BO_LOr: 2326 case BO_LT: 2327 case BO_GT: 2328 case BO_LE: 2329 case BO_GE: 2330 case BO_EQ: 2331 case BO_NE: 2332 return IntRange::forBoolType(); 2333 2334 // The type of these compound assignments is the type of the LHS, 2335 // so the RHS is not necessarily an integer. 2336 case BO_MulAssign: 2337 case BO_DivAssign: 2338 case BO_RemAssign: 2339 case BO_AddAssign: 2340 case BO_SubAssign: 2341 return IntRange::forValueOfType(C, E->getType()); 2342 2343 // Operations with opaque sources are black-listed. 2344 case BO_PtrMemD: 2345 case BO_PtrMemI: 2346 return IntRange::forValueOfType(C, E->getType()); 2347 2348 // Bitwise-and uses the *infinum* of the two source ranges. 2349 case BO_And: 2350 case BO_AndAssign: 2351 return IntRange::meet(GetExprRange(C, BO->getLHS(), MaxWidth), 2352 GetExprRange(C, BO->getRHS(), MaxWidth)); 2353 2354 // Left shift gets black-listed based on a judgement call. 2355 case BO_Shl: 2356 // ...except that we want to treat '1 << (blah)' as logically 2357 // positive. It's an important idiom. 2358 if (IntegerLiteral *I 2359 = dyn_cast<IntegerLiteral>(BO->getLHS()->IgnoreParenCasts())) { 2360 if (I->getValue() == 1) { 2361 IntRange R = IntRange::forValueOfType(C, E->getType()); 2362 return IntRange(R.Width, /*NonNegative*/ true); 2363 } 2364 } 2365 // fallthrough 2366 2367 case BO_ShlAssign: 2368 return IntRange::forValueOfType(C, E->getType()); 2369 2370 // Right shift by a constant can narrow its left argument. 2371 case BO_Shr: 2372 case BO_ShrAssign: { 2373 IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth); 2374 2375 // If the shift amount is a positive constant, drop the width by 2376 // that much. 2377 llvm::APSInt shift; 2378 if (BO->getRHS()->isIntegerConstantExpr(shift, C) && 2379 shift.isNonNegative()) { 2380 unsigned zext = shift.getZExtValue(); 2381 if (zext >= L.Width) 2382 L.Width = (L.NonNegative ? 0 : 1); 2383 else 2384 L.Width -= zext; 2385 } 2386 2387 return L; 2388 } 2389 2390 // Comma acts as its right operand. 2391 case BO_Comma: 2392 return GetExprRange(C, BO->getRHS(), MaxWidth); 2393 2394 // Black-list pointer subtractions. 2395 case BO_Sub: 2396 if (BO->getLHS()->getType()->isPointerType()) 2397 return IntRange::forValueOfType(C, E->getType()); 2398 // fallthrough 2399 2400 default: 2401 break; 2402 } 2403 2404 // Treat every other operator as if it were closed on the 2405 // narrowest type that encompasses both operands. 2406 IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth); 2407 IntRange R = GetExprRange(C, BO->getRHS(), MaxWidth); 2408 return IntRange::join(L, R); 2409 } 2410 2411 if (UnaryOperator *UO = dyn_cast<UnaryOperator>(E)) { 2412 switch (UO->getOpcode()) { 2413 // Boolean-valued operations are white-listed. 2414 case UO_LNot: 2415 return IntRange::forBoolType(); 2416 2417 // Operations with opaque sources are black-listed. 2418 case UO_Deref: 2419 case UO_AddrOf: // should be impossible 2420 return IntRange::forValueOfType(C, E->getType()); 2421 2422 default: 2423 return GetExprRange(C, UO->getSubExpr(), MaxWidth); 2424 } 2425 } 2426 2427 if (dyn_cast<OffsetOfExpr>(E)) { 2428 IntRange::forValueOfType(C, E->getType()); 2429 } 2430 2431 FieldDecl *BitField = E->getBitField(); 2432 if (BitField) { 2433 llvm::APSInt BitWidthAP = BitField->getBitWidth()->EvaluateAsInt(C); 2434 unsigned BitWidth = BitWidthAP.getZExtValue(); 2435 2436 return IntRange(BitWidth, BitField->getType()->isUnsignedIntegerType()); 2437 } 2438 2439 return IntRange::forValueOfType(C, E->getType()); 2440} 2441 2442IntRange GetExprRange(ASTContext &C, Expr *E) { 2443 return GetExprRange(C, E, C.getIntWidth(E->getType())); 2444} 2445 2446/// Checks whether the given value, which currently has the given 2447/// source semantics, has the same value when coerced through the 2448/// target semantics. 