SemaChecking.cpp revision 00e1f6f5fed4f33f67181578be0232accd342282
19be387c94ff8199f8031b7f11f06c52cce5ccf6djvr//===--- SemaChecking.cpp - Extra Semantic Checking -----------------------===// 22b06aaa2a6bcd363c25fb0c43f6bb906906594bdBehdad Esfahbod// 3035bc7da2318f0dcacecba2e1e24c0a4bf3cf80cBehdad Esfahbod// The LLVM Compiler Infrastructure 47842e56b97ce677b83bdab09cda48bc2d89ac75aJust// 57842e56b97ce677b83bdab09cda48bc2d89ac75aJust// This file is distributed under the University of Illinois Open Source 6d3d7250fc3147b6333e365e5be4ff29aa92ee506pabs// License. See LICENSE.TXT for details. 77842e56b97ce677b83bdab09cda48bc2d89ac75aJust// 87842e56b97ce677b83bdab09cda48bc2d89ac75aJust//===----------------------------------------------------------------------===// 97842e56b97ce677b83bdab09cda48bc2d89ac75aJust// 107842e56b97ce677b83bdab09cda48bc2d89ac75aJust// This file implements extra semantic analysis beyond what is enforced 117842e56b97ce677b83bdab09cda48bc2d89ac75aJust// by the C type system. 127842e56b97ce677b83bdab09cda48bc2d89ac75aJust// 137842e56b97ce677b83bdab09cda48bc2d89ac75aJust//===----------------------------------------------------------------------===// 147842e56b97ce677b83bdab09cda48bc2d89ac75aJust 157842e56b97ce677b83bdab09cda48bc2d89ac75aJust#include "clang/Sema/Initialization.h" 167842e56b97ce677b83bdab09cda48bc2d89ac75aJust#include "clang/Sema/Sema.h" 17035bc7da2318f0dcacecba2e1e24c0a4bf3cf80cBehdad Esfahbod#include "clang/Sema/SemaInternal.h" 187842e56b97ce677b83bdab09cda48bc2d89ac75aJust#include "clang/Sema/Initialization.h" 19035bc7da2318f0dcacecba2e1e24c0a4bf3cf80cBehdad Esfahbod#include "clang/Sema/ScopeInfo.h" 20035bc7da2318f0dcacecba2e1e24c0a4bf3cf80cBehdad Esfahbod#include "clang/Analysis/Analyses/FormatString.h" 21035bc7da2318f0dcacecba2e1e24c0a4bf3cf80cBehdad Esfahbod#include "clang/AST/ASTContext.h" 22035bc7da2318f0dcacecba2e1e24c0a4bf3cf80cBehdad Esfahbod#include "clang/AST/CharUnits.h" 23180ace6a5ff1399ec53bc696e8bef7cce6eef39aBehdad Esfahbod#include "clang/AST/DeclCXX.h" 24035bc7da2318f0dcacecba2e1e24c0a4bf3cf80cBehdad Esfahbod#include "clang/AST/DeclObjC.h" 257842e56b97ce677b83bdab09cda48bc2d89ac75aJust#include "clang/AST/ExprCXX.h" 26035bc7da2318f0dcacecba2e1e24c0a4bf3cf80cBehdad Esfahbod#include "clang/AST/ExprObjC.h" 27035bc7da2318f0dcacecba2e1e24c0a4bf3cf80cBehdad Esfahbod#include "clang/AST/EvaluatedExprVisitor.h" 28035bc7da2318f0dcacecba2e1e24c0a4bf3cf80cBehdad Esfahbod#include "clang/AST/DeclObjC.h" 29035bc7da2318f0dcacecba2e1e24c0a4bf3cf80cBehdad Esfahbod#include "clang/AST/StmtCXX.h" 30180ace6a5ff1399ec53bc696e8bef7cce6eef39aBehdad Esfahbod#include "clang/AST/StmtObjC.h" 31035bc7da2318f0dcacecba2e1e24c0a4bf3cf80cBehdad Esfahbod#include "clang/Lex/Preprocessor.h" 327842e56b97ce677b83bdab09cda48bc2d89ac75aJust#include "llvm/ADT/BitVector.h" 33731e4377e13b91d906d78f6e1e8c7ca2097c71e8Just#include "llvm/ADT/SmallString.h" 347842e56b97ce677b83bdab09cda48bc2d89ac75aJust#include "llvm/ADT/STLExtras.h" 35035bc7da2318f0dcacecba2e1e24c0a4bf3cf80cBehdad Esfahbod#include "llvm/Support/raw_ostream.h" 367842e56b97ce677b83bdab09cda48bc2d89ac75aJust#include "clang/Basic/TargetBuiltins.h" 37035bc7da2318f0dcacecba2e1e24c0a4bf3cf80cBehdad Esfahbod#include "clang/Basic/TargetInfo.h" 38035bc7da2318f0dcacecba2e1e24c0a4bf3cf80cBehdad Esfahbod#include "clang/Basic/ConvertUTF.h" 39035bc7da2318f0dcacecba2e1e24c0a4bf3cf80cBehdad Esfahbod#include <limits> 40035bc7da2318f0dcacecba2e1e24c0a4bf3cf80cBehdad Esfahbodusing namespace clang; 41035bc7da2318f0dcacecba2e1e24c0a4bf3cf80cBehdad Esfahbodusing namespace sema; 427842e56b97ce677b83bdab09cda48bc2d89ac75aJust 437842e56b97ce677b83bdab09cda48bc2d89ac75aJustSourceLocation Sema::getLocationOfStringLiteralByte(const StringLiteral *SL, 447842e56b97ce677b83bdab09cda48bc2d89ac75aJust unsigned ByteNo) const { 457842e56b97ce677b83bdab09cda48bc2d89ac75aJust return SL->getLocationOfByte(ByteNo, PP.getSourceManager(), 467842e56b97ce677b83bdab09cda48bc2d89ac75aJust PP.getLangOpts(), PP.getTargetInfo()); 477842e56b97ce677b83bdab09cda48bc2d89ac75aJust} 487842e56b97ce677b83bdab09cda48bc2d89ac75aJust 497842e56b97ce677b83bdab09cda48bc2d89ac75aJust/// Checks that a call expression's argument count is the desired number. 50035bc7da2318f0dcacecba2e1e24c0a4bf3cf80cBehdad Esfahbod/// This is useful when doing custom type-checking. Returns true on error. 511288f8ad41b183ccdc3a8657b751ce53e03953a7jvrstatic bool checkArgCount(Sema &S, CallExpr *call, unsigned desiredArgCount) { 52ca4c45681ef2ea9290c6845f8bf61dff281fc5c4jvr unsigned argCount = call->getNumArgs(); 531288f8ad41b183ccdc3a8657b751ce53e03953a7jvr if (argCount == desiredArgCount) return false; 54ca4c45681ef2ea9290c6845f8bf61dff281fc5c4jvr 557842e56b97ce677b83bdab09cda48bc2d89ac75aJust if (argCount < desiredArgCount) 56035bc7da2318f0dcacecba2e1e24c0a4bf3cf80cBehdad Esfahbod return S.Diag(call->getLocEnd(), diag::err_typecheck_call_too_few_args) 577842e56b97ce677b83bdab09cda48bc2d89ac75aJust << 0 /*function call*/ << desiredArgCount << argCount 587842e56b97ce677b83bdab09cda48bc2d89ac75aJust << call->getSourceRange(); 597842e56b97ce677b83bdab09cda48bc2d89ac75aJust 604c2f1b6a2c2e4d85d4987606795097d592b1586aBehdad Esfahbod // Highlight all the excess arguments. 614c2f1b6a2c2e4d85d4987606795097d592b1586aBehdad Esfahbod SourceRange range(call->getArg(desiredArgCount)->getLocStart(), 624c2f1b6a2c2e4d85d4987606795097d592b1586aBehdad Esfahbod call->getArg(argCount - 1)->getLocEnd()); 634c2f1b6a2c2e4d85d4987606795097d592b1586aBehdad Esfahbod 64180ace6a5ff1399ec53bc696e8bef7cce6eef39aBehdad Esfahbod return S.Diag(range.getBegin(), diag::err_typecheck_call_too_many_args) 654c2f1b6a2c2e4d85d4987606795097d592b1586aBehdad Esfahbod << 0 /*function call*/ << desiredArgCount << argCount 664c2f1b6a2c2e4d85d4987606795097d592b1586aBehdad Esfahbod << call->getArg(1)->getSourceRange(); 677842e56b97ce677b83bdab09cda48bc2d89ac75aJust} 68035bc7da2318f0dcacecba2e1e24c0a4bf3cf80cBehdad Esfahbod 69180ace6a5ff1399ec53bc696e8bef7cce6eef39aBehdad Esfahbod/// Check that the first argument to __builtin_annotation is an integer 70035bc7da2318f0dcacecba2e1e24c0a4bf3cf80cBehdad Esfahbod/// and the second argument is a non-wide string literal. 717842e56b97ce677b83bdab09cda48bc2d89ac75aJuststatic bool SemaBuiltinAnnotation(Sema &S, CallExpr *TheCall) { 727842e56b97ce677b83bdab09cda48bc2d89ac75aJust if (checkArgCount(S, TheCall, 2)) 737842e56b97ce677b83bdab09cda48bc2d89ac75aJust return true; 747842e56b97ce677b83bdab09cda48bc2d89ac75aJust 757842e56b97ce677b83bdab09cda48bc2d89ac75aJust // First argument should be an integer. 767842e56b97ce677b83bdab09cda48bc2d89ac75aJust Expr *ValArg = TheCall->getArg(0); 777842e56b97ce677b83bdab09cda48bc2d89ac75aJust QualType Ty = ValArg->getType(); 787842e56b97ce677b83bdab09cda48bc2d89ac75aJust if (!Ty->isIntegerType()) { 797842e56b97ce677b83bdab09cda48bc2d89ac75aJust S.Diag(ValArg->getLocStart(), diag::err_builtin_annotation_first_arg) 807842e56b97ce677b83bdab09cda48bc2d89ac75aJust << ValArg->getSourceRange(); 817842e56b97ce677b83bdab09cda48bc2d89ac75aJust return true; 827842e56b97ce677b83bdab09cda48bc2d89ac75aJust } 837842e56b97ce677b83bdab09cda48bc2d89ac75aJust 847842e56b97ce677b83bdab09cda48bc2d89ac75aJust // Second argument should be a constant string. 857842e56b97ce677b83bdab09cda48bc2d89ac75aJust Expr *StrArg = TheCall->getArg(1)->IgnoreParenCasts(); 863a9fd301808f5a8991ca9ac44028d1ecb22d307fBehdad Esfahbod StringLiteral *Literal = dyn_cast<StringLiteral>(StrArg); 877842e56b97ce677b83bdab09cda48bc2d89ac75aJust if (!Literal || !Literal->isAscii()) { 887842e56b97ce677b83bdab09cda48bc2d89ac75aJust S.Diag(StrArg->getLocStart(), diag::err_builtin_annotation_second_arg) 897842e56b97ce677b83bdab09cda48bc2d89ac75aJust << StrArg->getSourceRange(); 907842e56b97ce677b83bdab09cda48bc2d89ac75aJust return true; 917842e56b97ce677b83bdab09cda48bc2d89ac75aJust } 927842e56b97ce677b83bdab09cda48bc2d89ac75aJust 937842e56b97ce677b83bdab09cda48bc2d89ac75aJust TheCall->setType(Ty); 947842e56b97ce677b83bdab09cda48bc2d89ac75aJust return false; 957842e56b97ce677b83bdab09cda48bc2d89ac75aJust} 963a9fd301808f5a8991ca9ac44028d1ecb22d307fBehdad Esfahbod 973a9fd301808f5a8991ca9ac44028d1ecb22d307fBehdad EsfahbodExprResult 987842e56b97ce677b83bdab09cda48bc2d89ac75aJustSema::CheckBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) { 99 ExprResult TheCallResult(Owned(TheCall)); 100 101 // Find out if any arguments are required to be integer constant expressions. 