SemaChecking.cpp revision 1844a6e1074a5c1fd8eadac7fc7317c3e29d4653
1//===--- SemaChecking.cpp - Extra Semantic Checking -----------------------===// 2// 3// The LLVM Compiler Infrastructure 4// 5// This file is distributed under the University of Illinois Open Source 6// License. See LICENSE.TXT for details. 7// 8//===----------------------------------------------------------------------===// 9// 10// This file implements extra semantic analysis beyond what is enforced 11// by the C type system. 12// 13//===----------------------------------------------------------------------===// 14 15#include "clang/Sema/Sema.h" 16#include "clang/Sema/SemaInternal.h" 17#include "clang/Sema/ScopeInfo.h" 18#include "clang/Analysis/Analyses/FormatString.h" 19#include "clang/AST/ASTContext.h" 20#include "clang/AST/CharUnits.h" 21#include "clang/AST/DeclCXX.h" 22#include "clang/AST/DeclObjC.h" 23#include "clang/AST/ExprCXX.h" 24#include "clang/AST/ExprObjC.h" 25#include "clang/AST/DeclObjC.h" 26#include "clang/AST/StmtCXX.h" 27#include "clang/AST/StmtObjC.h" 28#include "clang/Lex/LiteralSupport.h" 29#include "clang/Lex/Preprocessor.h" 30#include "llvm/ADT/BitVector.h" 31#include "llvm/ADT/STLExtras.h" 32#include "llvm/Support/raw_ostream.h" 33#include "clang/Basic/TargetBuiltins.h" 34#include "clang/Basic/TargetInfo.h" 35#include "clang/Basic/ConvertUTF.h" 36 37#include <limits> 38using namespace clang; 39using namespace sema; 40 41/// getLocationOfStringLiteralByte - Return a source location that points to the 42/// specified byte of the specified string literal. 43/// 44/// Strings are amazingly complex. They can be formed from multiple tokens and 45/// can have escape sequences in them in addition to the usual trigraph and 46/// escaped newline business. This routine handles this complexity. 47/// 48SourceLocation Sema::getLocationOfStringLiteralByte(const StringLiteral *SL, 49 unsigned ByteNo) const { 50 assert(!SL->isWide() && "This doesn't work for wide strings yet"); 51 52 // Loop over all of the tokens in this string until we find the one that 53 // contains the byte we're looking for. 54 unsigned TokNo = 0; 55 while (1) { 56 assert(TokNo < SL->getNumConcatenated() && "Invalid byte number!"); 57 SourceLocation StrTokLoc = SL->getStrTokenLoc(TokNo); 58 59 // Get the spelling of the string so that we can get the data that makes up 60 // the string literal, not the identifier for the macro it is potentially 61 // expanded through. 62 SourceLocation StrTokSpellingLoc = SourceMgr.getSpellingLoc(StrTokLoc); 63 64 // Re-lex the token to get its length and original spelling. 65 std::pair<FileID, unsigned> LocInfo = 66 SourceMgr.getDecomposedLoc(StrTokSpellingLoc); 67 bool Invalid = false; 68 llvm::StringRef Buffer = SourceMgr.getBufferData(LocInfo.first, &Invalid); 69 if (Invalid) 70 return StrTokSpellingLoc; 71 72 const char *StrData = Buffer.data()+LocInfo.second; 73 74 // Create a langops struct and enable trigraphs. This is sufficient for 75 // relexing tokens. 76 LangOptions LangOpts; 77 LangOpts.Trigraphs = true; 78 79 // Create a lexer starting at the beginning of this token. 80 Lexer TheLexer(StrTokSpellingLoc, LangOpts, Buffer.begin(), StrData, 81 Buffer.end()); 82 Token TheTok; 83 TheLexer.LexFromRawLexer(TheTok); 84 85 // Use the StringLiteralParser to compute the length of the string in bytes. 86 StringLiteralParser SLP(&TheTok, 1, PP, /*Complain=*/false); 87 unsigned TokNumBytes = SLP.GetStringLength(); 88 89 // If the byte is in this token, return the location of the byte. 90 if (ByteNo < TokNumBytes || 91 (ByteNo == TokNumBytes && TokNo == SL->getNumConcatenated())) { 92 unsigned Offset = 93 StringLiteralParser::getOffsetOfStringByte(TheTok, ByteNo, PP, 94 /*Complain=*/false); 95 96 // Now that we know the offset of the token in the spelling, use the 97 // preprocessor to get the offset in the original source. 98 return PP.AdvanceToTokenCharacter(StrTokLoc, Offset); 99 } 100 101 // Move to the next string token. 102 ++TokNo; 103 ByteNo -= TokNumBytes; 104 } 105} 106 107/// CheckablePrintfAttr - does a function call have a "printf" attribute 108/// and arguments that merit checking? 109bool Sema::CheckablePrintfAttr(const FormatAttr *Format, CallExpr *TheCall) { 110 if (Format->getType() == "printf") return true; 111 if (Format->getType() == "printf0") { 112 // printf0 allows null "format" string; if so don't check format/args 113 unsigned format_idx = Format->getFormatIdx() - 1; 114 // Does the index refer to the implicit object argument? 115 if (isa<CXXMemberCallExpr>(TheCall)) { 116 if (format_idx == 0) 117 return false; 118 --format_idx; 119 } 120 if (format_idx < TheCall->getNumArgs()) { 121 Expr *Format = TheCall->getArg(format_idx)->IgnoreParenCasts(); 122 if (!Format->isNullPointerConstant(Context, 123 Expr::NPC_ValueDependentIsNull)) 124 return true; 125 } 126 } 127 return false; 128} 129 130ExprResult 131Sema::CheckBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) { 132 ExprResult TheCallResult(Owned(TheCall)); 133 134 // 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_Unknown; 551 CXXCastPath BasePath; 552 if (CheckCastTypes(Arg->getSourceRange(), ValType, Arg, Kind, BasePath)) 553 return ExprError(); 554 555 // Okay, we have something that *can* be converted to the right type. Check 556 // to see if there is a potentially weird extension going on here. This can 557 // happen when you do an atomic operation on something like an char* and 558 // pass in 42. The 42 gets converted to char. This is even more strange 559 // for things like 45.123 -> char, etc. 560 // FIXME: Do this check. 561 ImpCastExprToType(Arg, ValType, Kind, VK_RValue, &BasePath); 562 TheCall->setArg(i+1, Arg); 563 } 564 565 // Switch the DeclRefExpr to refer to the new decl. 566 DRE->setDecl(NewBuiltinDecl); 567 DRE->setType(NewBuiltinDecl->getType()); 568 569 // Set the callee in the CallExpr. 570 // FIXME: This leaks the original parens and implicit casts. 571 Expr *PromotedCall = DRE; 572 UsualUnaryConversions(PromotedCall); 573 TheCall->setCallee(PromotedCall); 574 575 // Change the result type of the call to match the original value type. This 576 // is arbitrary, but the codegen for these builtins ins design to handle it 577 // gracefully. 578 TheCall->setType(ResultType); 579 580 return move(TheCallResult); 581} 582 583 584/// CheckObjCString - Checks that the argument to the builtin 585/// CFString constructor is correct 586/// Note: It might also make sense to do the UTF-16 conversion here (would 587/// simplify the backend). 588bool Sema::CheckObjCString(Expr *Arg) { 589 Arg = Arg->IgnoreParenCasts(); 590 StringLiteral *Literal = dyn_cast<StringLiteral>(Arg); 591 592 if (!Literal || Literal->isWide()) { 593 Diag(Arg->getLocStart(), diag::err_cfstring_literal_not_string_constant) 594 << Arg->getSourceRange(); 595 return true; 596 } 597 598 size_t NulPos = Literal->getString().find('\0'); 599 if (NulPos != llvm::StringRef::npos) { 600 Diag(getLocationOfStringLiteralByte(Literal, NulPos), 601 diag::warn_cfstring_literal_contains_nul_character) 602 << Arg->getSourceRange(); 603 } 604 if (Literal->containsNonAsciiOrNull()) { 605 llvm::StringRef String = Literal->getString(); 606 unsigned NumBytes = String.size(); 607 llvm::SmallVector<UTF16, 128> ToBuf(NumBytes); 608 const UTF8 *FromPtr = (UTF8 *)String.data(); 609 UTF16 *ToPtr = &ToBuf[0]; 610 611 ConversionResult Result = ConvertUTF8toUTF16(&FromPtr, FromPtr + NumBytes, 612 &ToPtr, ToPtr + NumBytes, 613 strictConversion); 614 // Check for conversion failure. 615 if (Result != conversionOK) 616 Diag(Arg->getLocStart(), 617 diag::warn_cfstring_truncated) << Arg->getSourceRange(); 618 } 619 return false; 620} 621 622/// SemaBuiltinVAStart - Check the arguments to __builtin_va_start for validity. 623/// Emit an error and return true on failure, return false on success. 624bool Sema::SemaBuiltinVAStart(CallExpr *TheCall) { 625 Expr *Fn = TheCall->getCallee(); 626 if (TheCall->getNumArgs() > 2) { 627 Diag(TheCall->getArg(2)->getLocStart(), 628 diag::err_typecheck_call_too_many_args) 629 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 630 << Fn->getSourceRange() 631 << SourceRange(TheCall->getArg(2)->getLocStart(), 632 (*(TheCall->arg_end()-1))->getLocEnd()); 633 return true; 634 } 635 636 if (TheCall->getNumArgs() < 2) { 637 return Diag(TheCall->getLocEnd(), 638 diag::err_typecheck_call_too_few_args_at_least) 639 << 0 /*function call*/ << 2 << TheCall->getNumArgs(); 640 } 641 642 // Determine whether the current function is variadic or not. 643 BlockScopeInfo *CurBlock = getCurBlock(); 644 bool isVariadic; 645 if (CurBlock) 646 isVariadic = CurBlock->TheDecl->isVariadic(); 647 else if (FunctionDecl *FD = getCurFunctionDecl()) 648 isVariadic = FD->isVariadic(); 649 else 650 isVariadic = getCurMethodDecl()->isVariadic(); 651 652 if (!isVariadic) { 653 Diag(Fn->getLocStart(), diag::err_va_start_used_in_non_variadic_function); 654 return true; 655 } 656 657 // Verify that the second argument to the builtin is the last argument of the 658 // current function or method. 659 bool SecondArgIsLastNamedArgument = false; 660 const Expr *Arg = TheCall->getArg(1)->IgnoreParenCasts(); 661 662 if (const DeclRefExpr *DR = dyn_cast<DeclRefExpr>(Arg)) { 663 if (const ParmVarDecl *PV = dyn_cast<ParmVarDecl>(DR->getDecl())) { 664 // FIXME: This isn't correct for methods (results in bogus warning). 665 // Get the last formal in the current function. 666 const ParmVarDecl *LastArg; 667 if (CurBlock) 668 LastArg = *(CurBlock->TheDecl->param_end()-1); 669 else if (FunctionDecl *FD = getCurFunctionDecl()) 670 LastArg = *(FD->param_end()-1); 671 else 672 LastArg = *(getCurMethodDecl()->param_end()-1); 673 SecondArgIsLastNamedArgument = PV == LastArg; 674 } 675 } 676 677 if (!SecondArgIsLastNamedArgument) 678 Diag(TheCall->getArg(1)->getLocStart(), 679 diag::warn_second_parameter_of_va_start_not_last_named_argument); 680 return false; 681} 682 683/// SemaBuiltinUnorderedCompare - Handle functions like __builtin_isgreater and 684/// friends. This is declared to take (...), so we have to check everything. 685bool Sema::SemaBuiltinUnorderedCompare(CallExpr *TheCall) { 686 if (TheCall->getNumArgs() < 2) 687 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 688 << 0 << 2 << TheCall->getNumArgs()/*function call*/; 689 if (TheCall->getNumArgs() > 2) 690 return Diag(TheCall->getArg(2)->getLocStart(), 691 diag::err_typecheck_call_too_many_args) 692 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 693 << SourceRange(TheCall->getArg(2)->getLocStart(), 694 (*(TheCall->arg_end()-1))->getLocEnd()); 695 696 Expr *OrigArg0 = TheCall->getArg(0); 697 Expr *OrigArg1 = TheCall->getArg(1); 698 699 // Do standard promotions between the two arguments, returning their common 700 // type. 701 QualType Res = UsualArithmeticConversions(OrigArg0, OrigArg1, false); 702 703 // Make sure any conversions are pushed back into the call; this is 704 // type safe since unordered compare builtins are declared as "_Bool 705 // foo(...)". 