2449bool IsSameFloatAfterCast(const llvm::APFloat &value, 2450 const llvm::fltSemantics &Src, 2451 const llvm::fltSemantics &Tgt) { 2452 llvm::APFloat truncated = value; 2453 2454 bool ignored; 2455 truncated.convert(Src, llvm::APFloat::rmNearestTiesToEven, &ignored); 2456 truncated.convert(Tgt, llvm::APFloat::rmNearestTiesToEven, &ignored); 2457 2458 return truncated.bitwiseIsEqual(value); 2459} 2460 2461/// Checks whether the given value, which currently has the given 2462/// source semantics, has the same value when coerced through the 2463/// target semantics. 2464/// 2465/// The value might be a vector of floats (or a complex number). 2466bool IsSameFloatAfterCast(const APValue &value, 2467 const llvm::fltSemantics &Src, 2468 const llvm::fltSemantics &Tgt) { 2469 if (value.isFloat()) 2470 return IsSameFloatAfterCast(value.getFloat(), Src, Tgt); 2471 2472 if (value.isVector()) { 2473 for (unsigned i = 0, e = value.getVectorLength(); i != e; ++i) 2474 if (!IsSameFloatAfterCast(value.getVectorElt(i), Src, Tgt)) 2475 return false; 2476 return true; 2477 } 2478 2479 assert(value.isComplexFloat()); 2480 return (IsSameFloatAfterCast(value.getComplexFloatReal(), Src, Tgt) && 2481 IsSameFloatAfterCast(value.getComplexFloatImag(), Src, Tgt)); 2482} 2483 2484void AnalyzeImplicitConversions(Sema &S, Expr *E, SourceLocation CC); 2485 2486static bool IsZero(Sema &S, Expr *E) { 2487 // Suppress cases where we are comparing against an enum constant. 2488 if (const DeclRefExpr *DR = 2489 dyn_cast<DeclRefExpr>(E->IgnoreParenImpCasts())) 2490 if (isa<EnumConstantDecl>(DR->getDecl())) 2491 return false; 2492 2493 // Suppress cases where the '0' value is expanded from a macro. 2494 if (E->getLocStart().isMacroID()) 2495 return false; 2496 2497 llvm::APSInt Value; 2498 return E->isIntegerConstantExpr(Value, S.Context) && Value == 0; 2499} 2500 2501static bool HasEnumType(Expr *E) { 2502 // Strip off implicit integral promotions. 2503 while (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(E)) { 2504 if (ICE->getCastKind() != CK_IntegralCast && 2505 ICE->getCastKind() != CK_NoOp) 2506 break; 2507 E = ICE->getSubExpr(); 2508 } 2509 2510 return E->getType()->isEnumeralType(); 2511} 2512 2513void CheckTrivialUnsignedComparison(Sema &S, BinaryOperator *E) { 2514 BinaryOperatorKind op = E->getOpcode(); 2515 if (op == BO_LT && IsZero(S, E->getRHS())) { 2516 S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison) 2517 << "< 0" << "false" << HasEnumType(E->getLHS()) 2518 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 2519 } else if (op == BO_GE && IsZero(S, E->getRHS())) { 2520 S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison) 2521 << ">= 0" << "true" << HasEnumType(E->getLHS()) 2522 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 2523 } else if (op == BO_GT && IsZero(S, E->getLHS())) { 2524 S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison) 2525 << "0 >" << "false" << HasEnumType(E->getRHS()) 2526 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 2527 } else if (op == BO_LE && IsZero(S, E->getLHS())) { 2528 S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison) 2529 << "0 <=" << "true" << HasEnumType(E->getRHS()) 2530 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 2531 } 2532} 2533 2534/// Analyze the operands of the given comparison. Implements the 2535/// fallback case from AnalyzeComparison. 