102 unsigned ICEArguments = 0; 103 ASTContext::GetBuiltinTypeError Error; 104 Context.GetBuiltinType(BuiltinID, Error, &ICEArguments); 105 if (Error != ASTContext::GE_None) 106 ICEArguments = 0; // Don't diagnose previously diagnosed errors. 107 108 // If any arguments are required to be ICE's, check and diagnose. 109 for (unsigned ArgNo = 0; ICEArguments != 0; ++ArgNo) { 110 // Skip arguments not required to be ICE's. 111 if ((ICEArguments & (1 << ArgNo)) == 0) continue; 112 113 llvm::APSInt Result; 114 if (SemaBuiltinConstantArg(TheCall, ArgNo, Result)) 115 return true; 116 ICEArguments &= ~(1 << ArgNo); 117 } 118 119 switch (BuiltinID) { 120 case Builtin::BI__builtin___CFStringMakeConstantString: 121 assert(TheCall->getNumArgs() == 1 && 122 "Wrong # arguments to builtin CFStringMakeConstantString"); 123 if (CheckObjCString(TheCall->getArg(0))) 124 return ExprError(); 125 break; 126 case Builtin::BI__builtin_stdarg_start: 127 case Builtin::BI__builtin_va_start: 128 if (SemaBuiltinVAStart(TheCall)) 129 return ExprError(); 130 break; 131 case Builtin::BI__builtin_isgreater: 132 case Builtin::BI__builtin_isgreaterequal: 133 case Builtin::BI__builtin_isless: 134 case Builtin::BI__builtin_islessequal: 135 case Builtin::BI__builtin_islessgreater: 136 case Builtin::BI__builtin_isunordered: 137 if (SemaBuiltinUnorderedCompare(TheCall)) 138 return ExprError(); 139 break; 140 case Builtin::BI__builtin_fpclassify: 141 if (SemaBuiltinFPClassification(TheCall, 6)) 142 return ExprError(); 143 break; 144 case Builtin::BI__builtin_isfinite: 145 case Builtin::BI__builtin_isinf: 146 case Builtin::BI__builtin_isinf_sign: 147 case Builtin::BI__builtin_isnan: 148 case Builtin::BI__builtin_isnormal: 149 if (SemaBuiltinFPClassification(TheCall, 1)) 150 return ExprError(); 151 break; 152 case Builtin::BI__builtin_shufflevector: 153 return SemaBuiltinShuffleVector(TheCall); 154 // TheCall will be freed by the smart pointer here, but that's fine, since 155 // SemaBuiltinShuffleVector guts it, but then doesn't release it. 156 case Builtin::BI__builtin_prefetch: 157 if (SemaBuiltinPrefetch(TheCall)) 158 return ExprError(); 159 break; 160 case Builtin::BI__builtin_object_size: 161 if (SemaBuiltinObjectSize(TheCall)) 162 return ExprError(); 163 break; 164 case Builtin::BI__builtin_longjmp: 165 if (SemaBuiltinLongjmp(TheCall)) 166 return ExprError(); 167 break; 168 169 case Builtin::BI__builtin_classify_type: 170 if (checkArgCount(*this, TheCall, 1)) return true; 171 TheCall->setType(Context.IntTy); 172 break; 173 case Builtin::BI__builtin_constant_p: 174 if (checkArgCount(*this, TheCall, 1)) return true; 175 TheCall->setType(Context.IntTy); 176 break; 177 case Builtin::BI__sync_fetch_and_add: 178 case Builtin::BI__sync_fetch_and_add_1: 179 case Builtin::BI__sync_fetch_and_add_2: 180 case Builtin::BI__sync_fetch_and_add_4: 181 case Builtin::BI__sync_fetch_and_add_8: 182 case Builtin::BI__sync_fetch_and_add_16: 183 case Builtin::BI__sync_fetch_and_sub: 184 case Builtin::BI__sync_fetch_and_sub_1: 185 case Builtin::BI__sync_fetch_and_sub_2: 186 case Builtin::BI__sync_fetch_and_sub_4: 187 case Builtin::BI__sync_fetch_and_sub_8: 188 case Builtin::BI__sync_fetch_and_sub_16: 189 case Builtin::BI__sync_fetch_and_or: 190 case Builtin::BI__sync_fetch_and_or_1: 191 case Builtin::BI__sync_fetch_and_or_2: 192 case Builtin::BI__sync_fetch_and_or_4: 193 case Builtin::BI__sync_fetch_and_or_8: 194 case Builtin::BI__sync_fetch_and_or_16: 195 case Builtin::BI__sync_fetch_and_and: 196 case Builtin::BI__sync_fetch_and_and_1: 197 case Builtin::BI__sync_fetch_and_and_2: 198 case Builtin::BI__sync_fetch_and_and_4: 199 case Builtin::BI__sync_fetch_and_and_8: 200 case Builtin::BI__sync_fetch_and_and_16: 201 case Builtin::BI__sync_fetch_and_xor: 202 case Builtin::BI__sync_fetch_and_xor_1: 203 case Builtin::BI__sync_fetch_and_xor_2: 204 case Builtin::BI__sync_fetch_and_xor_4: 205 case Builtin::BI__sync_fetch_and_xor_8: 206 case Builtin::BI__sync_fetch_and_xor_16: 207 case Builtin::BI__sync_add_and_fetch: 208 case Builtin::BI__sync_add_and_fetch_1: 209 case Builtin::BI__sync_add_and_fetch_2: 210 case Builtin::BI__sync_add_and_fetch_4: 211 case Builtin::BI__sync_add_and_fetch_8: 212 case Builtin::BI__sync_add_and_fetch_16: 213 case Builtin::BI__sync_sub_and_fetch: 214 case Builtin::BI__sync_sub_and_fetch_1: 215 case Builtin::BI__sync_sub_and_fetch_2: 216 case Builtin::BI__sync_sub_and_fetch_4: 217 case Builtin::BI__sync_sub_and_fetch_8: 218 case Builtin::BI__sync_sub_and_fetch_16: 219 case Builtin::BI__sync_and_and_fetch: 220 case Builtin::BI__sync_and_and_fetch_1: 221 case Builtin::BI__sync_and_and_fetch_2: 222 case Builtin::BI__sync_and_and_fetch_4: 223 case Builtin::BI__sync_and_and_fetch_8: 224 case Builtin::BI__sync_and_and_fetch_16: 225 case Builtin::BI__sync_or_and_fetch: 226 case Builtin::BI__sync_or_and_fetch_1: 227 case Builtin::BI__sync_or_and_fetch_2: 228 case Builtin::BI__sync_or_and_fetch_4: 229 case Builtin::BI__sync_or_and_fetch_8: 230 case Builtin::BI__sync_or_and_fetch_16: 231 case Builtin::BI__sync_xor_and_fetch: 232 case Builtin::BI__sync_xor_and_fetch_1: 233 case Builtin::BI__sync_xor_and_fetch_2: 234 case Builtin::BI__sync_xor_and_fetch_4: 235 case Builtin::BI__sync_xor_and_fetch_8: 236 case Builtin::BI__sync_xor_and_fetch_16: 237 case Builtin::BI__sync_val_compare_and_swap: 238 case Builtin::BI__sync_val_compare_and_swap_1: 239 case Builtin::BI__sync_val_compare_and_swap_2: 240 case Builtin::BI__sync_val_compare_and_swap_4: 241 case Builtin::BI__sync_val_compare_and_swap_8: 242 case Builtin::BI__sync_val_compare_and_swap_16: 243 case Builtin::BI__sync_bool_compare_and_swap: 244 case Builtin::BI__sync_bool_compare_and_swap_1: 245 case Builtin::BI__sync_bool_compare_and_swap_2: 246 case Builtin::BI__sync_bool_compare_and_swap_4: 247 case Builtin::BI__sync_bool_compare_and_swap_8: 248 case Builtin::BI__sync_bool_compare_and_swap_16: 249 case Builtin::BI__sync_lock_test_and_set: 250 case Builtin::BI__sync_lock_test_and_set_1: 251 case Builtin::BI__sync_lock_test_and_set_2: 252 case Builtin::BI__sync_lock_test_and_set_4: 253 case Builtin::BI__sync_lock_test_and_set_8: 254 case Builtin::BI__sync_lock_test_and_set_16: 255 case Builtin::BI__sync_lock_release: 256 case Builtin::BI__sync_lock_release_1: 257 case Builtin::BI__sync_lock_release_2: 258 case Builtin::BI__sync_lock_release_4: 259 case Builtin::BI__sync_lock_release_8: 260 case Builtin::BI__sync_lock_release_16: 261 case Builtin::BI__sync_swap: 262 case Builtin::BI__sync_swap_1: 263 case Builtin::BI__sync_swap_2: 264 case Builtin::BI__sync_swap_4: 265 case Builtin::BI__sync_swap_8: 266 case Builtin::BI__sync_swap_16: 267 return SemaBuiltinAtomicOverloaded(move(TheCallResult)); 268#define BUILTIN(ID, TYPE, ATTRS) 269#define ATOMIC_BUILTIN(ID, TYPE, ATTRS) \ 270 case Builtin::BI##ID: \ 271 return SemaAtomicOpsOverloaded(move(TheCallResult), AtomicExpr::AO##ID); 272#include "clang/Basic/Builtins.def" 273 case Builtin::BI__builtin_annotation: 274 if (SemaBuiltinAnnotation(*this, TheCall)) 275 return ExprError(); 276 break; 277 } 278 279 // Since the target specific builtins for each arch overlap, only check those 280 // of the arch we are compiling for. 281 if (BuiltinID >= Builtin::FirstTSBuiltin) { 282 switch (Context.getTargetInfo().getTriple().getArch()) { 283 case llvm::Triple::arm: 284 case llvm::Triple::thumb: 285 if (CheckARMBuiltinFunctionCall(BuiltinID, TheCall)) 286 return ExprError(); 287 break; 288 default: 289 break; 290 } 291 } 292 293 return move(TheCallResult); 294} 295 296// Get the valid immediate range for the specified NEON type code. 297static unsigned RFT(unsigned t, bool shift = false) { 298 NeonTypeFlags Type(t); 299 int IsQuad = Type.isQuad(); 300 switch (Type.getEltType()) { 301 case NeonTypeFlags::Int8: 302 case NeonTypeFlags::Poly8: 303 return shift ? 7 : (8 << IsQuad) - 1; 304 case NeonTypeFlags::Int16: 305 case NeonTypeFlags::Poly16: 306 return shift ? 15 : (4 << IsQuad) - 1; 307 case NeonTypeFlags::Int32: 308 return shift ? 31 : (2 << IsQuad) - 1; 309 case NeonTypeFlags::Int64: 310 return shift ? 