706 TheCall->setArg(0, OrigArg0); 707 TheCall->setArg(1, OrigArg1); 708 709 if (OrigArg0->isTypeDependent() || OrigArg1->isTypeDependent()) 710 return false; 711 712 // If the common type isn't a real floating type, then the arguments were 713 // invalid for this operation. 714 if (!Res->isRealFloatingType()) 715 return Diag(OrigArg0->getLocStart(), 716 diag::err_typecheck_call_invalid_ordered_compare) 717 << OrigArg0->getType() << OrigArg1->getType() 718 << SourceRange(OrigArg0->getLocStart(), OrigArg1->getLocEnd()); 719 720 return false; 721} 722 723/// SemaBuiltinSemaBuiltinFPClassification - Handle functions like 724/// __builtin_isnan and friends. This is declared to take (...), so we have 725/// to check everything. We expect the last argument to be a floating point 726/// value. 727bool Sema::SemaBuiltinFPClassification(CallExpr *TheCall, unsigned NumArgs) { 728 if (TheCall->getNumArgs() < NumArgs) 729 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 730 << 0 << NumArgs << TheCall->getNumArgs()/*function call*/; 731 if (TheCall->getNumArgs() > NumArgs) 732 return Diag(TheCall->getArg(NumArgs)->getLocStart(), 733 diag::err_typecheck_call_too_many_args) 734 << 0 /*function call*/ << NumArgs << TheCall->getNumArgs() 735 << SourceRange(TheCall->getArg(NumArgs)->getLocStart(), 736 (*(TheCall->arg_end()-1))->getLocEnd()); 737 738 Expr *OrigArg = TheCall->getArg(NumArgs-1); 739 740 if (OrigArg->isTypeDependent()) 741 return false; 742 743 // This operation requires a non-_Complex floating-point number. 744 if (!OrigArg->getType()->isRealFloatingType()) 745 return Diag(OrigArg->getLocStart(), 746 diag::err_typecheck_call_invalid_unary_fp) 747 << OrigArg->getType() << OrigArg->getSourceRange(); 748 749 // If this is an implicit conversion from float -> double, remove it. 750 if (ImplicitCastExpr *Cast = dyn_cast<ImplicitCastExpr>(OrigArg)) { 751 Expr *CastArg = Cast->getSubExpr(); 752 if (CastArg->getType()->isSpecificBuiltinType(BuiltinType::Float)) { 753 assert(Cast->getType()->isSpecificBuiltinType(BuiltinType::Double) && 754 "promotion from float to double is the only expected cast here"); 755 Cast->setSubExpr(0); 756 TheCall->setArg(NumArgs-1, CastArg); 757 OrigArg = CastArg; 758 } 759 } 760 761 return false; 762} 763 764/// SemaBuiltinShuffleVector - Handle __builtin_shufflevector. 765// This is declared to take (...), so we have to check everything. 766ExprResult Sema::SemaBuiltinShuffleVector(CallExpr *TheCall) { 767 if (TheCall->getNumArgs() < 2) 768 return ExprError(Diag(TheCall->getLocEnd(), 769 diag::err_typecheck_call_too_few_args_at_least) 770 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 771 << TheCall->getSourceRange()); 772 773 // Determine which of the following types of shufflevector we're checking: 774 // 1) unary, vector mask: (lhs, mask) 775 // 2) binary, vector mask: (lhs, rhs, mask) 776 // 3) binary, scalar mask: (lhs, rhs, index, ..., index) 777 QualType resType = TheCall->getArg(0)->getType(); 778 unsigned numElements = 0; 779 780 if (!TheCall->getArg(0)->isTypeDependent() && 781 !TheCall->getArg(1)->isTypeDependent()) { 782 QualType LHSType = TheCall->getArg(0)->getType(); 783 QualType RHSType = TheCall->getArg(1)->getType(); 784 785 if (!LHSType->isVectorType() || !RHSType->isVectorType()) { 786 Diag(TheCall->getLocStart(), diag::err_shufflevector_non_vector) 787 << SourceRange(TheCall->getArg(0)->getLocStart(), 788 TheCall->getArg(1)->getLocEnd()); 789 return ExprError(); 790 } 791 792 numElements = LHSType->getAs<VectorType>()->getNumElements(); 793 unsigned numResElements = TheCall->getNumArgs() - 2; 794 795 // Check to see if we have a call with 2 vector arguments, the unary shuffle 796 // with mask. If so, verify that RHS is an integer vector type with the 797 // same number of elts as lhs. 798 if (TheCall->getNumArgs() == 2) { 799 if (!RHSType->hasIntegerRepresentation() || 800 RHSType->getAs<VectorType>()->getNumElements() != numElements) 801 Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector) 802 << SourceRange(TheCall->getArg(1)->getLocStart(), 803 TheCall->getArg(1)->getLocEnd()); 804 numResElements = numElements; 805 } 806 else if (!Context.hasSameUnqualifiedType(LHSType, RHSType)) { 807 Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector) 808 << SourceRange(TheCall->getArg(0)->getLocStart(), 809 TheCall->getArg(1)->getLocEnd()); 810 return ExprError(); 811 } else if (numElements != numResElements) { 812 QualType eltType = LHSType->getAs<VectorType>()->getElementType(); 813 resType = Context.getVectorType(eltType, numResElements, 814 VectorType::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 // Catch a format attribute mistakenly referring to the object argument. 1094 if (format_idx == 0) 1095 return; 1096 --format_idx; 1097 if(firstDataArg != 0) 1098 --firstDataArg; 1099 } 1100 1101 // CHECK: printf/scanf-like function is called with no format string. 1102 if (format_idx >= TheCall->getNumArgs()) { 1103 Diag(TheCall->getRParenLoc(), diag::warn_missing_format_string) 1104 << Fn->getSourceRange(); 1105 return; 1106 } 1107 1108 const Expr *OrigFormatExpr = TheCall->getArg(format_idx)->IgnoreParenCasts(); 1109 1110 // CHECK: format string is not a string literal. 1111 // 1112 // Dynamically generated format strings are difficult to 1113 // automatically vet at compile time. Requiring that format strings 1114 // are string literals: (1) permits the checking of format strings by 1115 // the compiler and thereby (2) can practically remove the source of 1116 // many format string exploits. 1117 1118 // Format string can be either ObjC string (e.g. @"%d") or 1119 // C string (e.g. "%d") 1120 // ObjC string uses the same format specifiers as C string, so we can use 1121 // the same format string checking logic for both ObjC and C strings. 1122 if (SemaCheckStringLiteral(OrigFormatExpr, TheCall, HasVAListArg, format_idx, 1123 firstDataArg, isPrintf)) 1124 return; // Literal format string found, check done! 1125 1126 // If there are no arguments specified, warn with -Wformat-security, otherwise 1127 // warn only with -Wformat-nonliteral. 1128 if (TheCall->getNumArgs() == format_idx+1) 1129 Diag(TheCall->getArg(format_idx)->getLocStart(), 1130 diag::warn_format_nonliteral_noargs) 1131 << OrigFormatExpr->getSourceRange(); 1132 else 1133 Diag(TheCall->getArg(format_idx)->getLocStart(), 1134 diag::warn_format_nonliteral) 1135 << OrigFormatExpr->getSourceRange(); 1136} 1137 1138namespace { 1139class CheckFormatHandler : public analyze_format_string::FormatStringHandler { 1140protected: 1141 Sema &S; 1142 const StringLiteral *FExpr; 1143 const Expr *OrigFormatExpr; 1144 const unsigned FirstDataArg; 1145 const unsigned NumDataArgs; 1146 const bool IsObjCLiteral; 1147 const char *Beg; // Start of format string. 1148 const bool HasVAListArg; 1149 const CallExpr *TheCall; 1150 unsigned FormatIdx; 1151 llvm::BitVector CoveredArgs; 1152 bool usesPositionalArgs; 1153 bool atFirstArg; 1154public: 1155 CheckFormatHandler(Sema &s, const StringLiteral *fexpr, 1156 const Expr *origFormatExpr, unsigned firstDataArg, 1157 unsigned numDataArgs, bool isObjCLiteral, 1158 const char *beg, bool hasVAListArg, 1159 const CallExpr *theCall, unsigned formatIdx) 1160 : S(s), FExpr(fexpr), OrigFormatExpr(origFormatExpr), 1161 FirstDataArg(firstDataArg), 1162 NumDataArgs(numDataArgs), 1163 IsObjCLiteral(isObjCLiteral), Beg(beg), 1164 HasVAListArg(hasVAListArg), 1165 TheCall(theCall), FormatIdx(formatIdx), 1166 usesPositionalArgs(false), atFirstArg(true) { 1167 CoveredArgs.resize(numDataArgs); 1168 CoveredArgs.reset(); 1169 } 1170 1171 void DoneProcessing(); 1172 1173 void HandleIncompleteSpecifier(const char *startSpecifier, 1174 unsigned specifierLen); 1175 1176 virtual void HandleInvalidPosition(const char *startSpecifier, 1177 unsigned specifierLen, 1178 analyze_format_string::PositionContext p); 1179 1180 virtual void HandleZeroPosition(const char *startPos, unsigned posLen); 1181 1182 void HandleNullChar(const char *nullCharacter); 1183 1184protected: 1185 bool HandleInvalidConversionSpecifier(unsigned argIndex, SourceLocation Loc, 1186 const char *startSpec, 1187 unsigned specifierLen, 1188 const char *csStart, unsigned csLen); 1189 1190 SourceRange getFormatStringRange(); 1191 CharSourceRange getSpecifierRange(const char *startSpecifier, 1192 unsigned specifierLen); 1193 SourceLocation getLocationOfByte(const char *x); 1194 1195 const Expr *getDataArg(unsigned i) const; 1196 1197 bool CheckNumArgs(const analyze_format_string::FormatSpecifier &FS, 1198 const analyze_format_string::ConversionSpecifier &CS, 1199 const char *startSpecifier, unsigned specifierLen, 1200 unsigned argIndex); 1201}; 1202} 1203 1204SourceRange CheckFormatHandler::getFormatStringRange() { 1205 return OrigFormatExpr->getSourceRange(); 1206} 1207 1208CharSourceRange CheckFormatHandler:: 1209getSpecifierRange(const char *startSpecifier, unsigned specifierLen) { 1210 SourceLocation Start = getLocationOfByte(startSpecifier); 1211 SourceLocation End = getLocationOfByte(startSpecifier + specifierLen - 1); 1212 1213 // Advance the end SourceLocation by one due to half-open ranges. 1214 End = End.getFileLocWithOffset(1); 1215 1216 return CharSourceRange::getCharRange(Start, End); 1217} 1218 1219SourceLocation CheckFormatHandler::getLocationOfByte(const char *x) { 1220 return S.getLocationOfStringLiteralByte(FExpr, x - Beg); 1221} 1222 1223void CheckFormatHandler::HandleIncompleteSpecifier(const char *startSpecifier, 1224 unsigned specifierLen){ 1225 SourceLocation Loc = getLocationOfByte(startSpecifier); 1226 S.Diag(Loc, diag::warn_printf_incomplete_specifier) 1227 << getSpecifierRange(startSpecifier, specifierLen); 1228} 1229 1230void 1231CheckFormatHandler::HandleInvalidPosition(const char *startPos, unsigned posLen, 1232 analyze_format_string::PositionContext p) { 1233 SourceLocation Loc = getLocationOfByte(startPos); 1234 S.Diag(Loc, diag::warn_format_invalid_positional_specifier) 1235 << (unsigned) p << getSpecifierRange(startPos, posLen); 1236} 1237 1238void CheckFormatHandler::HandleZeroPosition(const char *startPos, 1239 unsigned posLen) { 1240 SourceLocation Loc = getLocationOfByte(startPos); 1241 S.Diag(Loc, diag::warn_format_zero_positional_specifier) 1242 << getSpecifierRange(startPos, posLen); 1243} 1244 1245void CheckFormatHandler::HandleNullChar(const char *nullCharacter) { 1246 // The presence of a null character is likely an error. 1247 S.Diag(getLocationOfByte(nullCharacter), 1248 diag::warn_printf_format_string_contains_null_char) 1249 << getFormatStringRange(); 1250} 1251 1252const Expr *CheckFormatHandler::getDataArg(unsigned i) const { 1253 return TheCall->getArg(FirstDataArg + i); 1254} 1255 1256void CheckFormatHandler::DoneProcessing() { 1257 // Does the number of data arguments exceed the number of 1258 // format conversions in the format string? 1259 if (!HasVAListArg) { 1260 // Find any arguments that weren't covered. 