2536void AnalyzeImpConvsInComparison(Sema &S, BinaryOperator *E) { 2537 AnalyzeImplicitConversions(S, E->getLHS(), E->getOperatorLoc()); 2538 AnalyzeImplicitConversions(S, E->getRHS(), E->getOperatorLoc()); 2539} 2540 2541/// \brief Implements -Wsign-compare. 2542/// 2543/// \param lex the left-hand expression 2544/// \param rex the right-hand expression 2545/// \param OpLoc the location of the joining operator 2546/// \param BinOpc binary opcode or 0 2547void AnalyzeComparison(Sema &S, BinaryOperator *E) { 2548 // The type the comparison is being performed in. 2549 QualType T = E->getLHS()->getType(); 2550 assert(S.Context.hasSameUnqualifiedType(T, E->getRHS()->getType()) 2551 && "comparison with mismatched types"); 2552 2553 // We don't do anything special if this isn't an unsigned integral 2554 // comparison: we're only interested in integral comparisons, and 2555 // signed comparisons only happen in cases we don't care to warn about. 2556 if (!T->hasUnsignedIntegerRepresentation()) 2557 return AnalyzeImpConvsInComparison(S, E); 2558 2559 Expr *lex = E->getLHS()->IgnoreParenImpCasts(); 2560 Expr *rex = E->getRHS()->IgnoreParenImpCasts(); 2561 2562 // Check to see if one of the (unmodified) operands is of different 2563 // signedness. 2564 Expr *signedOperand, *unsignedOperand; 2565 if (lex->getType()->hasSignedIntegerRepresentation()) { 2566 assert(!rex->getType()->hasSignedIntegerRepresentation() && 2567 "unsigned comparison between two signed integer expressions?"); 2568 signedOperand = lex; 2569 unsignedOperand = rex; 2570 } else if (rex->getType()->hasSignedIntegerRepresentation()) { 2571 signedOperand = rex; 2572 unsignedOperand = lex; 2573 } else { 2574 CheckTrivialUnsignedComparison(S, E); 2575 return AnalyzeImpConvsInComparison(S, E); 2576 } 2577 2578 // Otherwise, calculate the effective range of the signed operand. 2579 IntRange signedRange = GetExprRange(S.Context, signedOperand); 2580 2581 // Go ahead and analyze implicit conversions in the operands. Note 2582 // that we skip the implicit conversions on both sides. 2583 AnalyzeImplicitConversions(S, lex, E->getOperatorLoc()); 2584 AnalyzeImplicitConversions(S, rex, E->getOperatorLoc()); 2585 2586 // If the signed range is non-negative, -Wsign-compare won't fire, 2587 // but we should still check for comparisons which are always true 2588 // or false. 2589 if (signedRange.NonNegative) 2590 return CheckTrivialUnsignedComparison(S, E); 2591 2592 // For (in)equality comparisons, if the unsigned operand is a 2593 // constant which cannot collide with a overflowed signed operand, 2594 // then reinterpreting the signed operand as unsigned will not 2595 // change the result of the comparison. 2596 if (E->isEqualityOp()) { 2597 unsigned comparisonWidth = S.Context.getIntWidth(T); 2598 IntRange unsignedRange = GetExprRange(S.Context, unsignedOperand); 2599 2600 // We should never be unable to prove that the unsigned operand is 2601 // non-negative. 