63 : (1 << IsQuad) - 1; 311 case NeonTypeFlags::Float16: 312 assert(!shift && "cannot shift float types!"); 313 return (4 << IsQuad) - 1; 314 case NeonTypeFlags::Float32: 315 assert(!shift && "cannot shift float types!"); 316 return (2 << IsQuad) - 1; 317 } 318 llvm_unreachable("Invalid NeonTypeFlag!"); 319} 320 321/// getNeonEltType - Return the QualType corresponding to the elements of 322/// the vector type specified by the NeonTypeFlags. This is used to check 323/// the pointer arguments for Neon load/store intrinsics. 324static QualType getNeonEltType(NeonTypeFlags Flags, ASTContext &Context) { 325 switch (Flags.getEltType()) { 326 case NeonTypeFlags::Int8: 327 return Flags.isUnsigned() ? Context.UnsignedCharTy : Context.SignedCharTy; 328 case NeonTypeFlags::Int16: 329 return Flags.isUnsigned() ? Context.UnsignedShortTy : Context.ShortTy; 330 case NeonTypeFlags::Int32: 331 return Flags.isUnsigned() ? Context.UnsignedIntTy : Context.IntTy; 332 case NeonTypeFlags::Int64: 333 return Flags.isUnsigned() ? Context.UnsignedLongLongTy : Context.LongLongTy; 334 case NeonTypeFlags::Poly8: 335 return Context.SignedCharTy; 336 case NeonTypeFlags::Poly16: 337 return Context.ShortTy; 338 case NeonTypeFlags::Float16: 339 return Context.UnsignedShortTy; 340 case NeonTypeFlags::Float32: 341 return Context.FloatTy; 342 } 343 llvm_unreachable("Invalid NeonTypeFlag!"); 344} 345 346bool Sema::CheckARMBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) { 347 llvm::APSInt Result; 348 349 unsigned mask = 0; 350 unsigned TV = 0; 351 int PtrArgNum = -1; 352 bool HasConstPtr = false; 353 switch (BuiltinID) { 354#define GET_NEON_OVERLOAD_CHECK 355#include "clang/Basic/arm_neon.inc" 356#undef GET_NEON_OVERLOAD_CHECK 357 } 358 359 // For NEON intrinsics which are overloaded on vector element type, validate 360 // the immediate which specifies which variant to emit. 361 unsigned ImmArg = TheCall->getNumArgs()-1; 362 if (mask) { 363 if (SemaBuiltinConstantArg(TheCall, ImmArg, Result)) 364 return true; 365 366 TV = Result.getLimitedValue(64); 367 if ((TV > 63) || (mask & (1 << TV)) == 0) 368 return Diag(TheCall->getLocStart(), diag::err_invalid_neon_type_code) 369 << TheCall->getArg(ImmArg)->getSourceRange(); 370 } 371 372 if (PtrArgNum >= 0) { 373 // Check that pointer arguments have the specified type. 374 Expr *Arg = TheCall->getArg(PtrArgNum); 375 if (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(Arg)) 376 Arg = ICE->getSubExpr(); 377 ExprResult RHS = DefaultFunctionArrayLvalueConversion(Arg); 378 QualType RHSTy = RHS.get()->getType(); 379 QualType EltTy = getNeonEltType(NeonTypeFlags(TV), Context); 380 if (HasConstPtr) 381 EltTy = EltTy.withConst(); 382 QualType LHSTy = Context.getPointerType(EltTy); 383 AssignConvertType ConvTy; 384 ConvTy = CheckSingleAssignmentConstraints(LHSTy, RHS); 385 if (RHS.isInvalid()) 386 return true; 387 if (DiagnoseAssignmentResult(ConvTy, Arg->getLocStart(), LHSTy, RHSTy, 388 RHS.get(), AA_Assigning)) 389 return true; 390 } 391 392 // For NEON intrinsics which take an immediate value as part of the 393 // instruction, range check them here. 394 unsigned i = 0, l = 0, u = 0; 395 switch (BuiltinID) { 396 default: return false; 397 case ARM::BI__builtin_arm_ssat: i = 1; l = 1; u = 31; break; 398 case ARM::BI__builtin_arm_usat: i = 1; u = 31; break; 399 case ARM::BI__builtin_arm_vcvtr_f: 400 case ARM::BI__builtin_arm_vcvtr_d: i = 1; u = 1; break; 401#define GET_NEON_IMMEDIATE_CHECK 402#include "clang/Basic/arm_neon.inc" 403#undef GET_NEON_IMMEDIATE_CHECK 404 }; 405 406 // Check that the immediate argument is actually a constant. 407 if (SemaBuiltinConstantArg(TheCall, i, Result)) 408 return true; 409 410 // Range check against the upper/lower values for this isntruction. 411 unsigned Val = Result.getZExtValue(); 412 if (Val < l || Val > (u + l)) 413 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 414 << l << u+l << TheCall->getArg(i)->getSourceRange(); 415 416 // FIXME: VFP Intrinsics should error if VFP not present. 417 return false; 418} 419 420/// CheckFunctionCall - Check a direct function call for various correctness 421/// and safety properties not strictly enforced by the C type system. 422bool Sema::CheckFunctionCall(FunctionDecl *FDecl, CallExpr *TheCall) { 423 // Get the IdentifierInfo* for the called function. 424 IdentifierInfo *FnInfo = FDecl->getIdentifier(); 425 426 // None of the checks below are needed for functions that don't have 427 // simple names (e.g., C++ conversion functions). 428 if (!FnInfo) 429 return false; 430 431 // FIXME: This mechanism should be abstracted to be less fragile and 432 // more efficient. For example, just map function ids to custom 433 // handlers. 434 435 // Printf and scanf checking. 436 for (specific_attr_iterator<FormatAttr> 437 i = FDecl->specific_attr_begin<FormatAttr>(), 438 e = FDecl->specific_attr_end<FormatAttr>(); i != e ; ++i) { 439 CheckFormatArguments(*i, TheCall); 440 } 441 442 for (specific_attr_iterator<NonNullAttr> 443 i = FDecl->specific_attr_begin<NonNullAttr>(), 444 e = FDecl->specific_attr_end<NonNullAttr>(); i != e; ++i) { 445 CheckNonNullArguments(*i, TheCall->getArgs(), 446 TheCall->getCallee()->getLocStart()); 447 } 448 449 unsigned CMId = FDecl->getMemoryFunctionKind(); 450 if (CMId == 0) 451 return false; 452 453 // Handle memory setting and copying functions. 454 if (CMId == Builtin::BIstrlcpy || CMId == Builtin::BIstrlcat) 455 CheckStrlcpycatArguments(TheCall, FnInfo); 456 else if (CMId == Builtin::BIstrncat) 457 CheckStrncatArguments(TheCall, FnInfo); 458 else 459 CheckMemaccessArguments(TheCall, CMId, FnInfo); 460 461 return false; 462} 463 464bool Sema::CheckObjCMethodCall(ObjCMethodDecl *Method, SourceLocation lbrac, 465 Expr **Args, unsigned NumArgs) { 466 for (specific_attr_iterator<FormatAttr> 467 i = Method->specific_attr_begin<FormatAttr>(), 468 e = Method->specific_attr_end<FormatAttr>(); i != e ; ++i) { 469 470 CheckFormatArguments(*i, Args, NumArgs, false, lbrac, 471 Method->getSourceRange()); 472 } 473 474 // diagnose nonnull arguments. 475 for (specific_attr_iterator<NonNullAttr> 476 i = Method->specific_attr_begin<NonNullAttr>(), 477 e = Method->specific_attr_end<NonNullAttr>(); i != e; ++i) { 478 CheckNonNullArguments(*i, Args, lbrac); 479 } 480 481 return false; 482} 483 484bool Sema::CheckBlockCall(NamedDecl *NDecl, CallExpr *TheCall) { 485 const VarDecl *V = dyn_cast<VarDecl>(NDecl); 486 if (!V) 487 return false; 488 489 QualType Ty = V->getType(); 490 if (!Ty->isBlockPointerType()) 491 return false; 492 493 // format string checking. 494 for (specific_attr_iterator<FormatAttr> 495 i = NDecl->specific_attr_begin<FormatAttr>(), 496 e = NDecl->specific_attr_end<FormatAttr>(); i != e ; ++i) { 497 CheckFormatArguments(*i, TheCall); 498 } 499 500 return false; 501} 502 503ExprResult Sema::SemaAtomicOpsOverloaded(ExprResult TheCallResult, 504 AtomicExpr::AtomicOp Op) { 505 CallExpr *TheCall = cast<CallExpr>(TheCallResult.get()); 506 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 507 508 // All these operations take one of the following forms: 509 enum { 510 // C __c11_atomic_init(A *, C) 511 Init, 512 // C __c11_atomic_load(A *, int) 513 Load, 514 // void __atomic_load(A *, CP, int) 515 Copy, 516 // C __c11_atomic_add(A *, M, int) 517 Arithmetic, 518 // C __atomic_exchange_n(A *, CP, int) 519 Xchg, 520 // void __atomic_exchange(A *, C *, CP, int) 521 GNUXchg, 522 // bool __c11_atomic_compare_exchange_strong(A *, C *, CP, int, int) 523 C11CmpXchg, 524 // bool __atomic_compare_exchange(A *, C *, CP, bool, int, int) 525 GNUCmpXchg 526 } Form = Init; 527 const unsigned NumArgs[] = { 2, 2, 3, 3, 3, 4, 5, 6 }; 528 const unsigned NumVals[] = { 1, 0, 1, 1, 1, 2, 2, 3 }; 529 // where: 530 // C is an appropriate type, 531 // A is volatile _Atomic(C) for __c11 builtins and is C for GNU builtins, 532 // CP is C for __c11 builtins and GNU _n builtins and is C * otherwise, 533 // M is C if C is an integer, and ptrdiff_t if C is a pointer, and 534 // the int parameters are for orderings. 