1261 CoveredArgs.flip(); 1262 signed notCoveredArg = CoveredArgs.find_first(); 1263 if (notCoveredArg >= 0) { 1264 assert((unsigned)notCoveredArg < NumDataArgs); 1265 S.Diag(getDataArg((unsigned) notCoveredArg)->getLocStart(), 1266 diag::warn_printf_data_arg_not_used) 1267 << getFormatStringRange(); 1268 } 1269 } 1270} 1271 1272bool 1273CheckFormatHandler::HandleInvalidConversionSpecifier(unsigned argIndex, 1274 SourceLocation Loc, 1275 const char *startSpec, 1276 unsigned specifierLen, 1277 const char *csStart, 1278 unsigned csLen) { 1279 1280 bool keepGoing = true; 1281 if (argIndex < NumDataArgs) { 1282 // Consider the argument coverered, even though the specifier doesn't 1283 // make sense. 1284 CoveredArgs.set(argIndex); 1285 } 1286 else { 1287 // If argIndex exceeds the number of data arguments we 1288 // don't issue a warning because that is just a cascade of warnings (and 1289 // they may have intended '%%' anyway). We don't want to continue processing 1290 // the format string after this point, however, as we will like just get 1291 // gibberish when trying to match arguments. 1292 keepGoing = false; 1293 } 1294 1295 S.Diag(Loc, diag::warn_format_invalid_conversion) 1296 << llvm::StringRef(csStart, csLen) 1297 << getSpecifierRange(startSpec, specifierLen); 1298 1299 return keepGoing; 1300} 1301 1302bool 1303CheckFormatHandler::CheckNumArgs( 1304 const analyze_format_string::FormatSpecifier &FS, 1305 const analyze_format_string::ConversionSpecifier &CS, 1306 const char *startSpecifier, unsigned specifierLen, unsigned argIndex) { 1307 1308 if (argIndex >= NumDataArgs) { 1309 if (FS.usesPositionalArg()) { 1310 S.Diag(getLocationOfByte(CS.getStart()), 1311 diag::warn_printf_positional_arg_exceeds_data_args) 1312 << (argIndex+1) << NumDataArgs 1313 << getSpecifierRange(startSpecifier, specifierLen); 1314 } 1315 else { 1316 S.Diag(getLocationOfByte(CS.getStart()), 1317 diag::warn_printf_insufficient_data_args) 1318 << getSpecifierRange(startSpecifier, specifierLen); 1319 } 1320 1321 return false; 1322 } 1323 return true; 1324} 1325 1326//===--- CHECK: Printf format string checking ------------------------------===// 1327 1328namespace { 1329class CheckPrintfHandler : public CheckFormatHandler { 1330public: 1331 CheckPrintfHandler(Sema &s, const StringLiteral *fexpr, 1332 const Expr *origFormatExpr, unsigned firstDataArg, 1333 unsigned numDataArgs, bool isObjCLiteral, 1334 const char *beg, bool hasVAListArg, 1335 const CallExpr *theCall, unsigned formatIdx) 1336 : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg, 1337 numDataArgs, isObjCLiteral, beg, hasVAListArg, 1338 theCall, formatIdx) {} 1339 1340 1341 bool HandleInvalidPrintfConversionSpecifier( 1342 const analyze_printf::PrintfSpecifier &FS, 1343 const char *startSpecifier, 1344 unsigned specifierLen); 1345 1346 bool HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier &FS, 1347 const char *startSpecifier, 1348 unsigned specifierLen); 1349 1350 bool HandleAmount(const analyze_format_string::OptionalAmount &Amt, unsigned k, 1351 const char *startSpecifier, unsigned specifierLen); 1352 void HandleInvalidAmount(const analyze_printf::PrintfSpecifier &FS, 1353 const analyze_printf::OptionalAmount &Amt, 1354 unsigned type, 1355 const char *startSpecifier, unsigned specifierLen); 1356 void HandleFlag(const analyze_printf::PrintfSpecifier &FS, 1357 const analyze_printf::OptionalFlag &flag, 1358 const char *startSpecifier, unsigned specifierLen); 1359 void HandleIgnoredFlag(const analyze_printf::PrintfSpecifier &FS, 1360 const analyze_printf::OptionalFlag &ignoredFlag, 1361 const analyze_printf::OptionalFlag &flag, 1362 const char *startSpecifier, unsigned specifierLen); 1363}; 1364} 1365 1366bool CheckPrintfHandler::HandleInvalidPrintfConversionSpecifier( 1367 const analyze_printf::PrintfSpecifier &FS, 1368 const char *startSpecifier, 1369 unsigned specifierLen) { 1370 const analyze_printf::PrintfConversionSpecifier &CS = 1371 FS.getConversionSpecifier(); 1372 1373 return HandleInvalidConversionSpecifier(FS.getArgIndex(), 1374 getLocationOfByte(CS.getStart()), 1375 startSpecifier, specifierLen, 1376 CS.getStart(), CS.getLength()); 1377} 1378 1379bool CheckPrintfHandler::HandleAmount( 1380 const analyze_format_string::OptionalAmount &Amt, 1381 unsigned k, const char *startSpecifier, 1382 unsigned specifierLen) { 1383 1384 if (Amt.hasDataArgument()) { 1385 if (!HasVAListArg) { 1386 unsigned argIndex = Amt.getArgIndex(); 1387 if (argIndex >= NumDataArgs) { 1388 S.Diag(getLocationOfByte(Amt.getStart()), 1389 diag::warn_printf_asterisk_missing_arg) 1390 << k << getSpecifierRange(startSpecifier, specifierLen); 1391 // Don't do any more checking. We will just emit 1392 // spurious errors. 1393 return false; 1394 } 1395 1396 // Type check the data argument. It should be an 'int'. 1397 // Although not in conformance with C99, we also allow the argument to be 1398 // an 'unsigned int' as that is a reasonably safe case. GCC also 1399 // doesn't emit a warning for that case. 1400 CoveredArgs.set(argIndex); 1401 const Expr *Arg = getDataArg(argIndex); 1402 QualType T = Arg->getType(); 1403 1404 const analyze_printf::ArgTypeResult &ATR = Amt.getArgType(S.Context); 1405 assert(ATR.isValid()); 1406 1407 if (!ATR.matchesType(S.Context, T)) { 1408 S.Diag(getLocationOfByte(Amt.getStart()), 1409 diag::warn_printf_asterisk_wrong_type) 1410 << k 1411 << ATR.getRepresentativeType(S.Context) << T 1412 << getSpecifierRange(startSpecifier, specifierLen) 1413 << Arg->getSourceRange(); 1414 // Don't do any more checking. We will just emit 1415 // spurious errors. 1416 return false; 1417 } 1418 } 1419 } 1420 return true; 1421} 1422 1423void CheckPrintfHandler::HandleInvalidAmount( 1424 const analyze_printf::PrintfSpecifier &FS, 1425 const analyze_printf::OptionalAmount &Amt, 1426 unsigned type, 1427 const char *startSpecifier, 1428 unsigned specifierLen) { 1429 const analyze_printf::PrintfConversionSpecifier &CS = 1430 FS.getConversionSpecifier(); 1431 switch (Amt.getHowSpecified()) { 1432 case analyze_printf::OptionalAmount::Constant: 1433 S.Diag(getLocationOfByte(Amt.getStart()), 1434 diag::warn_printf_nonsensical_optional_amount) 1435 << type 1436 << CS.toString() 1437 << getSpecifierRange(startSpecifier, specifierLen) 1438 << FixItHint::CreateRemoval(getSpecifierRange(Amt.getStart(), 1439 Amt.getConstantLength())); 1440 break; 1441 1442 default: 1443 S.Diag(getLocationOfByte(Amt.getStart()), 1444 diag::warn_printf_nonsensical_optional_amount) 1445 << type 1446 << CS.toString() 1447 << getSpecifierRange(startSpecifier, specifierLen); 1448 break; 1449 } 1450} 1451 1452void CheckPrintfHandler::HandleFlag(const analyze_printf::PrintfSpecifier &FS, 1453 const analyze_printf::OptionalFlag &flag, 1454 const char *startSpecifier, 1455 unsigned specifierLen) { 1456 // Warn about pointless flag with a fixit removal. 1457 const analyze_printf::PrintfConversionSpecifier &CS = 1458 FS.getConversionSpecifier(); 1459 S.Diag(getLocationOfByte(flag.getPosition()), 1460 diag::warn_printf_nonsensical_flag) 1461 << flag.toString() << CS.toString() 1462 << getSpecifierRange(startSpecifier, specifierLen) 1463 << FixItHint::CreateRemoval(getSpecifierRange(flag.getPosition(), 1)); 1464} 1465 1466void CheckPrintfHandler::HandleIgnoredFlag( 1467 const analyze_printf::PrintfSpecifier &FS, 1468 const analyze_printf::OptionalFlag &ignoredFlag, 1469 const analyze_printf::OptionalFlag &flag, 1470 const char *startSpecifier, 1471 unsigned specifierLen) { 1472 // Warn about ignored flag with a fixit removal. 1473 S.Diag(getLocationOfByte(ignoredFlag.getPosition()), 1474 diag::warn_printf_ignored_flag) 1475 << ignoredFlag.toString() << flag.toString() 1476 << getSpecifierRange(startSpecifier, specifierLen) 1477 << FixItHint::CreateRemoval(getSpecifierRange( 1478 ignoredFlag.getPosition(), 1)); 1479} 1480 1481bool 1482CheckPrintfHandler::HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier 1483 &FS, 1484 const char *startSpecifier, 1485 unsigned specifierLen) { 1486 1487 using namespace analyze_format_string; 1488 using namespace analyze_printf; 1489 const PrintfConversionSpecifier &CS = FS.getConversionSpecifier(); 1490 1491 if (FS.consumesDataArgument()) { 1492 if (atFirstArg) { 1493 atFirstArg = false; 1494 usesPositionalArgs = FS.usesPositionalArg(); 1495 } 1496 else if (usesPositionalArgs != FS.usesPositionalArg()) { 1497 // Cannot mix-and-match positional and non-positional arguments. 1498 S.Diag(getLocationOfByte(CS.getStart()), 1499 diag::warn_format_mix_positional_nonpositional_args) 1500 << getSpecifierRange(startSpecifier, specifierLen); 1501 return false; 1502 } 1503 } 1504 1505 // First check if the field width, precision, and conversion specifier 1506 // have matching data arguments. 1507 if (!HandleAmount(FS.getFieldWidth(), /* field width */ 0, 1508 startSpecifier, specifierLen)) { 1509 return false; 1510 } 1511 1512 if (!HandleAmount(FS.getPrecision(), /* precision */ 1, 1513 startSpecifier, specifierLen)) { 1514 return false; 1515 } 1516 1517 if (!CS.consumesDataArgument()) { 1518 // FIXME: Technically specifying a precision or field width here 1519 // makes no sense. Worth issuing a warning at some point. 1520 return true; 1521 } 1522 1523 // Consume the argument. 1524 unsigned argIndex = FS.getArgIndex(); 1525 if (argIndex < NumDataArgs) { 1526 // The check to see if the argIndex is valid will come later. 1527 // We set the bit here because we may exit early from this 1528 // function if we encounter some other error. 1529 CoveredArgs.set(argIndex); 1530 } 1531 1532 // Check for using an Objective-C specific conversion specifier 1533 // in a non-ObjC literal. 1534 if (!IsObjCLiteral && CS.isObjCArg()) { 1535 return HandleInvalidPrintfConversionSpecifier(FS, startSpecifier, 1536 specifierLen); 1537 } 1538 1539 // Check for invalid use of field width 1540 if (!FS.hasValidFieldWidth()) { 1541 HandleInvalidAmount(FS, FS.getFieldWidth(), /* field width */ 0, 1542 startSpecifier, specifierLen); 1543 } 1544 1545 // Check for invalid use of precision 1546 if (!FS.hasValidPrecision()) { 1547 HandleInvalidAmount(FS, FS.getPrecision(), /* precision */ 1, 1548 startSpecifier, specifierLen); 1549 } 1550 1551 // Check each flag does not conflict with any other component. 1552 if (!FS.hasValidLeadingZeros()) 1553 HandleFlag(FS, FS.hasLeadingZeros(), startSpecifier, specifierLen); 1554 if (!FS.hasValidPlusPrefix()) 1555 HandleFlag(FS, FS.hasPlusPrefix(), startSpecifier, specifierLen); 1556 if (!FS.hasValidSpacePrefix()) 1557 HandleFlag(FS, FS.hasSpacePrefix(), startSpecifier, specifierLen); 1558 if (!FS.hasValidAlternativeForm()) 1559 HandleFlag(FS, FS.hasAlternativeForm(), startSpecifier, specifierLen); 1560 if (!FS.hasValidLeftJustified()) 1561 HandleFlag(FS, FS.isLeftJustified(), startSpecifier, specifierLen); 1562 1563 // Check that flags are not ignored by another flag 1564 if (FS.hasSpacePrefix() && FS.hasPlusPrefix()) // ' ' ignored by '+' 1565 HandleIgnoredFlag(FS, FS.hasSpacePrefix(), FS.hasPlusPrefix(), 1566 startSpecifier, specifierLen); 1567 if (FS.hasLeadingZeros() && FS.isLeftJustified()) // '0' ignored by '-' 1568 HandleIgnoredFlag(FS, FS.hasLeadingZeros(), FS.isLeftJustified(), 1569 startSpecifier, specifierLen); 1570 1571 // Check the length modifier is valid with the given conversion specifier. 