2602 assert(unsignedRange.NonNegative && "unsigned range includes negative?"); 2603 2604 if (unsignedRange.Width < comparisonWidth) 2605 return; 2606 } 2607 2608 S.Diag(E->getOperatorLoc(), diag::warn_mixed_sign_comparison) 2609 << lex->getType() << rex->getType() 2610 << lex->getSourceRange() << rex->getSourceRange(); 2611} 2612 2613/// Analyzes an attempt to assign the given value to a bitfield. 2614/// 2615/// Returns true if there was something fishy about the attempt. 2616bool AnalyzeBitFieldAssignment(Sema &S, FieldDecl *Bitfield, Expr *Init, 2617 SourceLocation InitLoc) { 2618 assert(Bitfield->isBitField()); 2619 if (Bitfield->isInvalidDecl()) 2620 return false; 2621 2622 // White-list bool bitfields. 2623 if (Bitfield->getType()->isBooleanType()) 2624 return false; 2625 2626 Expr *OriginalInit = Init->IgnoreParenImpCasts(); 2627 2628 llvm::APSInt Width(32); 2629 Expr::EvalResult InitValue; 2630 if (!Bitfield->getBitWidth()->isIntegerConstantExpr(Width, S.Context) || 2631 !OriginalInit->Evaluate(InitValue, S.Context) || 2632 !InitValue.Val.isInt()) 2633 return false; 2634 2635 const llvm::APSInt &Value = InitValue.Val.getInt(); 2636 unsigned OriginalWidth = Value.getBitWidth(); 2637 unsigned FieldWidth = Width.getZExtValue(); 2638 2639 if (OriginalWidth <= FieldWidth) 2640 return false; 2641 2642 llvm::APSInt TruncatedValue = Value; 2643 TruncatedValue.trunc(FieldWidth); 2644 2645 // It's fairly common to write values into signed bitfields 2646 // that, if sign-extended, would end up becoming a different 2647 // value. We don't want to warn about that. 2648 if (Value.isSigned() && Value.isNegative()) 2649 TruncatedValue.sext(OriginalWidth); 2650 else 2651 TruncatedValue.zext(OriginalWidth); 2652 2653 if (Value == TruncatedValue) 2654 return false; 2655 2656 std::string PrettyValue = Value.toString(10); 2657 std::string PrettyTrunc = TruncatedValue.toString(10); 2658 2659 S.Diag(InitLoc, diag::warn_impcast_bitfield_precision_constant) 2660 << PrettyValue << PrettyTrunc << OriginalInit->getType() 2661 << Init->getSourceRange(); 2662 2663 return true; 2664} 2665 2666/// Analyze the given simple or compound assignment for warning-worthy 2667/// operations. 2668void AnalyzeAssignment(Sema &S, BinaryOperator *E) { 2669 // Just recurse on the LHS. 2670 AnalyzeImplicitConversions(S, E->getLHS(), E->getOperatorLoc()); 2671 2672 // We want to recurse on the RHS as normal unless we're assigning to 2673 // a bitfield. 2674 if (FieldDecl *Bitfield = E->getLHS()->getBitField()) { 2675 if (AnalyzeBitFieldAssignment(S, Bitfield, E->getRHS(), 2676 E->getOperatorLoc())) { 2677 // Recurse, ignoring any implicit conversions on the RHS. 2678 return AnalyzeImplicitConversions(S, E->getRHS()->IgnoreParenImpCasts(), 2679 E->getOperatorLoc()); 2680 } 2681 } 2682 2683 AnalyzeImplicitConversions(S, E->getRHS(), E->getOperatorLoc()); 2684} 2685 2686/// Diagnose an implicit cast; purely a helper for CheckImplicitConversion. 2687void DiagnoseImpCast(Sema &S, Expr *E, QualType T, SourceLocation CContext, 2688 unsigned diag) { 2689 S.Diag(E->getExprLoc(), diag) 2690 << E->getType() << T << E->getSourceRange() << SourceRange(CContext); 2691} 2692 2693std::string PrettyPrintInRange(const llvm::APSInt &Value, IntRange Range) { 2694 if (!Range.Width) return "0"; 2695 2696 llvm::APSInt ValueInRange = Value; 2697 ValueInRange.setIsSigned(!Range.NonNegative); 2698 ValueInRange.trunc(Range.Width); 2699 return ValueInRange.toString(10); 2700} 2701 2702void CheckImplicitConversion(Sema &S, Expr *E, QualType T, 2703 SourceLocation CC, bool *ICContext = 0) { 2704 if (E->isTypeDependent() || E->isValueDependent()) return; 2705 2706 const Type *Source = S.Context.getCanonicalType(E->getType()).getTypePtr(); 2707 const Type *Target = S.Context.getCanonicalType(T).getTypePtr(); 2708 if (Source == Target) return; 2709 if (Target->isDependentType()) return; 2710 2711 // If the conversion context location is invalid or instantiated 2712 // from a system macro, don't complain. 