535 536 assert(AtomicExpr::AO__c11_atomic_init == 0 && 537 AtomicExpr::AO__c11_atomic_fetch_xor + 1 == AtomicExpr::AO__atomic_load 538 && "need to update code for modified C11 atomics"); 539 bool IsC11 = Op >= AtomicExpr::AO__c11_atomic_init && 540 Op <= AtomicExpr::AO__c11_atomic_fetch_xor; 541 bool IsN = Op == AtomicExpr::AO__atomic_load_n || 542 Op == AtomicExpr::AO__atomic_store_n || 543 Op == AtomicExpr::AO__atomic_exchange_n || 544 Op == AtomicExpr::AO__atomic_compare_exchange_n; 545 bool IsAddSub = false; 546 547 switch (Op) { 548 case AtomicExpr::AO__c11_atomic_init: 549 Form = Init; 550 break; 551 552 case AtomicExpr::AO__c11_atomic_load: 553 case AtomicExpr::AO__atomic_load_n: 554 Form = Load; 555 break; 556 557 case AtomicExpr::AO__c11_atomic_store: 558 case AtomicExpr::AO__atomic_load: 559 case AtomicExpr::AO__atomic_store: 560 case AtomicExpr::AO__atomic_store_n: 561 Form = Copy; 562 break; 563 564 case AtomicExpr::AO__c11_atomic_fetch_add: 565 case AtomicExpr::AO__c11_atomic_fetch_sub: 566 case AtomicExpr::AO__atomic_fetch_add: 567 case AtomicExpr::AO__atomic_fetch_sub: 568 case AtomicExpr::AO__atomic_add_fetch: 569 case AtomicExpr::AO__atomic_sub_fetch: 570 IsAddSub = true; 571 // Fall through. 572 case AtomicExpr::AO__c11_atomic_fetch_and: 573 case AtomicExpr::AO__c11_atomic_fetch_or: 574 case AtomicExpr::AO__c11_atomic_fetch_xor: 575 case AtomicExpr::AO__atomic_fetch_and: 576 case AtomicExpr::AO__atomic_fetch_or: 577 case AtomicExpr::AO__atomic_fetch_xor: 578 case AtomicExpr::AO__atomic_fetch_nand: 579 case AtomicExpr::AO__atomic_and_fetch: 580 case AtomicExpr::AO__atomic_or_fetch: 581 case AtomicExpr::AO__atomic_xor_fetch: 582 case AtomicExpr::AO__atomic_nand_fetch: 583 Form = Arithmetic; 584 break; 585 586 case AtomicExpr::AO__c11_atomic_exchange: 587 case AtomicExpr::AO__atomic_exchange_n: 588 Form = Xchg; 589 break; 590 591 case AtomicExpr::AO__atomic_exchange: 592 Form = GNUXchg; 593 break; 594 595 case AtomicExpr::AO__c11_atomic_compare_exchange_strong: 596 case AtomicExpr::AO__c11_atomic_compare_exchange_weak: 597 Form = C11CmpXchg; 598 break; 599 600 case AtomicExpr::AO__atomic_compare_exchange: 601 case AtomicExpr::AO__atomic_compare_exchange_n: 602 Form = GNUCmpXchg; 603 break; 604 } 605 606 // Check we have the right number of arguments. 607 if (TheCall->getNumArgs() < NumArgs[Form]) { 608 Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 609 << 0 << NumArgs[Form] << TheCall->getNumArgs() 610 << TheCall->getCallee()->getSourceRange(); 611 return ExprError(); 612 } else if (TheCall->getNumArgs() > NumArgs[Form]) { 613 Diag(TheCall->getArg(NumArgs[Form])->getLocStart(), 614 diag::err_typecheck_call_too_many_args) 615 << 0 << NumArgs[Form] << TheCall->getNumArgs() 616 << TheCall->getCallee()->getSourceRange(); 617 return ExprError(); 618 } 619 620 // Inspect the first argument of the atomic operation. 621 Expr *Ptr = TheCall->getArg(0); 622 Ptr = DefaultFunctionArrayLvalueConversion(Ptr).get(); 623 const PointerType *pointerType = Ptr->getType()->getAs<PointerType>(); 624 if (!pointerType) { 625 Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer) 626 << Ptr->getType() << Ptr->getSourceRange(); 627 return ExprError(); 628 } 629 630 // For a __c11 builtin, this should be a pointer to an _Atomic type. 631 QualType AtomTy = pointerType->getPointeeType(); // 'A' 632 QualType ValType = AtomTy; // 'C' 633 if (IsC11) { 634 if (!AtomTy->isAtomicType()) { 635 Diag(DRE->getLocStart(), diag::err_atomic_op_needs_atomic) 636 << Ptr->getType() << Ptr->getSourceRange(); 637 return ExprError(); 638 } 639 ValType = AtomTy->getAs<AtomicType>()->getValueType(); 640 } 641 642 // For an arithmetic operation, the implied arithmetic must be well-formed. 643 if (Form == Arithmetic) { 644 // gcc does not enforce these rules for GNU atomics, but we do so for sanity. 645 if (IsAddSub && !ValType->isIntegerType() && !ValType->isPointerType()) { 646 Diag(DRE->getLocStart(), diag::err_atomic_op_needs_atomic_int_or_ptr) 647 << IsC11 << Ptr->getType() << Ptr->getSourceRange(); 648 return ExprError(); 649 } 650 if (!IsAddSub && !ValType->isIntegerType()) { 651 Diag(DRE->getLocStart(), diag::err_atomic_op_bitwise_needs_atomic_int) 652 << IsC11 << Ptr->getType() << Ptr->getSourceRange(); 653 return ExprError(); 654 } 655 } else if (IsN && !ValType->isIntegerType() && !ValType->isPointerType()) { 656 // For __atomic_*_n operations, the value type must be a scalar integral or 657 // pointer type which is 1, 2, 4, 8 or 16 bytes in length. 658 Diag(DRE->getLocStart(), diag::err_atomic_op_needs_atomic_int_or_ptr) 659 << IsC11 << Ptr->getType() << Ptr->getSourceRange(); 660 return ExprError(); 661 } 662 663 if (!IsC11 && !AtomTy.isTriviallyCopyableType(Context)) { 664 // For GNU atomics, require a trivially-copyable type. This is not part of 665 // the GNU atomics specification, but we enforce it for sanity. 666 Diag(DRE->getLocStart(), diag::err_atomic_op_needs_trivial_copy) 667 << Ptr->getType() << Ptr->getSourceRange(); 668 return ExprError(); 669 } 670 671 // FIXME: For any builtin other than a load, the ValType must not be 672 // const-qualified. 673 674 switch (ValType.getObjCLifetime()) { 675 case Qualifiers::OCL_None: 676 case Qualifiers::OCL_ExplicitNone: 677 // okay 678 break; 679 680 case Qualifiers::OCL_Weak: 681 case Qualifiers::OCL_Strong: 682 case Qualifiers::OCL_Autoreleasing: 683 // FIXME: Can this happen? By this point, ValType should be known 684 // to be trivially copyable. 685 Diag(DRE->getLocStart(), diag::err_arc_atomic_ownership) 686 << ValType << Ptr->getSourceRange(); 687 return ExprError(); 688 } 689 690 QualType ResultType = ValType; 691 if (Form == Copy || Form == GNUXchg || Form == Init) 692 ResultType = Context.VoidTy; 693 else if (Form == C11CmpXchg || Form == GNUCmpXchg) 694 ResultType = Context.BoolTy; 695 696 // The type of a parameter passed 'by value'. In the GNU atomics, such 697 // arguments are actually passed as pointers. 698 QualType ByValType = ValType; // 'CP' 699 if (!IsC11 && !IsN) 700 ByValType = Ptr->getType(); 701 702 // The first argument --- the pointer --- has a fixed type; we 703 // deduce the types of the rest of the arguments accordingly. Walk 704 // the remaining arguments, converting them to the deduced value type. 705 for (unsigned i = 1; i != NumArgs[Form]; ++i) { 706 QualType Ty; 707 if (i < NumVals[Form] + 1) { 708 switch (i) { 709 case 1: 710 // The second argument is the non-atomic operand. For arithmetic, this 711 // is always passed by value, and for a compare_exchange it is always 712 // passed by address. For the rest, GNU uses by-address and C11 uses 713 // by-value. 714 assert(Form != Load); 715 if (Form == Init || (Form == Arithmetic && ValType->isIntegerType())) 716 Ty = ValType; 717 else if (Form == Copy || Form == Xchg) 718 Ty = ByValType; 719 else if (Form == Arithmetic) 720 Ty = Context.getPointerDiffType(); 721 else 722 Ty = Context.getPointerType(ValType.getUnqualifiedType()); 723 break; 724 case 2: 725 // The third argument to compare_exchange / GNU exchange is a 726 // (pointer to a) desired value. 727 Ty = ByValType; 728 break; 729 case 3: 730 // The fourth argument to GNU compare_exchange is a 'weak' flag. 731 Ty = Context.BoolTy; 732 break; 733 } 734 } else { 735 // The order(s) are always converted to int. 736 Ty = Context.IntTy; 737 } 738 739 InitializedEntity Entity = 740 InitializedEntity::InitializeParameter(Context, Ty, false); 741 ExprResult Arg = TheCall->getArg(i); 742 Arg = PerformCopyInitialization(Entity, SourceLocation(), Arg); 743 if (Arg.isInvalid()) 744 return true; 745 TheCall->setArg(i, Arg.get()); 746 } 747 748 // Permute the arguments into a 'consistent' order. 749 SmallVector<Expr*, 5> SubExprs; 750 SubExprs.push_back(Ptr); 751 switch (Form) { 752 case Init: 753 // Note, AtomicExpr::getVal1() has a special case for this atomic. 754 SubExprs.push_back(TheCall->getArg(1)); // Val1 755 break; 756 case Load: 757 SubExprs.push_back(TheCall->getArg(1)); // Order 758 break; 759 case Copy: 760 case Arithmetic: 761 case Xchg: 762 SubExprs.push_back(TheCall->getArg(2)); // Order 763 SubExprs.push_back(TheCall->getArg(1)); // Val1 764 break; 765 case GNUXchg: 766 // Note, AtomicExpr::getVal2() has a special case for this atomic. 