1572 const LengthModifier &LM = FS.getLengthModifier(); 1573 if (!FS.hasValidLengthModifier()) 1574 S.Diag(getLocationOfByte(LM.getStart()), 1575 diag::warn_format_nonsensical_length) 1576 << LM.toString() << CS.toString() 1577 << getSpecifierRange(startSpecifier, specifierLen) 1578 << FixItHint::CreateRemoval(getSpecifierRange(LM.getStart(), 1579 LM.getLength())); 1580 1581 // Are we using '%n'? 1582 if (CS.getKind() == ConversionSpecifier::nArg) { 1583 // Issue a warning about this being a possible security issue. 1584 S.Diag(getLocationOfByte(CS.getStart()), diag::warn_printf_write_back) 1585 << getSpecifierRange(startSpecifier, specifierLen); 1586 // Continue checking the other format specifiers. 1587 return true; 1588 } 1589 1590 // The remaining checks depend on the data arguments. 1591 if (HasVAListArg) 1592 return true; 1593 1594 if (!CheckNumArgs(FS, CS, startSpecifier, specifierLen, argIndex)) 1595 return false; 1596 1597 // Now type check the data expression that matches the 1598 // format specifier. 1599 const Expr *Ex = getDataArg(argIndex); 1600 const analyze_printf::ArgTypeResult &ATR = FS.getArgType(S.Context); 1601 if (ATR.isValid() && !ATR.matchesType(S.Context, Ex->getType())) { 1602 // Check if we didn't match because of an implicit cast from a 'char' 1603 // or 'short' to an 'int'. This is done because printf is a varargs 1604 // function. 1605 if (const ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(Ex)) 1606 if (ICE->getType() == S.Context.IntTy) { 1607 // All further checking is done on the subexpression. 1608 Ex = ICE->getSubExpr(); 1609 if (ATR.matchesType(S.Context, Ex->getType())) 1610 return true; 1611 } 1612 1613 // We may be able to offer a FixItHint if it is a supported type. 1614 PrintfSpecifier fixedFS = FS; 1615 bool success = fixedFS.fixType(Ex->getType()); 1616 1617 if (success) { 1618 // Get the fix string from the fixed format specifier 1619 llvm::SmallString<128> buf; 1620 llvm::raw_svector_ostream os(buf); 1621 fixedFS.toString(os); 1622 1623 // FIXME: getRepresentativeType() perhaps should return a string 1624 // instead of a QualType to better handle when the representative 1625 // type is 'wint_t' (which is defined in the system headers). 1626 S.Diag(getLocationOfByte(CS.getStart()), 1627 diag::warn_printf_conversion_argument_type_mismatch) 1628 << ATR.getRepresentativeType(S.Context) << Ex->getType() 1629 << getSpecifierRange(startSpecifier, specifierLen) 1630 << Ex->getSourceRange() 1631 << FixItHint::CreateReplacement( 1632 getSpecifierRange(startSpecifier, specifierLen), 1633 os.str()); 1634 } 1635 else { 1636 S.Diag(getLocationOfByte(CS.getStart()), 1637 diag::warn_printf_conversion_argument_type_mismatch) 1638 << ATR.getRepresentativeType(S.Context) << Ex->getType() 1639 << getSpecifierRange(startSpecifier, specifierLen) 1640 << Ex->getSourceRange(); 1641 } 1642 } 1643 1644 return true; 1645} 1646 1647//===--- CHECK: Scanf format string checking ------------------------------===// 1648 1649namespace { 1650class CheckScanfHandler : public CheckFormatHandler { 1651public: 1652 CheckScanfHandler(Sema &s, const StringLiteral *fexpr, 1653 const Expr *origFormatExpr, unsigned firstDataArg, 1654 unsigned numDataArgs, bool isObjCLiteral, 1655 const char *beg, bool hasVAListArg, 1656 const CallExpr *theCall, unsigned formatIdx) 1657 : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg, 1658 numDataArgs, isObjCLiteral, beg, hasVAListArg, 1659 theCall, formatIdx) {} 1660 1661 bool HandleScanfSpecifier(const analyze_scanf::ScanfSpecifier &FS, 1662 const char *startSpecifier, 1663 unsigned specifierLen); 1664 1665 bool HandleInvalidScanfConversionSpecifier( 1666 const analyze_scanf::ScanfSpecifier &FS, 1667 const char *startSpecifier, 1668 unsigned specifierLen); 1669 1670 void HandleIncompleteScanList(const char *start, const char *end); 1671}; 1672} 1673 1674void CheckScanfHandler::HandleIncompleteScanList(const char *start, 1675 const char *end) { 1676 S.Diag(getLocationOfByte(end), diag::warn_scanf_scanlist_incomplete) 1677 << getSpecifierRange(start, end - start); 1678} 1679 1680bool CheckScanfHandler::HandleInvalidScanfConversionSpecifier( 1681 const analyze_scanf::ScanfSpecifier &FS, 1682 const char *startSpecifier, 1683 unsigned specifierLen) { 1684 1685 const analyze_scanf::ScanfConversionSpecifier &CS = 1686 FS.getConversionSpecifier(); 1687 1688 return HandleInvalidConversionSpecifier(FS.getArgIndex(), 1689 getLocationOfByte(CS.getStart()), 1690 startSpecifier, specifierLen, 1691 CS.getStart(), CS.getLength()); 1692} 1693 1694bool CheckScanfHandler::HandleScanfSpecifier( 1695 const analyze_scanf::ScanfSpecifier &FS, 1696 const char *startSpecifier, 1697 unsigned specifierLen) { 1698 1699 using namespace analyze_scanf; 1700 using namespace analyze_format_string; 1701 1702 const ScanfConversionSpecifier &CS = FS.getConversionSpecifier(); 1703 1704 // Handle case where '%' and '*' don't consume an argument. These shouldn't 1705 // be used to decide if we are using positional arguments consistently. 1706 if (FS.consumesDataArgument()) { 1707 if (atFirstArg) { 1708 atFirstArg = false; 1709 usesPositionalArgs = FS.usesPositionalArg(); 1710 } 1711 else if (usesPositionalArgs != FS.usesPositionalArg()) { 1712 // Cannot mix-and-match positional and non-positional arguments. 1713 S.Diag(getLocationOfByte(CS.getStart()), 1714 diag::warn_format_mix_positional_nonpositional_args) 1715 << getSpecifierRange(startSpecifier, specifierLen); 1716 return false; 1717 } 1718 } 1719 1720 // Check if the field with is non-zero. 1721 const OptionalAmount &Amt = FS.getFieldWidth(); 1722 if (Amt.getHowSpecified() == OptionalAmount::Constant) { 1723 if (Amt.getConstantAmount() == 0) { 1724 const CharSourceRange &R = getSpecifierRange(Amt.getStart(), 1725 Amt.getConstantLength()); 1726 S.Diag(getLocationOfByte(Amt.getStart()), 1727 diag::warn_scanf_nonzero_width) 1728 << R << FixItHint::CreateRemoval(R); 1729 } 1730 } 1731 1732 if (!FS.consumesDataArgument()) { 1733 // FIXME: Technically specifying a precision or field width here 1734 // makes no sense. Worth issuing a warning at some point. 1735 return true; 1736 } 1737 1738 // Consume the argument. 1739 unsigned argIndex = FS.getArgIndex(); 1740 if (argIndex < NumDataArgs) { 1741 // The check to see if the argIndex is valid will come later. 1742 // We set the bit here because we may exit early from this 1743 // function if we encounter some other error. 1744 CoveredArgs.set(argIndex); 1745 } 1746 1747 // Check the length modifier is valid with the given conversion specifier. 1748 const LengthModifier &LM = FS.getLengthModifier(); 1749 if (!FS.hasValidLengthModifier()) { 1750 S.Diag(getLocationOfByte(LM.getStart()), 1751 diag::warn_format_nonsensical_length) 1752 << LM.toString() << CS.toString() 1753 << getSpecifierRange(startSpecifier, specifierLen) 1754 << FixItHint::CreateRemoval(getSpecifierRange(LM.getStart(), 1755 LM.getLength())); 1756 } 1757 1758 // The remaining checks depend on the data arguments. 1759 if (HasVAListArg) 1760 return true; 1761 1762 if (!CheckNumArgs(FS, CS, startSpecifier, specifierLen, argIndex)) 1763 return false; 1764 1765 // FIXME: Check that the argument type matches the format specifier. 1766 1767 return true; 1768} 1769 1770void Sema::CheckFormatString(const StringLiteral *FExpr, 1771 const Expr *OrigFormatExpr, 1772 const CallExpr *TheCall, bool HasVAListArg, 1773 unsigned format_idx, unsigned firstDataArg, 1774 bool isPrintf) { 1775 1776 // CHECK: is the format string a wide literal? 1777 if (FExpr->isWide()) { 1778 Diag(FExpr->getLocStart(), 1779 diag::warn_format_string_is_wide_literal) 1780 << OrigFormatExpr->getSourceRange(); 1781 return; 1782 } 1783 1784 // Str - The format string. NOTE: this is NOT null-terminated! 1785 llvm::StringRef StrRef = FExpr->getString(); 1786 const char *Str = StrRef.data(); 1787 unsigned StrLen = StrRef.size(); 1788 1789 // CHECK: empty format string? 1790 if (StrLen == 0) { 1791 Diag(FExpr->getLocStart(), diag::warn_empty_format_string) 1792 << OrigFormatExpr->getSourceRange(); 1793 return; 1794 } 1795 1796 if (isPrintf) { 1797 CheckPrintfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg, 1798 TheCall->getNumArgs() - firstDataArg, 1799 isa<ObjCStringLiteral>(OrigFormatExpr), Str, 1800 HasVAListArg, TheCall, format_idx); 1801 1802 if (!analyze_format_string::ParsePrintfString(H, Str, Str + StrLen)) 1803 H.DoneProcessing(); 1804 } 1805 else { 1806 CheckScanfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg, 1807 TheCall->getNumArgs() - firstDataArg, 1808 isa<ObjCStringLiteral>(OrigFormatExpr), Str, 1809 HasVAListArg, TheCall, format_idx); 1810 1811 if (!analyze_format_string::ParseScanfString(H, Str, Str + StrLen)) 1812 H.DoneProcessing(); 1813 } 1814} 1815 1816//===--- CHECK: Return Address of Stack Variable --------------------------===// 1817 1818static DeclRefExpr* EvalVal(Expr *E); 1819static DeclRefExpr* EvalAddr(Expr* E); 1820 1821/// CheckReturnStackAddr - Check if a return statement returns the address 1822/// of a stack variable. 1823void 1824Sema::CheckReturnStackAddr(Expr *RetValExp, QualType lhsType, 1825 SourceLocation ReturnLoc) { 1826 1827 // Perform checking for returned stack addresses. 1828 if (lhsType->isPointerType() || lhsType->isBlockPointerType()) { 1829 if (DeclRefExpr *DR = EvalAddr(RetValExp)) 1830 Diag(DR->getLocStart(), diag::warn_ret_stack_addr) 1831 << DR->getDecl()->getDeclName() << RetValExp->getSourceRange(); 1832 1833 // Skip over implicit cast expressions when checking for block expressions. 1834 RetValExp = RetValExp->IgnoreParenCasts(); 1835 1836 if (BlockExpr *C = dyn_cast<BlockExpr>(RetValExp)) 1837 if (C->hasBlockDeclRefExprs()) 1838 Diag(C->getLocStart(), diag::err_ret_local_block) 1839 << C->getSourceRange(); 1840 1841 if (AddrLabelExpr *ALE = dyn_cast<AddrLabelExpr>(RetValExp)) 1842 Diag(ALE->getLocStart(), diag::warn_ret_addr_label) 1843 << ALE->getSourceRange(); 1844 1845 } else if (lhsType->isReferenceType()) { 1846 // Perform checking for stack values returned by reference. 1847 // Check for a reference to the stack 1848 if (DeclRefExpr *DR = EvalVal(RetValExp)) 1849 Diag(DR->getLocStart(), diag::warn_ret_stack_ref) 1850 << DR->getDecl()->getDeclName() << RetValExp->getSourceRange(); 1851 } 1852} 1853 1854/// EvalAddr - EvalAddr and EvalVal are mutually recursive functions that 1855/// check if the expression in a return statement evaluates to an address 1856/// to a location on the stack. The recursion is used to traverse the 1857/// AST of the return expression, with recursion backtracking when we 1858/// encounter a subexpression that (1) clearly does not lead to the address 1859/// of a stack variable or (2) is something we cannot determine leads to 1860/// the address of a stack variable based on such local checking. 