2713 if (CC.isInvalid() || 2714 (CC.isMacroID() && S.Context.getSourceManager().isInSystemHeader( 2715 S.Context.getSourceManager().getSpellingLoc(CC)))) 2716 return; 2717 2718 // Never diagnose implicit casts to bool. 2719 if (Target->isSpecificBuiltinType(BuiltinType::Bool)) 2720 return; 2721 2722 // Strip vector types. 2723 if (isa<VectorType>(Source)) { 2724 if (!isa<VectorType>(Target)) 2725 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_vector_scalar); 2726 2727 Source = cast<VectorType>(Source)->getElementType().getTypePtr(); 2728 Target = cast<VectorType>(Target)->getElementType().getTypePtr(); 2729 } 2730 2731 // Strip complex types. 2732 if (isa<ComplexType>(Source)) { 2733 if (!isa<ComplexType>(Target)) 2734 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_complex_scalar); 2735 2736 Source = cast<ComplexType>(Source)->getElementType().getTypePtr(); 2737 Target = cast<ComplexType>(Target)->getElementType().getTypePtr(); 2738 } 2739 2740 const BuiltinType *SourceBT = dyn_cast<BuiltinType>(Source); 2741 const BuiltinType *TargetBT = dyn_cast<BuiltinType>(Target); 2742 2743 // If the source is floating point... 2744 if (SourceBT && SourceBT->isFloatingPoint()) { 2745 // ...and the target is floating point... 2746 if (TargetBT && TargetBT->isFloatingPoint()) { 2747 // ...then warn if we're dropping FP rank. 2748 2749 // Builtin FP kinds are ordered by increasing FP rank. 2750 if (SourceBT->getKind() > TargetBT->getKind()) { 2751 // Don't warn about float constants that are precisely 2752 // representable in the target type. 2753 Expr::EvalResult result; 2754 if (E->Evaluate(result, S.Context)) { 2755 // Value might be a float, a float vector, or a float complex. 2756 if (IsSameFloatAfterCast(result.Val, 2757 S.Context.getFloatTypeSemantics(QualType(TargetBT, 0)), 2758 S.Context.getFloatTypeSemantics(QualType(SourceBT, 0)))) 2759 return; 2760 } 2761 2762 DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_float_precision); 2763 } 2764 return; 2765 } 2766 2767 // If the target is integral, always warn. 2768 if ((TargetBT && TargetBT->isInteger())) 2769 // TODO: don't warn for integer values? 2770 DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_float_integer); 2771 2772 return; 2773 } 2774 2775 if (!Source->isIntegerType() || !Target->isIntegerType()) 2776 return; 2777 2778 IntRange SourceRange = GetExprRange(S.Context, E); 2779 IntRange TargetRange = IntRange::forTargetOfCanonicalType(S.Context, Target); 2780 2781 if (SourceRange.Width > TargetRange.Width) { 2782 // If the source is a constant, use a default-on diagnostic. 