767 SubExprs.push_back(TheCall->getArg(3)); // Order 768 SubExprs.push_back(TheCall->getArg(1)); // Val1 769 SubExprs.push_back(TheCall->getArg(2)); // Val2 770 break; 771 case C11CmpXchg: 772 SubExprs.push_back(TheCall->getArg(3)); // Order 773 SubExprs.push_back(TheCall->getArg(1)); // Val1 774 SubExprs.push_back(TheCall->getArg(4)); // OrderFail 775 SubExprs.push_back(TheCall->getArg(2)); // Val2 776 break; 777 case GNUCmpXchg: 778 SubExprs.push_back(TheCall->getArg(4)); // Order 779 SubExprs.push_back(TheCall->getArg(1)); // Val1 780 SubExprs.push_back(TheCall->getArg(5)); // OrderFail 781 SubExprs.push_back(TheCall->getArg(2)); // Val2 782 SubExprs.push_back(TheCall->getArg(3)); // Weak 783 break; 784 } 785 786 return Owned(new (Context) AtomicExpr(TheCall->getCallee()->getLocStart(), 787 SubExprs.data(), SubExprs.size(), 788 ResultType, Op, 789 TheCall->getRParenLoc())); 790} 791 792 793/// checkBuiltinArgument - Given a call to a builtin function, perform 794/// normal type-checking on the given argument, updating the call in 795/// place. This is useful when a builtin function requires custom 796/// type-checking for some of its arguments but not necessarily all of 797/// them. 798/// 799/// Returns true on error. 800static bool checkBuiltinArgument(Sema &S, CallExpr *E, unsigned ArgIndex) { 801 FunctionDecl *Fn = E->getDirectCallee(); 802 assert(Fn && "builtin call without direct callee!"); 803 804 ParmVarDecl *Param = Fn->getParamDecl(ArgIndex); 805 InitializedEntity Entity = 806 InitializedEntity::InitializeParameter(S.Context, Param); 807 808 ExprResult Arg = E->getArg(0); 809 Arg = S.PerformCopyInitialization(Entity, SourceLocation(), Arg); 810 if (Arg.isInvalid()) 811 return true; 812 813 E->setArg(ArgIndex, Arg.take()); 814 return false; 815} 816 817/// SemaBuiltinAtomicOverloaded - We have a call to a function like 818/// __sync_fetch_and_add, which is an overloaded function based on the pointer 819/// type of its first argument. The main ActOnCallExpr routines have already 820/// promoted the types of arguments because all of these calls are prototyped as 821/// void(...). 822/// 823/// This function goes through and does final semantic checking for these 824/// builtins, 825ExprResult 826Sema::SemaBuiltinAtomicOverloaded(ExprResult TheCallResult) { 827 CallExpr *TheCall = (CallExpr *)TheCallResult.get(); 828 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 829 FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl()); 830 831 // Ensure that we have at least one argument to do type inference from. 832 if (TheCall->getNumArgs() < 1) { 833 Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args_at_least) 834 << 0 << 1 << TheCall->getNumArgs() 835 << TheCall->getCallee()->getSourceRange(); 836 return ExprError(); 837 } 838 839 // Inspect the first argument of the atomic builtin. This should always be 840 // a pointer type, whose element is an integral scalar or pointer type. 841 // Because it is a pointer type, we don't have to worry about any implicit 842 // casts here. 843 // FIXME: We don't allow floating point scalars as input. 844 Expr *FirstArg = TheCall->getArg(0); 845 ExprResult FirstArgResult = DefaultFunctionArrayLvalueConversion(FirstArg); 846 if (FirstArgResult.isInvalid()) 847 return ExprError(); 848 FirstArg = FirstArgResult.take(); 849 TheCall->setArg(0, FirstArg); 850 851 const PointerType *pointerType = FirstArg->getType()->getAs<PointerType>(); 852 if (!pointerType) { 853 Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer) 854 << FirstArg->getType() << FirstArg->getSourceRange(); 855 return ExprError(); 856 } 857 858 QualType ValType = pointerType->getPointeeType(); 859 if (!ValType->isIntegerType() && !ValType->isAnyPointerType() && 860 !ValType->isBlockPointerType()) { 861 Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer_intptr) 862 << FirstArg->getType() << FirstArg->getSourceRange(); 863 return ExprError(); 864 } 865 866 switch (ValType.getObjCLifetime()) { 867 case Qualifiers::OCL_None: 868 case Qualifiers::OCL_ExplicitNone: 869 // okay 870 break; 871 872 case Qualifiers::OCL_Weak: 873 case Qualifiers::OCL_Strong: 874 case Qualifiers::OCL_Autoreleasing: 875 Diag(DRE->getLocStart(), diag::err_arc_atomic_ownership) 876 << ValType << FirstArg->getSourceRange(); 877 return ExprError(); 878 } 879 880 // Strip any qualifiers off ValType. 881 ValType = ValType.getUnqualifiedType(); 882 883 // The majority of builtins return a value, but a few have special return 884 // types, so allow them to override appropriately below. 885 QualType ResultType = ValType; 886 887 // We need to figure out which concrete builtin this maps onto. For example, 888 // __sync_fetch_and_add with a 2 byte object turns into 889 // __sync_fetch_and_add_2. 890#define BUILTIN_ROW(x) \ 891 { Builtin::BI##x##_1, Builtin::BI##x##_2, Builtin::BI##x##_4, \ 892 Builtin::BI##x##_8, Builtin::BI##x##_16 } 893 894 static const unsigned BuiltinIndices[][5] = { 895 BUILTIN_ROW(__sync_fetch_and_add), 896 BUILTIN_ROW(__sync_fetch_and_sub), 897 BUILTIN_ROW(__sync_fetch_and_or), 898 BUILTIN_ROW(__sync_fetch_and_and), 899 BUILTIN_ROW(__sync_fetch_and_xor), 900 901 BUILTIN_ROW(__sync_add_and_fetch), 902 BUILTIN_ROW(__sync_sub_and_fetch), 903 BUILTIN_ROW(__sync_and_and_fetch), 904 BUILTIN_ROW(__sync_or_and_fetch), 905 BUILTIN_ROW(__sync_xor_and_fetch), 906 907 BUILTIN_ROW(__sync_val_compare_and_swap), 908 BUILTIN_ROW(__sync_bool_compare_and_swap), 909 BUILTIN_ROW(__sync_lock_test_and_set), 910 BUILTIN_ROW(__sync_lock_release), 911 BUILTIN_ROW(__sync_swap) 912 }; 913#undef BUILTIN_ROW 914 915 // Determine the index of the size. 916 unsigned SizeIndex; 917 switch (Context.getTypeSizeInChars(ValType).getQuantity()) { 918 case 1: SizeIndex = 0; break; 919 case 2: SizeIndex = 1; break; 920 case 4: SizeIndex = 2; break; 921 case 8: SizeIndex = 3; break; 922 case 16: SizeIndex = 4; break; 923 default: 924 Diag(DRE->getLocStart(), diag::err_atomic_builtin_pointer_size) 925 << FirstArg->getType() << FirstArg->getSourceRange(); 926 return ExprError(); 927 } 928 929 // Each of these builtins has one pointer argument, followed by some number of 930 // values (0, 1 or 2) followed by a potentially empty varags list of stuff 931 // that we ignore. Find out which row of BuiltinIndices to read from as well 932 // as the number of fixed args. 933 unsigned BuiltinID = FDecl->getBuiltinID(); 934 unsigned BuiltinIndex, NumFixed = 1; 935 switch (BuiltinID) { 936 default: llvm_unreachable("Unknown overloaded atomic builtin!"); 937 case Builtin::BI__sync_fetch_and_add: 938 case Builtin::BI__sync_fetch_and_add_1: 939 case Builtin::BI__sync_fetch_and_add_2: 940 case Builtin::BI__sync_fetch_and_add_4: 941 case Builtin::BI__sync_fetch_and_add_8: 942 case Builtin::BI__sync_fetch_and_add_16: 943 BuiltinIndex = 0; 944 break; 945 946 case Builtin::BI__sync_fetch_and_sub: 947 case Builtin::BI__sync_fetch_and_sub_1: 948 case Builtin::BI__sync_fetch_and_sub_2: 949 case Builtin::BI__sync_fetch_and_sub_4: 950 case Builtin::BI__sync_fetch_and_sub_8: 951 case Builtin::BI__sync_fetch_and_sub_16: 952 BuiltinIndex = 1; 953 break; 954 955 case Builtin::BI__sync_fetch_and_or: 956 case Builtin::BI__sync_fetch_and_or_1: 957 case Builtin::BI__sync_fetch_and_or_2: 958 case Builtin::BI__sync_fetch_and_or_4: 959 case Builtin::BI__sync_fetch_and_or_8: 960 case Builtin::BI__sync_fetch_and_or_16: 961 BuiltinIndex = 2; 962 break; 963 964 case Builtin::BI__sync_fetch_and_and: 965 case Builtin::BI__sync_fetch_and_and_1: 966 case Builtin::BI__sync_fetch_and_and_2: 967 case Builtin::BI__sync_fetch_and_and_4: 968 case Builtin::BI__sync_fetch_and_and_8: 969 case Builtin::BI__sync_fetch_and_and_16: 970 BuiltinIndex = 3; 971 break; 972 973 case Builtin::BI__sync_fetch_and_xor: 974 case Builtin::BI__sync_fetch_and_xor_1: 975 case Builtin::BI__sync_fetch_and_xor_2: 976 case Builtin::BI__sync_fetch_and_xor_4: 977 case Builtin::BI__sync_fetch_and_xor_8: 978 case Builtin::BI__sync_fetch_and_xor_16: 979 BuiltinIndex = 4; 980 break; 981 982 case Builtin::BI__sync_add_and_fetch: 983 case Builtin::BI__sync_add_and_fetch_1: 984 case Builtin::BI__sync_add_and_fetch_2: 985 case Builtin::BI__sync_add_and_fetch_4: 986 case Builtin::BI__sync_add_and_fetch_8: 987 case Builtin::BI__sync_add_and_fetch_16: 988 BuiltinIndex = 5; 989 break; 990 991 case Builtin::BI__sync_sub_and_fetch: 992 case Builtin::BI__sync_sub_and_fetch_1: 993 case Builtin::BI__sync_sub_and_fetch_2: 994 case Builtin::BI__sync_sub_and_fetch_4: 995 case Builtin::BI__sync_sub_and_fetch_8: 996 case Builtin::BI__sync_sub_and_fetch_16: 997 BuiltinIndex = 6; 998 break; 999 1000 case Builtin::BI__sync_and_and_fetch: 1001 case