1861/// 1862/// EvalAddr processes expressions that are pointers that are used as 1863/// references (and not L-values). EvalVal handles all other values. 1864/// At the base case of the recursion is a check for a DeclRefExpr* in 1865/// the refers to a stack variable. 1866/// 1867/// This implementation handles: 1868/// 1869/// * pointer-to-pointer casts 1870/// * implicit conversions from array references to pointers 1871/// * taking the address of fields 1872/// * arbitrary interplay between "&" and "*" operators 1873/// * pointer arithmetic from an address of a stack variable 1874/// * taking the address of an array element where the array is on the stack 1875static DeclRefExpr* EvalAddr(Expr *E) { 1876 // We should only be called for evaluating pointer expressions. 1877 assert((E->getType()->isAnyPointerType() || 1878 E->getType()->isBlockPointerType() || 1879 E->getType()->isObjCQualifiedIdType()) && 1880 "EvalAddr only works on pointers"); 1881 1882 // Our "symbolic interpreter" is just a dispatch off the currently 1883 // viewed AST node. We then recursively traverse the AST by calling 1884 // EvalAddr and EvalVal appropriately. 1885 switch (E->getStmtClass()) { 1886 case Stmt::ParenExprClass: 1887 // Ignore parentheses. 1888 return EvalAddr(cast<ParenExpr>(E)->getSubExpr()); 1889 1890 case Stmt::UnaryOperatorClass: { 1891 // The only unary operator that make sense to handle here 1892 // is AddrOf. All others don't make sense as pointers. 1893 UnaryOperator *U = cast<UnaryOperator>(E); 1894 1895 if (U->getOpcode() == UO_AddrOf) 1896 return EvalVal(U->getSubExpr()); 1897 else 1898 return NULL; 1899 } 1900 1901 case Stmt::BinaryOperatorClass: { 1902 // Handle pointer arithmetic. All other binary operators are not valid 1903 // in this context. 1904 BinaryOperator *B = cast<BinaryOperator>(E); 1905 BinaryOperatorKind op = B->getOpcode(); 1906 1907 if (op != BO_Add && op != BO_Sub) 1908 return NULL; 1909 1910 Expr *Base = B->getLHS(); 1911 1912 // Determine which argument is the real pointer base. It could be 1913 // the RHS argument instead of the LHS. 1914 if (!Base->getType()->isPointerType()) Base = B->getRHS(); 1915 1916 assert (Base->getType()->isPointerType()); 1917 return EvalAddr(Base); 1918 } 1919 1920 // For conditional operators we need to see if either the LHS or RHS are 1921 // valid DeclRefExpr*s. If one of them is valid, we return it. 1922 case Stmt::ConditionalOperatorClass: { 1923 ConditionalOperator *C = cast<ConditionalOperator>(E); 1924 1925 // Handle the GNU extension for missing LHS. 1926 if (Expr *lhsExpr = C->getLHS()) { 1927 // In C++, we can have a throw-expression, which has 'void' type. 1928 if (!lhsExpr->getType()->isVoidType()) 1929 if (DeclRefExpr* LHS = EvalAddr(lhsExpr)) 1930 return LHS; 1931 } 1932 1933 // In C++, we can have a throw-expression, which has 'void' type. 1934 if (C->getRHS()->getType()->isVoidType()) 1935 return NULL; 1936 1937 return EvalAddr(C->getRHS()); 1938 } 1939 1940 // For casts, we need to handle conversions from arrays to 1941 // pointer values, and pointer-to-pointer conversions. 1942 case Stmt::ImplicitCastExprClass: 1943 case Stmt::CStyleCastExprClass: 1944 case Stmt::CXXFunctionalCastExprClass: { 1945 Expr* SubExpr = cast<CastExpr>(E)->getSubExpr(); 1946 QualType T = SubExpr->getType(); 1947 1948 if (SubExpr->getType()->isPointerType() || 1949 SubExpr->getType()->isBlockPointerType() || 1950 SubExpr->getType()->isObjCQualifiedIdType()) 1951 return EvalAddr(SubExpr); 1952 else if (T->isArrayType()) 1953 return EvalVal(SubExpr); 1954 else 1955 return 0; 1956 } 1957 1958 // C++ casts. For dynamic casts, static casts, and const casts, we 1959 // are always converting from a pointer-to-pointer, so we just blow 1960 // through the cast. In the case the dynamic cast doesn't fail (and 1961 // return NULL), we take the conservative route and report cases 1962 // where we return the address of a stack variable. For Reinterpre 1963 // FIXME: The comment about is wrong; we're not always converting 1964 // from pointer to pointer. I'm guessing that this code should also 1965 // handle references to objects. 1966 case Stmt::CXXStaticCastExprClass: 1967 case Stmt::CXXDynamicCastExprClass: 1968 case Stmt::CXXConstCastExprClass: 1969 case Stmt::CXXReinterpretCastExprClass: { 1970 Expr *S = cast<CXXNamedCastExpr>(E)->getSubExpr(); 1971 if (S->getType()->isPointerType() || S->getType()->isBlockPointerType()) 1972 return EvalAddr(S); 1973 else 1974 return NULL; 1975 } 1976 1977 // Everything else: we simply don't reason about them. 1978 default: 1979 return NULL; 1980 } 1981} 1982 1983 1984/// EvalVal - This function is complements EvalAddr in the mutual recursion. 1985/// See the comments for EvalAddr for more details. 1986static DeclRefExpr* EvalVal(Expr *E) { 1987do { 1988 // We should only be called for evaluating non-pointer expressions, or 1989 // expressions with a pointer type that are not used as references but instead 1990 // are l-values (e.g., DeclRefExpr with a pointer type). 1991 1992 // Our "symbolic interpreter" is just a dispatch off the currently 1993 // viewed AST node. We then recursively traverse the AST by calling 1994 // EvalAddr and EvalVal appropriately. 1995 switch (E->getStmtClass()) { 1996 case Stmt::ImplicitCastExprClass: { 1997 ImplicitCastExpr *IE = cast<ImplicitCastExpr>(E); 1998 if (IE->getValueKind() == VK_LValue) { 1999 E = IE->getSubExpr(); 2000 continue; 2001 } 2002 return NULL; 2003 } 2004 2005 case Stmt::DeclRefExprClass: { 2006 // DeclRefExpr: the base case. When we hit a DeclRefExpr we are looking 2007 // at code that refers to a variable's name. We check if it has local 2008 // storage within the function, and if so, return the expression. 2009 DeclRefExpr *DR = cast<DeclRefExpr>(E); 2010 2011 if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl())) 2012 if (V->hasLocalStorage() && !V->getType()->isReferenceType()) return DR; 2013 2014 return NULL; 2015 } 2016 2017 case Stmt::ParenExprClass: { 2018 // Ignore parentheses. 2019 E = cast<ParenExpr>(E)->getSubExpr(); 2020 continue; 2021 } 2022 2023 case Stmt::UnaryOperatorClass: { 2024 // The only unary operator that make sense to handle here 2025 // is Deref. All others don't resolve to a "name." This includes 2026 // handling all sorts of rvalues passed to a unary operator. 2027 UnaryOperator *U = cast<UnaryOperator>(E); 2028 2029 if (U->getOpcode() == UO_Deref) 2030 return EvalAddr(U->getSubExpr()); 2031 2032 return NULL; 2033 } 2034 2035 case Stmt::ArraySubscriptExprClass: { 2036 // Array subscripts are potential references to data on the stack. We 2037 // retrieve the DeclRefExpr* for the array variable if it indeed 2038 // has local storage. 2039 return EvalAddr(cast<ArraySubscriptExpr>(E)->getBase()); 2040 } 2041 2042 case Stmt::ConditionalOperatorClass: { 2043 // For conditional operators we need to see if either the LHS or RHS are 2044 // non-NULL DeclRefExpr's. If one is non-NULL, we return it. 2045 ConditionalOperator *C = cast<ConditionalOperator>(E); 2046 2047 // Handle the GNU extension for missing LHS. 2048 if (Expr *lhsExpr = C->getLHS()) 2049 if (DeclRefExpr *LHS = EvalVal(lhsExpr)) 2050 return LHS; 2051 2052 return EvalVal(C->getRHS()); 2053 } 2054 2055 // Accesses to members are potential references to data on the stack. 2056 case Stmt::MemberExprClass: { 2057 MemberExpr *M = cast<MemberExpr>(E); 2058 2059 // Check for indirect access. We only want direct field accesses. 2060 if (M->isArrow()) 2061 return NULL; 2062 2063 // Check whether the member type is itself a reference, in which case 2064 // we're not going to refer to the member, but to what the member refers to. 2065 if (M->getMemberDecl()->getType()->isReferenceType()) 2066 return NULL; 2067 2068 return EvalVal(M->getBase()); 2069 } 2070 2071 // Everything else: we simply don't reason about them. 2072 default: 2073 return NULL; 2074 } 2075} while (true); 2076} 2077 2078//===--- CHECK: Floating-Point comparisons (-Wfloat-equal) ---------------===// 2079 2080/// Check for comparisons of floating point operands using != and ==. 2081/// Issue a warning if these are no self-comparisons, as they are not likely 2082/// to do what the programmer intended. 2083void Sema::CheckFloatComparison(SourceLocation loc, Expr* lex, Expr *rex) { 2084 bool EmitWarning = true; 2085 2086 Expr* LeftExprSansParen = lex->IgnoreParens(); 2087 Expr* RightExprSansParen = rex->IgnoreParens(); 2088 2089 // Special case: check for x == x (which is OK). 2090 // Do not emit warnings for such cases. 2091 if (DeclRefExpr* DRL = dyn_cast<DeclRefExpr>(LeftExprSansParen)) 2092 if (DeclRefExpr* DRR = dyn_cast<DeclRefExpr>(RightExprSansParen)) 2093 if (DRL->getDecl() == DRR->getDecl()) 2094 EmitWarning = false; 2095 2096 2097 // Special case: check for comparisons against literals that can be exactly 2098 // represented by APFloat. In such cases, do not emit a warning. This 2099 // is a heuristic: often comparison against such literals are used to 2100 // detect if a value in a variable has not changed. This clearly can 2101 // lead to false negatives. 2102 if (EmitWarning) { 2103 if (FloatingLiteral* FLL = dyn_cast<FloatingLiteral>(LeftExprSansParen)) { 2104 if (FLL->isExact()) 2105 EmitWarning = false; 2106 } else 2107 if (FloatingLiteral* FLR = dyn_cast<FloatingLiteral>(RightExprSansParen)){ 2108 if (FLR->isExact()) 2109 EmitWarning = false; 2110 } 2111 } 2112 2113 // Check for comparisons with builtin types. 2114 if (EmitWarning) 2115 if (CallExpr* CL = dyn_cast<CallExpr>(LeftExprSansParen)) 2116 if (CL->isBuiltinCall(Context)) 2117 EmitWarning = false; 2118 2119 if (EmitWarning) 2120 if (CallExpr* CR = dyn_cast<CallExpr>(RightExprSansParen)) 2121 if (CR->isBuiltinCall(Context)) 2122 EmitWarning = false; 2123 2124 // Emit the diagnostic. 2125 if (EmitWarning) 2126 Diag(loc, diag::warn_floatingpoint_eq) 2127 << lex->getSourceRange() << rex->getSourceRange(); 2128} 2129 2130//===--- CHECK: Integer mixed-sign comparisons (-Wsign-compare) --------===// 2131//===--- CHECK: Lossy implicit conversions (-Wconversion) --------------===// 2132 2133namespace { 2134 2135/// Structure recording the 'active' range of an integer-valued 2136/// expression. 2137struct IntRange { 2138 /// The number of bits active in the int. 2139 unsigned Width; 2140 2141 /// True if the int is known not to have negative values. 2142 bool NonNegative; 2143 2144 IntRange(unsigned Width, bool NonNegative) 2145 : Width(Width), NonNegative(NonNegative) 2146 {} 2147 2148 /// Returns the range of the bool type. 2149 static IntRange forBoolType() { 2150 return IntRange(1, true); 2151 } 2152 2153 /// Returns the range of an opaque value of the given integral type. 2154 static IntRange forValueOfType(ASTContext &C, QualType T) { 2155 return forValueOfCanonicalType(C, 2156 T->getCanonicalTypeInternal().getTypePtr()); 2157 } 2158 2159 /// Returns the range of an opaque value of a canonical integral type. 2160 static IntRange forValueOfCanonicalType(ASTContext &C, const Type *T) { 2161 assert(T->isCanonicalUnqualified()); 2162 2163 if (const VectorType *VT = dyn_cast<VectorType>(T)) 2164 T = VT->getElementType().getTypePtr(); 2165 if (const ComplexType *CT = dyn_cast<ComplexType>(T)) 2166 T = CT->getElementType().getTypePtr(); 2167 2168 // For enum types, use the known bit width of the enumerators. 