2783 // TODO: this should happen for bitfield stores, too. 2784 llvm::APSInt Value(32); 2785 if (E->isIntegerConstantExpr(Value, S.Context)) { 2786 std::string PrettySourceValue = Value.toString(10); 2787 std::string PrettyTargetValue = PrettyPrintInRange(Value, TargetRange); 2788 2789 S.Diag(E->getExprLoc(), diag::warn_impcast_integer_precision_constant) 2790 << PrettySourceValue << PrettyTargetValue 2791 << E->getType() << T << E->getSourceRange() << clang::SourceRange(CC); 2792 return; 2793 } 2794 2795 // People want to build with -Wshorten-64-to-32 and not -Wconversion 2796 // and by god we'll let them. 2797 if (SourceRange.Width == 64 && TargetRange.Width == 32) 2798 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_integer_64_32); 2799 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_integer_precision); 2800 } 2801 2802 if ((TargetRange.NonNegative && !SourceRange.NonNegative) || 2803 (!TargetRange.NonNegative && SourceRange.NonNegative && 2804 SourceRange.Width == TargetRange.Width)) { 2805 unsigned DiagID = diag::warn_impcast_integer_sign; 2806 2807 // Traditionally, gcc has warned about this under -Wsign-compare. 2808 // We also want to warn about it in -Wconversion. 2809 // So if -Wconversion is off, use a completely identical diagnostic 2810 // in the sign-compare group. 2811 // The conditional-checking code will 2812 if (ICContext) { 2813 DiagID = diag::warn_impcast_integer_sign_conditional; 2814 *ICContext = true; 2815 } 2816 2817 return DiagnoseImpCast(S, E, T, CC, DiagID); 2818 } 2819 2820 return; 2821} 2822 2823void CheckConditionalOperator(Sema &S, ConditionalOperator *E, QualType T); 2824 2825void CheckConditionalOperand(Sema &S, Expr *E, QualType T, 2826 SourceLocation CC, bool &ICContext) { 2827 E = E->IgnoreParenImpCasts(); 2828 2829 if (isa<ConditionalOperator>(E)) 2830 return CheckConditionalOperator(S, cast<ConditionalOperator>(E), T); 2831 2832 AnalyzeImplicitConversions(S, E, CC); 2833 if (E->getType() != T) 2834 return CheckImplicitConversion(S, E, T, CC, &ICContext); 2835 return; 2836} 2837 2838void CheckConditionalOperator(Sema &S, ConditionalOperator *E, QualType T) { 2839 SourceLocation CC = E->getQuestionLoc(); 2840 2841 AnalyzeImplicitConversions(S, E->getCond(), CC); 2842 2843 bool Suspicious = false; 2844 CheckConditionalOperand(S, E->getTrueExpr(), T, CC, Suspicious); 2845 CheckConditionalOperand(S, E->getFalseExpr(), T, CC, Suspicious); 2846 2847 // If -Wconversion would have warned about either of the candidates 2848 // for a signedness conversion to the context type... 2849 if (!Suspicious) return; 2850 2851 // ...but it's currently ignored... 2852 if (S.Diags.getDiagnosticLevel(diag::warn_impcast_integer_sign_conditional)) 2853 return; 2854 2855 // ...and -Wsign-compare isn't... 2856 if (!S.Diags.getDiagnosticLevel(diag::warn_mixed_sign_conditional)) 2857 return; 2858 2859 // ...then check whether it would have warned about either of the 2860 // candidates for a signedness conversion to the condition type. 2861 if (E->getType() != T) { 2862 Suspicious = false; 2863 CheckImplicitConversion(S, E->getTrueExpr()->IgnoreParenImpCasts(), 2864 E->getType(), CC, &Suspicious); 2865 if (!Suspicious) 2866 CheckImplicitConversion(S, E->getFalseExpr()->IgnoreParenImpCasts(), 2867 E->getType(), CC, &Suspicious); 2868 if (!Suspicious) 2869 return; 2870 } 2871 2872 // If so, emit a diagnostic under -Wsign-compare. 2873 Expr *lex = E->getTrueExpr()->IgnoreParenImpCasts(); 2874 Expr *rex = E->getFalseExpr()->IgnoreParenImpCasts(); 2875 S.Diag(E->getQuestionLoc(), diag::warn_mixed_sign_conditional) 2876 << lex->getType() << rex->getType() 2877 << lex->getSourceRange() << rex->getSourceRange(); 2878} 2879 2880/// AnalyzeImplicitConversions - Find and report any interesting 2881/// implicit conversions in the given expression. There are a couple 2882/// of competing diagnostics here, -Wconversion and -Wsign-compare. 