Builtin::BI__sync_and_and_fetch_1: 1002 case Builtin::BI__sync_and_and_fetch_2: 1003 case Builtin::BI__sync_and_and_fetch_4: 1004 case Builtin::BI__sync_and_and_fetch_8: 1005 case Builtin::BI__sync_and_and_fetch_16: 1006 BuiltinIndex = 7; 1007 break; 1008 1009 case Builtin::BI__sync_or_and_fetch: 1010 case Builtin::BI__sync_or_and_fetch_1: 1011 case Builtin::BI__sync_or_and_fetch_2: 1012 case Builtin::BI__sync_or_and_fetch_4: 1013 case Builtin::BI__sync_or_and_fetch_8: 1014 case Builtin::BI__sync_or_and_fetch_16: 1015 BuiltinIndex = 8; 1016 break; 1017 1018 case Builtin::BI__sync_xor_and_fetch: 1019 case Builtin::BI__sync_xor_and_fetch_1: 1020 case Builtin::BI__sync_xor_and_fetch_2: 1021 case Builtin::BI__sync_xor_and_fetch_4: 1022 case Builtin::BI__sync_xor_and_fetch_8: 1023 case Builtin::BI__sync_xor_and_fetch_16: 1024 BuiltinIndex = 9; 1025 break; 1026 1027 case Builtin::BI__sync_val_compare_and_swap: 1028 case Builtin::BI__sync_val_compare_and_swap_1: 1029 case Builtin::BI__sync_val_compare_and_swap_2: 1030 case Builtin::BI__sync_val_compare_and_swap_4: 1031 case Builtin::BI__sync_val_compare_and_swap_8: 1032 case Builtin::BI__sync_val_compare_and_swap_16: 1033 BuiltinIndex = 10; 1034 NumFixed = 2; 1035 break; 1036 1037 case Builtin::BI__sync_bool_compare_and_swap: 1038 case Builtin::BI__sync_bool_compare_and_swap_1: 1039 case Builtin::BI__sync_bool_compare_and_swap_2: 1040 case Builtin::BI__sync_bool_compare_and_swap_4: 1041 case Builtin::BI__sync_bool_compare_and_swap_8: 1042 case Builtin::BI__sync_bool_compare_and_swap_16: 1043 BuiltinIndex = 11; 1044 NumFixed = 2; 1045 ResultType = Context.BoolTy; 1046 break; 1047 1048 case Builtin::BI__sync_lock_test_and_set: 1049 case Builtin::BI__sync_lock_test_and_set_1: 1050 case Builtin::BI__sync_lock_test_and_set_2: 1051 case Builtin::BI__sync_lock_test_and_set_4: 1052 case Builtin::BI__sync_lock_test_and_set_8: 1053 case Builtin::BI__sync_lock_test_and_set_16: 1054 BuiltinIndex = 12; 1055 break; 1056 1057 case Builtin::BI__sync_lock_release: 1058 case Builtin::BI__sync_lock_release_1: 1059 case Builtin::BI__sync_lock_release_2: 1060 case Builtin::BI__sync_lock_release_4: 1061 case Builtin::BI__sync_lock_release_8: 1062 case Builtin::BI__sync_lock_release_16: 1063 BuiltinIndex = 13; 1064 NumFixed = 0; 1065 ResultType = Context.VoidTy; 1066 break; 1067 1068 case Builtin::BI__sync_swap: 1069 case Builtin::BI__sync_swap_1: 1070 case Builtin::BI__sync_swap_2: 1071 case Builtin::BI__sync_swap_4: 1072 case Builtin::BI__sync_swap_8: 1073 case Builtin::BI__sync_swap_16: 1074 BuiltinIndex = 14; 1075 break; 1076 } 1077 1078 // Now that we know how many fixed arguments we expect, first check that we 1079 // have at least that many. 1080 if (TheCall->getNumArgs() < 1+NumFixed) { 1081 Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args_at_least) 1082 << 0 << 1+NumFixed << TheCall->getNumArgs() 1083 << TheCall->getCallee()->getSourceRange(); 1084 return ExprError(); 1085 } 1086 1087 // Get the decl for the concrete builtin from this, we can tell what the 1088 // concrete integer type we should convert to is. 1089 unsigned NewBuiltinID = BuiltinIndices[BuiltinIndex][SizeIndex]; 1090 const char *NewBuiltinName = Context.BuiltinInfo.GetName(NewBuiltinID); 1091 IdentifierInfo *NewBuiltinII = PP.getIdentifierInfo(NewBuiltinName); 1092 FunctionDecl *NewBuiltinDecl = 1093 cast<FunctionDecl>(LazilyCreateBuiltin(NewBuiltinII, NewBuiltinID, 1094 TUScope, false, DRE->getLocStart())); 1095 1096 // The first argument --- the pointer --- has a fixed type; we 1097 // deduce the types of the rest of the arguments accordingly. Walk 1098 // the remaining arguments, converting them to the deduced value type. 1099 for (unsigned i = 0; i != NumFixed; ++i) { 1100 ExprResult Arg = TheCall->getArg(i+1); 1101 1102 // GCC does an implicit conversion to the pointer or integer ValType. This 1103 // can fail in some cases (1i -> int**), check for this error case now. 1104 // Initialize the argument. 1105 InitializedEntity Entity = InitializedEntity::InitializeParameter(Context, 1106 ValType, /*consume*/ false); 1107 Arg = PerformCopyInitialization(Entity, SourceLocation(), Arg); 1108 if (Arg.isInvalid()) 1109 return ExprError(); 1110 1111 // Okay, we have something that *can* be converted to the right type. Check 1112 // to see if there is a potentially weird extension going on here. This can 1113 // happen when you do an atomic operation on something like an char* and 1114 // pass in 42. The 42 gets converted to char. This is even more strange 1115 // for things like 45.123 -> char, etc. 1116 // FIXME: Do this check. 1117 TheCall->setArg(i+1, Arg.take()); 1118 } 1119 1120 ASTContext& Context = this->getASTContext(); 1121 1122 // Create a new DeclRefExpr to refer to the new decl. 1123 DeclRefExpr* NewDRE = DeclRefExpr::Create( 1124 Context, 1125 DRE->getQualifierLoc(), 1126 SourceLocation(), 1127 NewBuiltinDecl, 1128 /*enclosing*/ false, 1129 DRE->getLocation(), 1130 NewBuiltinDecl->getType(), 1131 DRE->getValueKind()); 1132 1133 // Set the callee in the CallExpr. 1134 // FIXME: This leaks the original parens and implicit casts. 1135 ExprResult PromotedCall = UsualUnaryConversions(NewDRE); 1136 if (PromotedCall.isInvalid()) 1137 return ExprError(); 1138 TheCall->setCallee(PromotedCall.take()); 1139 1140 // Change the result type of the call to match the original value type. This 1141 // is arbitrary, but the codegen for these builtins ins design to handle it 1142 // gracefully. 1143 TheCall->setType(ResultType); 1144 1145 return move(TheCallResult); 1146} 1147 1148/// CheckObjCString - Checks that the argument to the builtin 1149/// CFString constructor is correct 1150/// Note: It might also make sense to do the UTF-16 conversion here (would 1151/// simplify the backend). 1152bool Sema::CheckObjCString(Expr *Arg) { 1153 Arg = Arg->IgnoreParenCasts(); 1154 StringLiteral *Literal = dyn_cast<StringLiteral>(Arg); 1155 1156 if (!Literal || !Literal->isAscii()) { 1157 Diag(Arg->getLocStart(), diag::err_cfstring_literal_not_string_constant) 1158 << Arg->getSourceRange(); 1159 return true; 1160 } 1161 1162 if (Literal->containsNonAsciiOrNull()) { 1163 StringRef String = Literal->getString(); 1164 unsigned NumBytes = String.size(); 1165 SmallVector<UTF16, 128> ToBuf(NumBytes); 1166 const UTF8 *FromPtr = (UTF8 *)String.data(); 1167 UTF16 *ToPtr = &ToBuf[0]; 1168 1169 ConversionResult Result = ConvertUTF8toUTF16(&FromPtr, FromPtr + NumBytes, 1170 &ToPtr, ToPtr + NumBytes, 1171 strictConversion); 1172 // Check for conversion failure. 1173 if (Result != conversionOK) 1174 Diag(Arg->getLocStart(), 1175 diag::warn_cfstring_truncated) << Arg->getSourceRange(); 1176 } 1177 return false; 1178} 1179 1180/// SemaBuiltinVAStart - Check the arguments to __builtin_va_start for validity. 1181/// Emit an error and return true on failure, return false on success. 1182bool Sema::SemaBuiltinVAStart(CallExpr *TheCall) { 1183 Expr *Fn = TheCall->getCallee(); 1184 if (TheCall->getNumArgs() > 2) { 1185 Diag(TheCall->getArg(2)->getLocStart(), 1186 diag::err_typecheck_call_too_many_args) 1187 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 1188 << Fn->getSourceRange() 1189 << SourceRange(TheCall->getArg(2)->getLocStart(), 1190 (*(TheCall->arg_end()-1))->getLocEnd()); 1191 return true; 1192 } 1193 1194 if (TheCall->getNumArgs() < 2) { 1195 return Diag(TheCall->getLocEnd(), 1196 diag::err_typecheck_call_too_few_args_at_least) 1197 << 0 /*function call*/ << 2 << TheCall->getNumArgs(); 1198 } 1199 1200 // Type-check the first argument normally. 1201 if (checkBuiltinArgument(*this, TheCall, 0)) 1202 return true; 1203 1204 // Determine whether the current function is variadic or not. 1205 BlockScopeInfo *CurBlock = getCurBlock(); 1206 bool isVariadic; 1207 if (CurBlock) 1208 isVariadic = CurBlock->TheDecl->isVariadic(); 1209 else if (FunctionDecl *FD = getCurFunctionDecl()) 1210 isVariadic = FD->isVariadic(); 1211 else 1212 isVariadic = getCurMethodDecl()->isVariadic(); 1213 1214 if (!isVariadic) { 1215 Diag(Fn->getLocStart(), diag::err_va_start_used_in_non_variadic_function); 1216 return true; 1217 } 1218 1219 // Verify that the second argument to the builtin is the last argument of the 1220 // current function or method. 