2169 if (const EnumType *ET = dyn_cast<EnumType>(T)) { 2170 EnumDecl *Enum = ET->getDecl(); 2171 if (!Enum->isDefinition()) 2172 return IntRange(C.getIntWidth(QualType(T, 0)), false); 2173 2174 unsigned NumPositive = Enum->getNumPositiveBits(); 2175 unsigned NumNegative = Enum->getNumNegativeBits(); 2176 2177 return IntRange(std::max(NumPositive, NumNegative), NumNegative == 0); 2178 } 2179 2180 const BuiltinType *BT = cast<BuiltinType>(T); 2181 assert(BT->isInteger()); 2182 2183 return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger()); 2184 } 2185 2186 /// Returns the "target" range of a canonical integral type, i.e. 2187 /// the range of values expressible in the type. 2188 /// 2189 /// This matches forValueOfCanonicalType except that enums have the 2190 /// full range of their type, not the range of their enumerators. 2191 static IntRange forTargetOfCanonicalType(ASTContext &C, const Type *T) { 2192 assert(T->isCanonicalUnqualified()); 2193 2194 if (const VectorType *VT = dyn_cast<VectorType>(T)) 2195 T = VT->getElementType().getTypePtr(); 2196 if (const ComplexType *CT = dyn_cast<ComplexType>(T)) 2197 T = CT->getElementType().getTypePtr(); 2198 if (const EnumType *ET = dyn_cast<EnumType>(T)) 2199 T = ET->getDecl()->getIntegerType().getTypePtr(); 2200 2201 const BuiltinType *BT = cast<BuiltinType>(T); 2202 assert(BT->isInteger()); 2203 2204 return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger()); 2205 } 2206 2207 /// Returns the supremum of two ranges: i.e. their conservative merge. 2208 static IntRange join(IntRange L, IntRange R) { 2209 return IntRange(std::max(L.Width, R.Width), 2210 L.NonNegative && R.NonNegative); 2211 } 2212 2213 /// Returns the infinum of two ranges: i.e. their aggressive merge. 2214 static IntRange meet(IntRange L, IntRange R) { 2215 return IntRange(std::min(L.Width, R.Width), 2216 L.NonNegative || R.NonNegative); 2217 } 2218}; 2219 2220IntRange GetValueRange(ASTContext &C, llvm::APSInt &value, unsigned MaxWidth) { 2221 if (value.isSigned() && value.isNegative()) 2222 return IntRange(value.getMinSignedBits(), false); 2223 2224 if (value.getBitWidth() > MaxWidth) 2225 value.trunc(MaxWidth); 2226 2227 // isNonNegative() just checks the sign bit without considering 2228 // signedness. 2229 return IntRange(value.getActiveBits(), true); 2230} 2231 2232IntRange GetValueRange(ASTContext &C, APValue &result, QualType Ty, 2233 unsigned MaxWidth) { 2234 if (result.isInt()) 2235 return GetValueRange(C, result.getInt(), MaxWidth); 2236 2237 if (result.isVector()) { 2238 IntRange R = GetValueRange(C, result.getVectorElt(0), Ty, MaxWidth); 2239 for (unsigned i = 1, e = result.getVectorLength(); i != e; ++i) { 2240 IntRange El = GetValueRange(C, result.getVectorElt(i), Ty, MaxWidth); 2241 R = IntRange::join(R, El); 2242 } 2243 return R; 2244 } 2245 2246 if (result.isComplexInt()) { 2247 IntRange R = GetValueRange(C, result.getComplexIntReal(), MaxWidth); 2248 IntRange I = GetValueRange(C, result.getComplexIntImag(), MaxWidth); 2249 return IntRange::join(R, I); 2250 } 2251 2252 // This can happen with lossless casts to intptr_t of "based" lvalues. 2253 // Assume it might use arbitrary bits. 2254 // FIXME: The only reason we need to pass the type in here is to get 2255 // the sign right on this one case. It would be nice if APValue 2256 // preserved this. 2257 assert(result.isLValue()); 2258 return IntRange(MaxWidth, Ty->isUnsignedIntegerType()); 2259} 2260 2261/// Pseudo-evaluate the given integer expression, estimating the 2262/// range of values it might take. 2263/// 2264/// \param MaxWidth - the width to which the value will be truncated 2265IntRange GetExprRange(ASTContext &C, Expr *E, unsigned MaxWidth) { 2266 E = E->IgnoreParens(); 2267 2268 // Try a full evaluation first. 2269 Expr::EvalResult result; 2270 if (E->Evaluate(result, C)) 2271 return GetValueRange(C, result.Val, E->getType(), MaxWidth); 2272 2273 // I think we only want to look through implicit casts here; if the 2274 // user has an explicit widening cast, we should treat the value as 2275 // being of the new, wider type. 2276 if (ImplicitCastExpr *CE = dyn_cast<ImplicitCastExpr>(E)) { 2277 if (CE->getCastKind() == CK_NoOp) 2278 return GetExprRange(C, CE->getSubExpr(), MaxWidth); 2279 2280 IntRange OutputTypeRange = IntRange::forValueOfType(C, CE->getType()); 2281 2282 bool isIntegerCast = (CE->getCastKind() == CK_IntegralCast); 2283 if (!isIntegerCast && CE->getCastKind() == CK_Unknown) 2284 isIntegerCast = CE->getSubExpr()->getType()->isIntegerType(); 2285 2286 // Assume that non-integer casts can span the full range of the type. 2287 if (!isIntegerCast) 2288 return OutputTypeRange; 2289 2290 IntRange SubRange 2291 = GetExprRange(C, CE->getSubExpr(), 2292 std::min(MaxWidth, OutputTypeRange.Width)); 2293 2294 // Bail out if the subexpr's range is as wide as the cast type. 2295 if (SubRange.Width >= OutputTypeRange.Width) 2296 return OutputTypeRange; 2297 2298 // Otherwise, we take the smaller width, and we're non-negative if 2299 // either the output type or the subexpr is. 2300 return IntRange(SubRange.Width, 2301 SubRange.NonNegative || OutputTypeRange.NonNegative); 2302 } 2303 2304 if (ConditionalOperator *CO = dyn_cast<ConditionalOperator>(E)) { 2305 // If we can fold the condition, just take that operand. 2306 bool CondResult; 2307 if (CO->getCond()->EvaluateAsBooleanCondition(CondResult, C)) 2308 return GetExprRange(C, CondResult ? CO->getTrueExpr() 2309 : CO->getFalseExpr(), 2310 MaxWidth); 2311 2312 // Otherwise, conservatively merge. 2313 IntRange L = GetExprRange(C, CO->getTrueExpr(), MaxWidth); 2314 IntRange R = GetExprRange(C, CO->getFalseExpr(), MaxWidth); 2315 return IntRange::join(L, R); 2316 } 2317 2318 if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) { 2319 switch (BO->getOpcode()) { 2320 2321 // Boolean-valued operations are single-bit and positive. 2322 case BO_LAnd: 2323 case BO_LOr: 2324 case BO_LT: 2325 case BO_GT: 2326 case BO_LE: 2327 case BO_GE: 2328 case BO_EQ: 2329 case BO_NE: 2330 return IntRange::forBoolType(); 2331 2332 // The type of these compound assignments is the type of the LHS, 2333 // so the RHS is not necessarily an integer. 2334 case BO_MulAssign: 2335 case BO_DivAssign: 2336 case BO_RemAssign: 2337 case BO_AddAssign: 2338 case BO_SubAssign: 2339 return IntRange::forValueOfType(C, E->getType()); 2340 2341 // Operations with opaque sources are black-listed. 2342 case BO_PtrMemD: 2343 case BO_PtrMemI: 2344 return IntRange::forValueOfType(C, E->getType()); 2345 2346 // Bitwise-and uses the *infinum* of the two source ranges. 2347 case BO_And: 2348 case BO_AndAssign: 2349 return IntRange::meet(GetExprRange(C, BO->getLHS(), MaxWidth), 2350 GetExprRange(C, BO->getRHS(), MaxWidth)); 2351 2352 // Left shift gets black-listed based on a judgement call. 2353 case BO_Shl: 2354 // ...except that we want to treat '1 << (blah)' as logically 2355 // positive. It's an important idiom. 2356 if (IntegerLiteral *I 2357 = dyn_cast<IntegerLiteral>(BO->getLHS()->IgnoreParenCasts())) { 2358 if (I->getValue() == 1) { 2359 IntRange R = IntRange::forValueOfType(C, E->getType()); 2360 return IntRange(R.Width, /*NonNegative*/ true); 2361 } 2362 } 2363 // fallthrough 2364 2365 case BO_ShlAssign: 2366 return IntRange::forValueOfType(C, E->getType()); 2367 2368 // Right shift by a constant can narrow its left argument. 2369 case BO_Shr: 2370 case BO_ShrAssign: { 2371 IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth); 2372 2373 // If the shift amount is a positive constant, drop the width by 2374 // that much. 2375 llvm::APSInt shift; 2376 if (BO->getRHS()->isIntegerConstantExpr(shift, C) && 2377 shift.isNonNegative()) { 2378 unsigned zext = shift.getZExtValue(); 2379 if (zext >= L.Width) 2380 L.Width = (L.NonNegative ? 0 : 1); 2381 else 2382 L.Width -= zext; 2383 } 2384 2385 return L; 2386 } 2387 2388 // Comma acts as its right operand. 2389 case BO_Comma: 2390 return GetExprRange(C, BO->getRHS(), MaxWidth); 2391 2392 // Black-list pointer subtractions. 2393 case BO_Sub: 2394 if (BO->getLHS()->getType()->isPointerType()) 2395 return IntRange::forValueOfType(C, E->getType()); 2396 // fallthrough 2397 2398 default: 2399 break; 2400 } 2401 2402 // Treat every other operator as if it were closed on the 2403 // narrowest type that encompasses both operands. 2404 IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth); 2405 IntRange R = GetExprRange(C, BO->getRHS(), MaxWidth); 2406 return IntRange::join(L, R); 2407 } 2408 2409 if (UnaryOperator *UO = dyn_cast<UnaryOperator>(E)) { 2410 switch (UO->getOpcode()) { 2411 // Boolean-valued operations are white-listed. 2412 case UO_LNot: 2413 return IntRange::forBoolType(); 2414 2415 // Operations with opaque sources are black-listed. 2416 case UO_Deref: 2417 case UO_AddrOf: // should be impossible 2418 return IntRange::forValueOfType(C, E->getType()); 2419 2420 default: 2421 return GetExprRange(C, UO->getSubExpr(), MaxWidth); 2422 } 2423 } 2424 2425 if (dyn_cast<OffsetOfExpr>(E)) { 2426 IntRange::forValueOfType(C, E->getType()); 2427 } 2428 2429 FieldDecl *BitField = E->getBitField(); 2430 if (BitField) { 2431 llvm::APSInt BitWidthAP = BitField->getBitWidth()->EvaluateAsInt(C); 2432 unsigned BitWidth = BitWidthAP.getZExtValue(); 2433 2434 return IntRange(BitWidth, BitField->getType()->isUnsignedIntegerType()); 2435 } 2436 2437 return IntRange::forValueOfType(C, E->getType()); 2438} 2439 2440IntRange GetExprRange(ASTContext &C, Expr *E) { 2441 return GetExprRange(C, E, C.getIntWidth(E->getType())); 2442} 2443 2444/// Checks whether the given value, which currently has the given 2445/// source semantics, has the same value when coerced through the 2446/// target semantics. 2447bool IsSameFloatAfterCast(const llvm::APFloat &value, 2448 const llvm::fltSemantics &Src, 2449 const llvm::fltSemantics &Tgt) { 2450 llvm::APFloat truncated = value; 2451 2452 bool ignored; 2453 truncated.convert(Src, llvm::APFloat::rmNearestTiesToEven, &ignored); 2454 truncated.convert(Tgt, llvm::APFloat::rmNearestTiesToEven, &ignored); 2455 2456 return truncated.bitwiseIsEqual(value); 2457} 2458 2459/// Checks whether the given value, which currently has the given 2460/// source semantics, has the same value when coerced through the 2461/// target semantics. 2462/// 2463/// The value might be a vector of floats (or a complex number). 