2883void AnalyzeImplicitConversions(Sema &S, Expr *OrigE, SourceLocation CC) { 2884 QualType T = OrigE->getType(); 2885 Expr *E = OrigE->IgnoreParenImpCasts(); 2886 2887 // For conditional operators, we analyze the arguments as if they 2888 // were being fed directly into the output. 2889 if (isa<ConditionalOperator>(E)) { 2890 ConditionalOperator *CO = cast<ConditionalOperator>(E); 2891 CheckConditionalOperator(S, CO, T); 2892 return; 2893 } 2894 2895 // Go ahead and check any implicit conversions we might have skipped. 2896 // The non-canonical typecheck is just an optimization; 2897 // CheckImplicitConversion will filter out dead implicit conversions. 2898 if (E->getType() != T) 2899 CheckImplicitConversion(S, E, T, CC); 2900 2901 // Now continue drilling into this expression. 2902 2903 // Skip past explicit casts. 2904 if (isa<ExplicitCastExpr>(E)) { 2905 E = cast<ExplicitCastExpr>(E)->getSubExpr()->IgnoreParenImpCasts(); 2906 return AnalyzeImplicitConversions(S, E, CC); 2907 } 2908 2909 if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) { 2910 // Do a somewhat different check with comparison operators. 2911 if (BO->isComparisonOp()) 2912 return AnalyzeComparison(S, BO); 2913 2914 // And with assignments and compound assignments. 2915 if (BO->isAssignmentOp()) 2916 return AnalyzeAssignment(S, BO); 2917 } 2918 2919 // These break the otherwise-useful invariant below. Fortunately, 2920 // we don't really need to recurse into them, because any internal 2921 // expressions should have been analyzed already when they were 2922 // built into statements. 2923 if (isa<StmtExpr>(E)) return; 2924 2925 // Don't descend into unevaluated contexts. 2926 if (isa<SizeOfAlignOfExpr>(E)) return; 2927 2928 // Now just recurse over the expression's children. 2929 CC = E->getExprLoc(); 2930 for (Stmt::child_iterator I = E->child_begin(), IE = E->child_end(); 2931 I != IE; ++I) 2932 AnalyzeImplicitConversions(S, cast<Expr>(*I), CC); 2933} 2934 2935} // end anonymous namespace 2936 2937/// Diagnoses "dangerous" implicit conversions within the given 2938/// expression (which is a full expression). Implements -Wconversion 2939/// and -Wsign-compare. 2940/// 2941/// \param CC the "context" location of the implicit conversion, i.e. 2942/// the most location of the syntactic entity requiring the implicit 2943/// conversion 2944void Sema::CheckImplicitConversions(Expr *E, SourceLocation CC) { 2945 // Don't diagnose in unevaluated contexts. 2946 if (ExprEvalContexts.back().Context == Sema::Unevaluated) 2947 return; 2948 2949 // Don't diagnose for value- or type-dependent expressions. 2950 if (E->isTypeDependent() || E->isValueDependent()) 2951 return; 2952 2953 // This is not the right CC for (e.g.) a variable initialization. 2954 AnalyzeImplicitConversions(*this, E, CC); 2955} 2956 2957void Sema::CheckBitFieldInitialization(SourceLocation InitLoc, 2958 FieldDecl *BitField, 2959 Expr *Init) { 2960 (void) AnalyzeBitFieldAssignment(*this, BitField, Init, InitLoc); 2961} 2962 2963/// CheckParmsForFunctionDef - Check that the parameters of the given 2964/// function are appropriate for the definition of a function. This 2965/// takes care of any checks that cannot be performed on the 2966/// declaration itself, e.g., that the types of each of the function 2967/// parameters are complete. 