1221 bool SecondArgIsLastNamedArgument = false; 1222 const Expr *Arg = TheCall->getArg(1)->IgnoreParenCasts(); 1223 1224 if (const DeclRefExpr *DR = dyn_cast<DeclRefExpr>(Arg)) { 1225 if (const ParmVarDecl *PV = dyn_cast<ParmVarDecl>(DR->getDecl())) { 1226 // FIXME: This isn't correct for methods (results in bogus warning). 1227 // Get the last formal in the current function. 1228 const ParmVarDecl *LastArg; 1229 if (CurBlock) 1230 LastArg = *(CurBlock->TheDecl->param_end()-1); 1231 else if (FunctionDecl *FD = getCurFunctionDecl()) 1232 LastArg = *(FD->param_end()-1); 1233 else 1234 LastArg = *(getCurMethodDecl()->param_end()-1); 1235 SecondArgIsLastNamedArgument = PV == LastArg; 1236 } 1237 } 1238 1239 if (!SecondArgIsLastNamedArgument) 1240 Diag(TheCall->getArg(1)->getLocStart(), 1241 diag::warn_second_parameter_of_va_start_not_last_named_argument); 1242 return false; 1243} 1244 1245/// SemaBuiltinUnorderedCompare - Handle functions like __builtin_isgreater and 1246/// friends. This is declared to take (...), so we have to check everything. 1247bool Sema::SemaBuiltinUnorderedCompare(CallExpr *TheCall) { 1248 if (TheCall->getNumArgs() < 2) 1249 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 1250 << 0 << 2 << TheCall->getNumArgs()/*function call*/; 1251 if (TheCall->getNumArgs() > 2) 1252 return Diag(TheCall->getArg(2)->getLocStart(), 1253 diag::err_typecheck_call_too_many_args) 1254 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 1255 << SourceRange(TheCall->getArg(2)->getLocStart(), 1256 (*(TheCall->arg_end()-1))->getLocEnd()); 1257 1258 ExprResult OrigArg0 = TheCall->getArg(0); 1259 ExprResult OrigArg1 = TheCall->getArg(1); 1260 1261 // Do standard promotions between the two arguments, returning their common 1262 // type. 1263 QualType Res = UsualArithmeticConversions(OrigArg0, OrigArg1, false); 1264 if (OrigArg0.isInvalid() || OrigArg1.isInvalid()) 1265 return true; 1266 1267 // Make sure any conversions are pushed back into the call; this is 1268 // type safe since unordered compare builtins are declared as "_Bool 1269 // foo(...)". 1270 TheCall->setArg(0, OrigArg0.get()); 1271 TheCall->setArg(1, OrigArg1.get()); 1272 1273 if (OrigArg0.get()->isTypeDependent() || OrigArg1.get()->isTypeDependent()) 1274 return false; 1275 1276 // If the common type isn't a real floating type, then the arguments were 1277 // invalid for this operation. 1278 if (!Res->isRealFloatingType()) 1279 return Diag(OrigArg0.get()->getLocStart(), 1280 diag::err_typecheck_call_invalid_ordered_compare) 1281 << OrigArg0.get()->getType() << OrigArg1.get()->getType() 1282 << SourceRange(OrigArg0.get()->getLocStart(), OrigArg1.get()->getLocEnd()); 1283 1284 return false; 1285} 1286 1287/// SemaBuiltinSemaBuiltinFPClassification - Handle functions like 1288/// __builtin_isnan and friends. This is declared to take (...), so we have 1289/// to check everything. We expect the last argument to be a floating point 1290/// value. 1291bool Sema::SemaBuiltinFPClassification(CallExpr *TheCall, unsigned NumArgs) { 1292 if (TheCall->getNumArgs() < NumArgs) 1293 return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args) 1294 << 0 << NumArgs << TheCall->getNumArgs()/*function call*/; 1295 if (TheCall->getNumArgs() > NumArgs) 1296 return Diag(TheCall->getArg(NumArgs)->getLocStart(), 1297 diag::err_typecheck_call_too_many_args) 1298 << 0 /*function call*/ << NumArgs << TheCall->getNumArgs() 1299 << SourceRange(TheCall->getArg(NumArgs)->getLocStart(), 1300 (*(TheCall->arg_end()-1))->getLocEnd()); 1301 1302 Expr *OrigArg = TheCall->getArg(NumArgs-1); 1303 1304 if (OrigArg->isTypeDependent()) 1305 return false; 1306 1307 // This operation requires a non-_Complex floating-point number. 1308 if (!OrigArg->getType()->isRealFloatingType()) 1309 return Diag(OrigArg->getLocStart(), 1310 diag::err_typecheck_call_invalid_unary_fp) 1311 << OrigArg->getType() << OrigArg->getSourceRange(); 1312 1313 // If this is an implicit conversion from float -> double, remove it. 1314 if (ImplicitCastExpr *Cast = dyn_cast<ImplicitCastExpr>(OrigArg)) { 1315 Expr *CastArg = Cast->getSubExpr(); 1316 if (CastArg->getType()->isSpecificBuiltinType(BuiltinType::Float)) { 1317 assert(Cast->getType()->isSpecificBuiltinType(BuiltinType::Double) && 1318 "promotion from float to double is the only expected cast here"); 1319 Cast->setSubExpr(0); 1320 TheCall->setArg(NumArgs-1, CastArg); 1321 } 1322 } 1323 1324 return false; 1325} 1326 1327/// SemaBuiltinShuffleVector - Handle __builtin_shufflevector. 1328// This is declared to take (...), so we have to check everything. 1329ExprResult Sema::SemaBuiltinShuffleVector(CallExpr *TheCall) { 1330 if (TheCall->getNumArgs() < 2) 1331 return ExprError(Diag(TheCall->getLocEnd(), 1332 diag::err_typecheck_call_too_few_args_at_least) 1333 << 0 /*function call*/ << 2 << TheCall->getNumArgs() 1334 << TheCall->getSourceRange()); 1335 1336 // Determine which of the following types of shufflevector we're checking: 1337 // 1) unary, vector mask: (lhs, mask) 1338 // 2) binary, vector mask: (lhs, rhs, mask) 1339 // 3) binary, scalar mask: (lhs, rhs, index, ..., index) 1340 QualType resType = TheCall->getArg(0)->getType(); 1341 unsigned numElements = 0; 1342 1343 if (!TheCall->getArg(0)->isTypeDependent() && 1344 !TheCall->getArg(1)->isTypeDependent()) { 1345 QualType LHSType = TheCall->getArg(0)->getType(); 1346 QualType RHSType = TheCall->getArg(1)->getType(); 1347 1348 if (!LHSType->isVectorType() || !RHSType->isVectorType()) { 1349 Diag(TheCall->getLocStart(), diag::err_shufflevector_non_vector) 1350 << SourceRange(TheCall->getArg(0)->getLocStart(), 1351 TheCall->getArg(1)->getLocEnd()); 1352 return ExprError(); 1353 } 1354 1355 numElements = LHSType->getAs<VectorType>()->getNumElements(); 1356 unsigned numResElements = TheCall->getNumArgs() - 2; 1357 1358 // Check to see if we have a call with 2 vector arguments, the unary shuffle 1359 // with mask. If so, verify that RHS is an integer vector type with the 1360 // same number of elts as lhs. 1361 if (TheCall->getNumArgs() == 2) { 1362 if (!RHSType->hasIntegerRepresentation() || 1363 RHSType->getAs<VectorType>()->getNumElements() != numElements) 1364 Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector) 1365 << SourceRange(TheCall->getArg(1)->getLocStart(), 1366 TheCall->getArg(1)->getLocEnd()); 1367 numResElements = numElements; 1368 } 1369 else if (!Context.hasSameUnqualifiedType(LHSType, RHSType)) { 1370 Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector) 1371 << SourceRange(TheCall->getArg(0)->getLocStart(), 1372 TheCall->getArg(1)->getLocEnd()); 1373 return ExprError(); 1374 } else if (numElements != numResElements) { 1375 QualType eltType = LHSType->getAs<VectorType>()->getElementType(); 1376 resType = Context.getVectorType(eltType, numResElements, 1377 VectorType::GenericVector); 1378 } 1379 } 1380 1381 for (unsigned i = 2; i < TheCall->getNumArgs(); i++) { 1382 if (TheCall->getArg(i)->isTypeDependent() || 1383 TheCall->getArg(i)->isValueDependent()) 1384 continue; 1385 1386 llvm::APSInt Result(32); 1387 if (!TheCall->getArg(i)->isIntegerConstantExpr(Result, Context)) 1388 return ExprError(Diag(TheCall->getLocStart(), 1389 diag::err_shufflevector_nonconstant_argument) 1390 << TheCall->getArg(i)->getSourceRange()); 1391 1392 if (Result.getActiveBits() > 64 || Result.getZExtValue() >= numElements*2) 1393 return ExprError(Diag(TheCall->getLocStart(), 1394 diag::err_shufflevector_argument_too_large) 1395 << TheCall->getArg(i)->getSourceRange()); 1396 } 1397 1398 SmallVector<Expr*, 32> exprs; 1399 1400 for (unsigned i = 0, e = TheCall->getNumArgs(); i != e; i++) { 1401 exprs.push_back(TheCall->getArg(i)); 1402 TheCall->setArg(i, 0); 1403 } 1404 1405 return Owned(new (Context) ShuffleVectorExpr(Context, exprs.begin(), 1406 exprs.size(), resType, 1407 TheCall->getCallee()->getLocStart(), 1408 TheCall->getRParenLoc())); 1409} 1410 1411/// SemaBuiltinPrefetch - Handle __builtin_prefetch. 1412// This is declared to take (const void*, ...) and can take two 1413// optional constant int args. 1414bool Sema::SemaBuiltinPrefetch(CallExpr *TheCall) { 1415 unsigned NumArgs = TheCall->getNumArgs(); 1416 1417 if (NumArgs > 3) 1418 return Diag(TheCall->getLocEnd(), 1419 diag::err_typecheck_call_too_many_args_at_most) 1420 << 0 /*function call*/ << 3 << NumArgs 1421 << TheCall->getSourceRange(); 1422 1423 // Argument 0 is checked for us and the remaining arguments must be 1424 // constant integers. 