2464bool IsSameFloatAfterCast(const APValue &value, 2465 const llvm::fltSemantics &Src, 2466 const llvm::fltSemantics &Tgt) { 2467 if (value.isFloat()) 2468 return IsSameFloatAfterCast(value.getFloat(), Src, Tgt); 2469 2470 if (value.isVector()) { 2471 for (unsigned i = 0, e = value.getVectorLength(); i != e; ++i) 2472 if (!IsSameFloatAfterCast(value.getVectorElt(i), Src, Tgt)) 2473 return false; 2474 return true; 2475 } 2476 2477 assert(value.isComplexFloat()); 2478 return (IsSameFloatAfterCast(value.getComplexFloatReal(), Src, Tgt) && 2479 IsSameFloatAfterCast(value.getComplexFloatImag(), Src, Tgt)); 2480} 2481 2482void AnalyzeImplicitConversions(Sema &S, Expr *E, SourceLocation CC); 2483 2484static bool IsZero(Sema &S, Expr *E) { 2485 // Suppress cases where we are comparing against an enum constant. 2486 if (const DeclRefExpr *DR = 2487 dyn_cast<DeclRefExpr>(E->IgnoreParenImpCasts())) 2488 if (isa<EnumConstantDecl>(DR->getDecl())) 2489 return false; 2490 2491 // Suppress cases where the '0' value is expanded from a macro. 2492 if (E->getLocStart().isMacroID()) 2493 return false; 2494 2495 llvm::APSInt Value; 2496 return E->isIntegerConstantExpr(Value, S.Context) && Value == 0; 2497} 2498 2499static bool HasEnumType(Expr *E) { 2500 // Strip off implicit integral promotions. 2501 while (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(E)) { 2502 if (ICE->getCastKind() != CK_IntegralCast && 2503 ICE->getCastKind() != CK_NoOp) 2504 break; 2505 E = ICE->getSubExpr(); 2506 } 2507 2508 return E->getType()->isEnumeralType(); 2509} 2510 2511void CheckTrivialUnsignedComparison(Sema &S, BinaryOperator *E) { 2512 BinaryOperatorKind op = E->getOpcode(); 2513 if (op == BO_LT && IsZero(S, E->getRHS())) { 2514 S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison) 2515 << "< 0" << "false" << HasEnumType(E->getLHS()) 2516 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 2517 } else if (op == BO_GE && IsZero(S, E->getRHS())) { 2518 S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison) 2519 << ">= 0" << "true" << HasEnumType(E->getLHS()) 2520 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 2521 } else if (op == BO_GT && IsZero(S, E->getLHS())) { 2522 S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison) 2523 << "0 >" << "false" << HasEnumType(E->getRHS()) 2524 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 2525 } else if (op == BO_LE && IsZero(S, E->getLHS())) { 2526 S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison) 2527 << "0 <=" << "true" << HasEnumType(E->getRHS()) 2528 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 2529 } 2530} 2531 2532/// Analyze the operands of the given comparison. Implements the 2533/// fallback case from AnalyzeComparison. 2534void AnalyzeImpConvsInComparison(Sema &S, BinaryOperator *E) { 2535 AnalyzeImplicitConversions(S, E->getLHS(), E->getOperatorLoc()); 2536 AnalyzeImplicitConversions(S, E->getRHS(), E->getOperatorLoc()); 2537} 2538 2539/// \brief Implements -Wsign-compare. 2540/// 2541/// \param lex the left-hand expression 2542/// \param rex the right-hand expression 2543/// \param OpLoc the location of the joining operator 2544/// \param BinOpc binary opcode or 0 2545void AnalyzeComparison(Sema &S, BinaryOperator *E) { 2546 // The type the comparison is being performed in. 2547 QualType T = E->getLHS()->getType(); 2548 assert(S.Context.hasSameUnqualifiedType(T, E->getRHS()->getType()) 2549 && "comparison with mismatched types"); 2550 2551 // We don't do anything special if this isn't an unsigned integral 2552 // comparison: we're only interested in integral comparisons, and 2553 // signed comparisons only happen in cases we don't care to warn about. 2554 if (!T->hasUnsignedIntegerRepresentation()) 2555 return AnalyzeImpConvsInComparison(S, E); 2556 2557 Expr *lex = E->getLHS()->IgnoreParenImpCasts(); 2558 Expr *rex = E->getRHS()->IgnoreParenImpCasts(); 2559 2560 // Check to see if one of the (unmodified) operands is of different 2561 // signedness. 2562 Expr *signedOperand, *unsignedOperand; 2563 if (lex->getType()->hasSignedIntegerRepresentation()) { 2564 assert(!rex->getType()->hasSignedIntegerRepresentation() && 2565 "unsigned comparison between two signed integer expressions?"); 2566 signedOperand = lex; 2567 unsignedOperand = rex; 2568 } else if (rex->getType()->hasSignedIntegerRepresentation()) { 2569 signedOperand = rex; 2570 unsignedOperand = lex; 2571 } else { 2572 CheckTrivialUnsignedComparison(S, E); 2573 return AnalyzeImpConvsInComparison(S, E); 2574 } 2575 2576 // Otherwise, calculate the effective range of the signed operand. 2577 IntRange signedRange = GetExprRange(S.Context, signedOperand); 2578 2579 // Go ahead and analyze implicit conversions in the operands. Note 2580 // that we skip the implicit conversions on both sides. 2581 AnalyzeImplicitConversions(S, lex, E->getOperatorLoc()); 2582 AnalyzeImplicitConversions(S, rex, E->getOperatorLoc()); 2583 2584 // If the signed range is non-negative, -Wsign-compare won't fire, 2585 // but we should still check for comparisons which are always true 2586 // or false. 2587 if (signedRange.NonNegative) 2588 return CheckTrivialUnsignedComparison(S, E); 2589 2590 // For (in)equality comparisons, if the unsigned operand is a 2591 // constant which cannot collide with a overflowed signed operand, 2592 // then reinterpreting the signed operand as unsigned will not 2593 // change the result of the comparison. 2594 if (E->isEqualityOp()) { 2595 unsigned comparisonWidth = S.Context.getIntWidth(T); 2596 IntRange unsignedRange = GetExprRange(S.Context, unsignedOperand); 2597 2598 // We should never be unable to prove that the unsigned operand is 2599 // non-negative. 2600 assert(unsignedRange.NonNegative && "unsigned range includes negative?"); 2601 2602 if (unsignedRange.Width < comparisonWidth) 2603 return; 2604 } 2605 2606 S.Diag(E->getOperatorLoc(), diag::warn_mixed_sign_comparison) 2607 << lex->getType() << rex->getType() 2608 << lex->getSourceRange() << rex->getSourceRange(); 2609} 2610 2611/// Analyze the given simple or compound assignment for warning-worthy 2612/// operations. 2613void AnalyzeAssignment(Sema &S, BinaryOperator *E) { 2614 // Just recurse on the LHS. 2615 AnalyzeImplicitConversions(S, E->getLHS(), E->getOperatorLoc()); 2616 2617 // We want to recurse on the RHS as normal unless we're assigning to 2618 // a bitfield. 2619 if (FieldDecl *Bitfield = E->getLHS()->getBitField()) { 2620 assert(Bitfield->isBitField()); 2621 2622 Expr *RHS = E->getRHS()->IgnoreParenImpCasts(); 2623 2624 llvm::APSInt Width(32); 2625 Expr::EvalResult RHSValue; 2626 if (!Bitfield->isInvalidDecl() && 2627 Bitfield->getBitWidth()->isIntegerConstantExpr(Width, S.Context) && 2628 RHS->Evaluate(RHSValue, S.Context) && RHSValue.Val.isInt()) { 2629 const llvm::APSInt &Value = RHSValue.Val.getInt(); 2630 unsigned OriginalWidth = Value.getBitWidth(); 2631 unsigned FieldWidth = Width.getZExtValue(); 2632 2633 if (OriginalWidth > FieldWidth) { 2634 llvm::APSInt TruncatedValue = Value; 2635 TruncatedValue.trunc(FieldWidth); 2636 2637 // It's fairly common to write values into signed bitfields 2638 // that, if sign-extended, would end up becoming a different 2639 // value. We don't want to warn about that. 2640 if (Value.isSigned() && Value.isNegative()) 2641 TruncatedValue.sext(OriginalWidth); 2642 else 2643 TruncatedValue.zext(OriginalWidth); 2644 2645 if (Value != TruncatedValue) { 2646 std::string PrettyValue = Value.toString(10); 2647 std::string PrettyTrunc = TruncatedValue.toString(10); 2648 2649 S.Diag(E->getOperatorLoc(), 2650 diag::warn_impcast_bitfield_precision_constant) 2651 << PrettyValue << PrettyTrunc << RHS->getType() 2652 << E->getRHS()->getSourceRange(); 2653 2654 // Recurse, ignoring any implicit conversions on the RHS. 2655 return AnalyzeImplicitConversions(S, RHS, E->getOperatorLoc()); 2656 } 2657 } 2658 } 2659 } 2660 2661 AnalyzeImplicitConversions(S, E->getRHS(), E->getOperatorLoc()); 2662} 2663 2664/// Diagnose an implicit cast; purely a helper for CheckImplicitConversion. 2665void DiagnoseImpCast(Sema &S, Expr *E, QualType T, SourceLocation CContext, 2666 unsigned diag) { 2667 S.Diag(E->getExprLoc(), diag) 2668 << E->getType() << T << E->getSourceRange() << SourceRange(CContext); 2669} 2670 2671std::string PrettyPrintInRange(const llvm::APSInt &Value, IntRange Range) { 2672 if (!Range.Width) return "0"; 2673 2674 llvm::APSInt ValueInRange = Value; 2675 ValueInRange.setIsSigned(!Range.NonNegative); 2676 ValueInRange.trunc(Range.Width); 2677 return ValueInRange.toString(10); 2678} 2679 2680void CheckImplicitConversion(Sema &S, Expr *E, QualType T, 2681 SourceLocation CC, bool *ICContext = 0) { 2682 if (E->isTypeDependent() || E->isValueDependent()) return; 2683 2684 const Type *Source = S.Context.getCanonicalType(E->getType()).getTypePtr(); 2685 const Type *Target = S.Context.getCanonicalType(T).getTypePtr(); 2686 if (Source == Target) return; 2687 if (Target->isDependentType()) return; 2688 2689 // If the conversion context location is invalid or instantiated 2690 // from a system macro, don't complain. 2691 if (CC.isInvalid() || 2692 (CC.isMacroID() && S.Context.getSourceManager().isInSystemHeader( 2693 S.Context.getSourceManager().getSpellingLoc(CC)))) 2694 return; 2695 2696 // Never diagnose implicit casts to bool. 2697 if (Target->isSpecificBuiltinType(BuiltinType::Bool)) 2698 return; 2699 2700 // Strip vector types. 2701 if (isa<VectorType>(Source)) { 2702 if (!isa<VectorType>(Target)) 2703 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_vector_scalar); 2704 2705 Source = cast<VectorType>(Source)->getElementType().getTypePtr(); 2706 Target = cast<VectorType>(Target)->getElementType().getTypePtr(); 2707 } 2708 2709 // Strip complex types. 2710 if (isa<ComplexType>(Source)) { 2711 if (!isa<ComplexType>(Target)) 2712 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_complex_scalar); 2713 2714 Source = cast<ComplexType>(Source)->getElementType().getTypePtr(); 2715 Target = cast<ComplexType>(Target)->getElementType().getTypePtr(); 2716 } 2717 2718 const BuiltinType *SourceBT = dyn_cast<BuiltinType>(Source); 2719 const BuiltinType *TargetBT = dyn_cast<BuiltinType>(Target); 2720 2721 // If the source is floating point... 2722 if (SourceBT && SourceBT->isFloatingPoint()) { 2723 // ...and the target is floating point... 2724 if (TargetBT && TargetBT->isFloatingPoint()) { 2725 // ...then warn if we're dropping FP rank. 2726 2727 // Builtin FP kinds are ordered by increasing FP rank. 2728 if (SourceBT->getKind() > TargetBT->getKind()) { 2729 // Don't warn about float constants that are precisely 2730 // representable in the target type. 2731 Expr::EvalResult result; 2732 if (E->Evaluate(result, S.Context)) { 2733 // Value might be a float, a float vector, or a float complex. 2734 if (IsSameFloatAfterCast(result.Val, 2735 S.Context.getFloatTypeSemantics(QualType(TargetBT, 0)), 2736 S.Context.getFloatTypeSemantics(QualType(SourceBT, 0)))) 2737 return; 2738 } 2739 2740 DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_float_precision); 2741 } 2742 return; 2743 } 2744 2745 // If the target is integral, always warn. 2746 if ((TargetBT && TargetBT->isInteger())) 2747 // TODO: don't warn for integer values? 