2968bool Sema::CheckParmsForFunctionDef(ParmVarDecl **P, ParmVarDecl **PEnd, 2969 bool CheckParameterNames) { 2970 bool HasInvalidParm = false; 2971 for (; P != PEnd; ++P) { 2972 ParmVarDecl *Param = *P; 2973 2974 // C99 6.7.5.3p4: the parameters in a parameter type list in a 2975 // function declarator that is part of a function definition of 2976 // that function shall not have incomplete type. 2977 // 2978 // This is also C++ [dcl.fct]p6. 2979 if (!Param->isInvalidDecl() && 2980 RequireCompleteType(Param->getLocation(), Param->getType(), 2981 diag::err_typecheck_decl_incomplete_type)) { 2982 Param->setInvalidDecl(); 2983 HasInvalidParm = true; 2984 } 2985 2986 // C99 6.9.1p5: If the declarator includes a parameter type list, the 2987 // declaration of each parameter shall include an identifier. 2988 if (CheckParameterNames && 2989 Param->getIdentifier() == 0 && 2990 !Param->isImplicit() && 2991 !getLangOptions().CPlusPlus) 2992 Diag(Param->getLocation(), diag::err_parameter_name_omitted); 2993 2994 // C99 6.7.5.3p12: 2995 // If the function declarator is not part of a definition of that 2996 // function, parameters may have incomplete type and may use the [*] 2997 // notation in their sequences of declarator specifiers to specify 2998 // variable length array types. 2999 QualType PType = Param->getOriginalType(); 3000 if (const ArrayType *AT = Context.getAsArrayType(PType)) { 3001 if (AT->getSizeModifier() == ArrayType::Star) { 3002 // FIXME: This diagnosic should point the the '[*]' if source-location 3003 // information is added for it. 3004 Diag(Param->getLocation(), diag::err_array_star_in_function_definition); 3005 } 3006 } 3007 } 3008 3009 return HasInvalidParm; 3010} 3011 3012/// CheckCastAlign - Implements -Wcast-align, which warns when a 3013/// pointer cast increases the alignment requirements. 3014void Sema::CheckCastAlign(Expr *Op, QualType T, SourceRange TRange) { 3015 // This is actually a lot of work to potentially be doing on every 3016 // cast; don't do it if we're ignoring -Wcast_align (as is the default). 3017 if (getDiagnostics().getDiagnosticLevel(diag::warn_cast_align) 3018 == Diagnostic::Ignored) 3019 return; 3020 3021 // Ignore dependent types. 3022 if (T->isDependentType() || Op->getType()->isDependentType()) 3023 return; 3024 3025 // Require that the destination be a pointer type. 3026 const PointerType *DestPtr = T->getAs<PointerType>(); 3027 if (!DestPtr) return; 3028 3029 // If the destination has alignment 1, we're done. 3030 QualType DestPointee = DestPtr->getPointeeType(); 3031 if (DestPointee->isIncompleteType()) return; 3032 CharUnits DestAlign = Context.getTypeAlignInChars(DestPointee); 3033 if (DestAlign.isOne()) return; 3034 3035 // Require that the source be a pointer type. 3036 const PointerType *SrcPtr = Op->getType()->getAs<PointerType>(); 3037 if (!SrcPtr) return; 3038 QualType SrcPointee = SrcPtr->getPointeeType(); 3039 3040 // Whitelist casts from cv void*. We already implicitly 3041 // whitelisted casts to cv void*, since they have alignment 1. 3042 // Also whitelist casts involving incomplete types, which implicitly 3043 // includes 'void'. 3044 if (SrcPointee->isIncompleteType()) return; 3045 3046 CharUnits SrcAlign = Context.getTypeAlignInChars(SrcPointee); 3047 if (SrcAlign >= DestAlign) return; 3048 3049 Diag(TRange.getBegin(), diag::warn_cast_align) 3050 << Op->getType() << T 3051 << static_cast<unsigned>(SrcAlign.getQuantity()) 3052 << static_cast<unsigned>(DestAlign.getQuantity()) 3053 << TRange << Op->getSourceRange(); 3054} 3055 3056