2748 DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_float_integer); 2749 2750 return; 2751 } 2752 2753 if (!Source->isIntegerType() || !Target->isIntegerType()) 2754 return; 2755 2756 IntRange SourceRange = GetExprRange(S.Context, E); 2757 IntRange TargetRange = IntRange::forTargetOfCanonicalType(S.Context, Target); 2758 2759 if (SourceRange.Width > TargetRange.Width) { 2760 // If the source is a constant, use a default-on diagnostic. 2761 // TODO: this should happen for bitfield stores, too. 2762 llvm::APSInt Value(32); 2763 if (E->isIntegerConstantExpr(Value, S.Context)) { 2764 std::string PrettySourceValue = Value.toString(10); 2765 std::string PrettyTargetValue = PrettyPrintInRange(Value, TargetRange); 2766 2767 S.Diag(E->getExprLoc(), diag::warn_impcast_integer_precision_constant) 2768 << PrettySourceValue << PrettyTargetValue 2769 << E->getType() << T << E->getSourceRange() << clang::SourceRange(CC); 2770 return; 2771 } 2772 2773 // People want to build with -Wshorten-64-to-32 and not -Wconversion 2774 // and by god we'll let them. 2775 if (SourceRange.Width == 64 && TargetRange.Width == 32) 2776 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_integer_64_32); 2777 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_integer_precision); 2778 } 2779 2780 if ((TargetRange.NonNegative && !SourceRange.NonNegative) || 2781 (!TargetRange.NonNegative && SourceRange.NonNegative && 2782 SourceRange.Width == TargetRange.Width)) { 2783 unsigned DiagID = diag::warn_impcast_integer_sign; 2784 2785 // Traditionally, gcc has warned about this under -Wsign-compare. 2786 // We also want to warn about it in -Wconversion. 2787 // So if -Wconversion is off, use a completely identical diagnostic 2788 // in the sign-compare group. 2789 // The conditional-checking code will 2790 if (ICContext) { 2791 DiagID = diag::warn_impcast_integer_sign_conditional; 2792 *ICContext = true; 2793 } 2794 2795 return DiagnoseImpCast(S, E, T, CC, DiagID); 2796 } 2797 2798 return; 2799} 2800 2801void CheckConditionalOperator(Sema &S, ConditionalOperator *E, QualType T); 2802 2803void CheckConditionalOperand(Sema &S, Expr *E, QualType T, 2804 SourceLocation CC, bool &ICContext) { 2805 E = E->IgnoreParenImpCasts(); 2806 2807 if (isa<ConditionalOperator>(E)) 2808 return CheckConditionalOperator(S, cast<ConditionalOperator>(E), T); 2809 2810 AnalyzeImplicitConversions(S, E, CC); 2811 if (E->getType() != T) 2812 return CheckImplicitConversion(S, E, T, CC, &ICContext); 2813 return; 2814} 2815 2816void CheckConditionalOperator(Sema &S, ConditionalOperator *E, QualType T) { 2817 SourceLocation CC = E->getQuestionLoc(); 2818 2819 AnalyzeImplicitConversions(S, E->getCond(), CC); 2820 2821 bool Suspicious = false; 2822 CheckConditionalOperand(S, E->getTrueExpr(), T, CC, Suspicious); 2823 CheckConditionalOperand(S, E->getFalseExpr(), T, CC, Suspicious); 2824 2825 // If -Wconversion would have warned about either of the candidates 2826 // for a signedness conversion to the context type... 2827 if (!Suspicious) return; 2828 2829 // ...but it's currently ignored... 2830 if (S.Diags.getDiagnosticLevel(diag::warn_impcast_integer_sign_conditional)) 2831 return; 2832 2833 // ...and -Wsign-compare isn't... 2834 if (!S.Diags.getDiagnosticLevel(diag::warn_mixed_sign_conditional)) 2835 return; 2836 2837 // ...then check whether it would have warned about either of the 2838 // candidates for a signedness conversion to the condition type. 2839 if (E->getType() != T) { 2840 Suspicious = false; 2841 CheckImplicitConversion(S, E->getTrueExpr()->IgnoreParenImpCasts(), 2842 E->getType(), CC, &Suspicious); 2843 if (!Suspicious) 2844 CheckImplicitConversion(S, E->getFalseExpr()->IgnoreParenImpCasts(), 2845 E->getType(), CC, &Suspicious); 2846 if (!Suspicious) 2847 return; 2848 } 2849 2850 // If so, emit a diagnostic under -Wsign-compare. 2851 Expr *lex = E->getTrueExpr()->IgnoreParenImpCasts(); 2852 Expr *rex = E->getFalseExpr()->IgnoreParenImpCasts(); 2853 S.Diag(E->getQuestionLoc(), diag::warn_mixed_sign_conditional) 2854 << lex->getType() << rex->getType() 2855 << lex->getSourceRange() << rex->getSourceRange(); 2856} 2857 2858/// AnalyzeImplicitConversions - Find and report any interesting 2859/// implicit conversions in the given expression. There are a couple 2860/// of competing diagnostics here, -Wconversion and -Wsign-compare. 2861void AnalyzeImplicitConversions(Sema &S, Expr *OrigE, SourceLocation CC) { 2862 QualType T = OrigE->getType(); 2863 Expr *E = OrigE->IgnoreParenImpCasts(); 2864 2865 // For conditional operators, we analyze the arguments as if they 2866 // were being fed directly into the output. 2867 if (isa<ConditionalOperator>(E)) { 2868 ConditionalOperator *CO = cast<ConditionalOperator>(E); 2869 CheckConditionalOperator(S, CO, T); 2870 return; 2871 } 2872 2873 // Go ahead and check any implicit conversions we might have skipped. 2874 // The non-canonical typecheck is just an optimization; 2875 // CheckImplicitConversion will filter out dead implicit conversions. 2876 if (E->getType() != T) 2877 CheckImplicitConversion(S, E, T, CC); 2878 2879 // Now continue drilling into this expression. 2880 2881 // Skip past explicit casts. 2882 if (isa<ExplicitCastExpr>(E)) { 2883 E = cast<ExplicitCastExpr>(E)->getSubExpr()->IgnoreParenImpCasts(); 2884 return AnalyzeImplicitConversions(S, E, CC); 2885 } 2886 2887 if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) { 2888 // Do a somewhat different check with comparison operators. 2889 if (BO->isComparisonOp()) 2890 return AnalyzeComparison(S, BO); 2891 2892 // And with assignments and compound assignments. 2893 if (BO->isAssignmentOp()) 2894 return AnalyzeAssignment(S, BO); 2895 } 2896 2897 // These break the otherwise-useful invariant below. Fortunately, 2898 // we don't really need to recurse into them, because any internal 2899 // expressions should have been analyzed already when they were 2900 // built into statements. 2901 if (isa<StmtExpr>(E)) return; 2902 2903 // Don't descend into unevaluated contexts. 2904 if (isa<SizeOfAlignOfExpr>(E)) return; 2905 2906 // Now just recurse over the expression's children. 2907 CC = E->getExprLoc(); 2908 for (Stmt::child_iterator I = E->child_begin(), IE = E->child_end(); 2909 I != IE; ++I) 2910 AnalyzeImplicitConversions(S, cast<Expr>(*I), CC); 2911} 2912 2913} // end anonymous namespace 2914 2915/// Diagnoses "dangerous" implicit conversions within the given 2916/// expression (which is a full expression). Implements -Wconversion 2917/// and -Wsign-compare. 2918/// 2919/// \param CC the "context" location of the implicit conversion, i.e. 2920/// the most location of the syntactic entity requiring the implicit 2921/// conversion 2922void Sema::CheckImplicitConversions(Expr *E, SourceLocation CC) { 2923 // Don't diagnose in unevaluated contexts. 2924 if (ExprEvalContexts.back().Context == Sema::Unevaluated) 2925 return; 2926 2927 // Don't diagnose for value- or type-dependent expressions. 2928 if (E->isTypeDependent() || E->isValueDependent()) 2929 return; 2930 2931 // This is not the right CC for (e.g.) a variable initialization. 2932 AnalyzeImplicitConversions(*this, E, CC); 2933} 2934 2935/// CheckParmsForFunctionDef - Check that the parameters of the given 2936/// function are appropriate for the definition of a function. This 2937/// takes care of any checks that cannot be performed on the 2938/// declaration itself, e.g., that the types of each of the function 2939/// parameters are complete. 2940bool Sema::CheckParmsForFunctionDef(ParmVarDecl **P, ParmVarDecl **PEnd, 2941 bool CheckParameterNames) { 2942 bool HasInvalidParm = false; 2943 for (; P != PEnd; ++P) { 2944 ParmVarDecl *Param = *P; 2945 2946 // C99 6.7.5.3p4: the parameters in a parameter type list in a 2947 // function declarator that is part of a function definition of 2948 // that function shall not have incomplete type. 2949 // 2950 // This is also C++ [dcl.fct]p6. 2951 if (!Param->isInvalidDecl() && 2952 RequireCompleteType(Param->getLocation(), Param->getType(), 2953 diag::err_typecheck_decl_incomplete_type)) { 2954 Param->setInvalidDecl(); 2955 HasInvalidParm = true; 2956 } 2957 2958 // C99 6.9.1p5: If the declarator includes a parameter type list, the 2959 // declaration of each parameter shall include an identifier. 2960 if (CheckParameterNames && 2961 Param->getIdentifier() == 0 && 2962 !Param->isImplicit() && 2963 !getLangOptions().CPlusPlus) 2964 Diag(Param->getLocation(), diag::err_parameter_name_omitted); 2965 2966 // C99 6.7.5.3p12: 2967 // If the function declarator is not part of a definition of that 2968 // function, parameters may have incomplete type and may use the [*] 2969 // notation in their sequences of declarator specifiers to specify 2970 // variable length array types. 2971 QualType PType = Param->getOriginalType(); 2972 if (const ArrayType *AT = Context.getAsArrayType(PType)) { 2973 if (AT->getSizeModifier() == ArrayType::Star) { 2974 // FIXME: This diagnosic should point the the '[*]' if source-location 2975 // information is added for it. 2976 Diag(Param->getLocation(), diag::err_array_star_in_function_definition); 2977 } 2978 } 2979 } 2980 2981 return HasInvalidParm; 2982} 2983 2984/// CheckCastAlign - Implements -Wcast-align, which warns when a 2985/// pointer cast increases the alignment requirements. 2986void Sema::CheckCastAlign(Expr *Op, QualType T, SourceRange TRange) { 2987 // This is actually a lot of work to potentially be doing on every 2988 // cast; don't do it if we're ignoring -Wcast_align (as is the default). 2989 if (getDiagnostics().getDiagnosticLevel(diag::warn_cast_align) 2990 == Diagnostic::Ignored) 2991 return; 2992 2993 // Ignore dependent types. 2994 if (T->isDependentType() || Op->getType()->isDependentType()) 2995 return; 2996 2997 // Require that the destination be a pointer type. 2998 const PointerType *DestPtr = T->getAs<PointerType>(); 2999 if (!DestPtr) return; 3000 3001 // If the destination has alignment 1, we're done. 3002 QualType DestPointee = DestPtr->getPointeeType(); 3003 if (DestPointee->isIncompleteType()) return; 3004 CharUnits DestAlign = Context.getTypeAlignInChars(DestPointee); 3005 if (DestAlign.isOne()) return; 3006 3007 // Require that the source be a pointer type. 3008 const PointerType *SrcPtr = Op->getType()->getAs<PointerType>(); 3009 if (!SrcPtr) return; 3010 QualType SrcPointee = SrcPtr->getPointeeType(); 3011 3012 // Whitelist casts from cv void*. We already implicitly 3013 // whitelisted casts to cv void*, since they have alignment 1. 3014 // Also whitelist casts involving incomplete types, which implicitly 3015 // includes 'void'. 3016 if (SrcPointee->isIncompleteType()) return; 3017 3018 CharUnits SrcAlign = Context.getTypeAlignInChars(SrcPointee); 3019 if (SrcAlign >= DestAlign) return; 3020 3021 Diag(TRange.getBegin(), diag::warn_cast_align) 3022 << Op->getType() << T 3023 << static_cast<unsigned>(SrcAlign.getQuantity()) 3024 << static_cast<unsigned>(DestAlign.getQuantity()) 3025 << TRange << Op->getSourceRange(); 3026} 3027 3028