1425 for (unsigned i = 1; i != NumArgs; ++i) { 1426 Expr *Arg = TheCall->getArg(i); 1427 1428 llvm::APSInt Result; 1429 if (SemaBuiltinConstantArg(TheCall, i, Result)) 1430 return true; 1431 1432 // FIXME: gcc issues a warning and rewrites these to 0. These 1433 // seems especially odd for the third argument since the default 1434 // is 3. 1435 if (i == 1) { 1436 if (Result.getLimitedValue() > 1) 1437 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 1438 << "0" << "1" << Arg->getSourceRange(); 1439 } else { 1440 if (Result.getLimitedValue() > 3) 1441 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 1442 << "0" << "3" << Arg->getSourceRange(); 1443 } 1444 } 1445 1446 return false; 1447} 1448 1449/// SemaBuiltinConstantArg - Handle a check if argument ArgNum of CallExpr 1450/// TheCall is a constant expression. 1451bool Sema::SemaBuiltinConstantArg(CallExpr *TheCall, int ArgNum, 1452 llvm::APSInt &Result) { 1453 Expr *Arg = TheCall->getArg(ArgNum); 1454 DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts()); 1455 FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl()); 1456 1457 if (Arg->isTypeDependent() || Arg->isValueDependent()) return false; 1458 1459 if (!Arg->isIntegerConstantExpr(Result, Context)) 1460 return Diag(TheCall->getLocStart(), diag::err_constant_integer_arg_type) 1461 << FDecl->getDeclName() << Arg->getSourceRange(); 1462 1463 return false; 1464} 1465 1466/// SemaBuiltinObjectSize - Handle __builtin_object_size(void *ptr, 1467/// int type). This simply type checks that type is one of the defined 1468/// constants (0-3). 1469// For compatibility check 0-3, llvm only handles 0 and 2. 1470bool Sema::SemaBuiltinObjectSize(CallExpr *TheCall) { 1471 llvm::APSInt Result; 1472 1473 // Check constant-ness first. 1474 if (SemaBuiltinConstantArg(TheCall, 1, Result)) 1475 return true; 1476 1477 Expr *Arg = TheCall->getArg(1); 1478 if (Result.getSExtValue() < 0 || Result.getSExtValue() > 3) { 1479 return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range) 1480 << "0" << "3" << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 1481 } 1482 1483 return false; 1484} 1485 1486/// SemaBuiltinLongjmp - Handle __builtin_longjmp(void *env[5], int val). 1487/// This checks that val is a constant 1. 1488bool Sema::SemaBuiltinLongjmp(CallExpr *TheCall) { 1489 Expr *Arg = TheCall->getArg(1); 1490 llvm::APSInt Result; 1491 1492 // TODO: This is less than ideal. Overload this to take a value. 1493 if (SemaBuiltinConstantArg(TheCall, 1, Result)) 1494 return true; 1495 1496 if (Result != 1) 1497 return Diag(TheCall->getLocStart(), diag::err_builtin_longjmp_invalid_val) 1498 << SourceRange(Arg->getLocStart(), Arg->getLocEnd()); 1499 1500 return false; 1501} 1502 1503// Handle i > 1 ? "x" : "y", recursively. 1504bool Sema::SemaCheckStringLiteral(const Expr *E, Expr **Args, 1505 unsigned NumArgs, bool HasVAListArg, 1506 unsigned format_idx, unsigned firstDataArg, 1507 FormatStringType Type, bool inFunctionCall) { 1508 tryAgain: 1509 if (E->isTypeDependent() || E->isValueDependent()) 1510 return false; 1511 1512 E = E->IgnoreParenCasts(); 1513 1514 if (E->isNullPointerConstant(Context, Expr::NPC_ValueDependentIsNotNull)) 1515 // Technically -Wformat-nonliteral does not warn about this case. 1516 // The behavior of printf and friends in this case is implementation 1517 // dependent. Ideally if the format string cannot be null then 1518 // it should have a 'nonnull' attribute in the function prototype. 1519 return true; 1520 1521 switch (E->getStmtClass()) { 1522 case Stmt::BinaryConditionalOperatorClass: 1523 case Stmt::ConditionalOperatorClass: { 1524 const AbstractConditionalOperator *C = cast<AbstractConditionalOperator>(E); 1525 return SemaCheckStringLiteral(C->getTrueExpr(), Args, NumArgs, HasVAListArg, 1526 format_idx, firstDataArg, Type, 1527 inFunctionCall) 1528 && SemaCheckStringLiteral(C->getFalseExpr(), Args, NumArgs, HasVAListArg, 1529 format_idx, firstDataArg, Type, 1530 inFunctionCall); 1531 } 1532 1533 case Stmt::ImplicitCastExprClass: { 1534 E = cast<ImplicitCastExpr>(E)->getSubExpr(); 1535 goto tryAgain; 1536 } 1537 1538 case Stmt::OpaqueValueExprClass: 1539 if (const Expr *src = cast<OpaqueValueExpr>(E)->getSourceExpr()) { 1540 E = src; 1541 goto tryAgain; 1542 } 1543 return false; 1544 1545 case Stmt::PredefinedExprClass: 1546 // While __func__, etc., are technically not string literals, they 1547 // cannot contain format specifiers and thus are not a security 1548 // liability. 1549 return true; 1550 1551 case Stmt::DeclRefExprClass: { 1552 const DeclRefExpr *DR = cast<DeclRefExpr>(E); 1553 1554 // As an exception, do not flag errors for variables binding to 1555 // const string literals. 1556 if (const VarDecl *VD = dyn_cast<VarDecl>(DR->getDecl())) { 1557 bool isConstant = false; 1558 QualType T = DR->getType(); 1559 1560 if (const ArrayType *AT = Context.getAsArrayType(T)) { 1561 isConstant = AT->getElementType().isConstant(Context); 1562 } else if (const PointerType *PT = T->getAs<PointerType>()) { 1563 isConstant = T.isConstant(Context) && 1564 PT->getPointeeType().isConstant(Context); 1565 } else if (T->isObjCObjectPointerType()) { 1566 // In ObjC, there is usually no "const ObjectPointer" type, 1567 // so don't check if the pointee type is constant. 1568 isConstant = T.isConstant(Context); 1569 } 1570 1571 if (isConstant) { 1572 if (const Expr *Init = VD->getAnyInitializer()) 1573 return SemaCheckStringLiteral(Init, Args, NumArgs, 1574 HasVAListArg, format_idx, firstDataArg, 1575 Type, /*inFunctionCall*/false); 1576 } 1577 1578 // For vprintf* functions (i.e., HasVAListArg==true), we add a 1579 // special check to see if the format string is a function parameter 1580 // of the function calling the printf function. If the function 1581 // has an attribute indicating it is a printf-like function, then we 1582 // should suppress warnings concerning non-literals being used in a call 1583 // to a vprintf function. For example: 1584 // 1585 // void 1586 // logmessage(char const *fmt __attribute__ (format (printf, 1, 2)), ...){ 1587 // va_list ap; 1588 // va_start(ap, fmt); 1589 // vprintf(fmt, ap); // Do NOT emit a warning about "fmt". 1590 // ... 1591 // 1592 if (HasVAListArg) { 1593 if (const ParmVarDecl *PV = dyn_cast<ParmVarDecl>(VD)) { 1594 if (const NamedDecl *ND = dyn_cast<NamedDecl>(PV->getDeclContext())) { 1595 int PVIndex = PV->getFunctionScopeIndex() + 1; 1596 for (specific_attr_iterator<FormatAttr> 1597 i = ND->specific_attr_begin<FormatAttr>(), 1598 e = ND->specific_attr_end<FormatAttr>(); i != e ; ++i) { 1599 FormatAttr *PVFormat = *i; 1600 // adjust for implicit parameter 1601 if (const CXXMethodDecl *MD = dyn_cast<CXXMethodDecl>(ND)) 1602 if (MD->isInstance()) 1603 ++PVIndex; 1604 // We also check if the formats are compatible. 1605 // We can't pass a 'scanf' string to a 'printf' function. 1606 if (PVIndex == PVFormat->getFormatIdx() && 1607 Type == GetFormatStringType(PVFormat)) 1608 return true; 1609 } 1610 } 1611 } 1612 } 1613 } 1614 1615 return false; 1616 } 1617 1618 case Stmt::CallExprClass: 1619 case Stmt::CXXMemberCallExprClass: { 1620 const CallExpr *CE = cast<CallExpr>(E); 1621 if (const NamedDecl *ND = dyn_cast_or_null<NamedDecl>(CE->getCalleeDecl())) { 1622 if (const FormatArgAttr *FA = ND->getAttr<FormatArgAttr>()) { 1623 unsigned ArgIndex = FA->getFormatIdx(); 1624 if (const CXXMethodDecl *MD = dyn_cast<CXXMethodDecl>(ND)) 1625 if (MD->isInstance()) 1626 --ArgIndex; 1627 const Expr *Arg = CE->getArg(ArgIndex - 1); 1628 1629 return SemaCheckStringLiteral(Arg, Args, NumArgs, HasVAListArg, 1630 format_idx, firstDataArg, Type, 1631 inFunctionCall); 1632 } 1633 } 1634 1635 return false; 1636 } 1637 case Stmt::ObjCStringLiteralClass: 1638 case Stmt::StringLiteralClass: { 1639 const StringLiteral *StrE = NULL; 1640 1641 if (const ObjCStringLiteral *ObjCFExpr = dyn_cast<ObjCStringLiteral>(E)) 1642 StrE = ObjCFExpr->getString(); 1643 else 1644 StrE = cast<StringLiteral>(E); 1645 1646 if (StrE) { 1647 CheckFormatString(StrE, E, Args, NumArgs, HasVAListArg, format_idx, 1648 firstDataArg, Type, inFunctionCall); 1649 return true; 1650 } 1651 1652 return false; 1653 } 1654 1655 default: 1656 return false; 1657 } 1658} 1659 1660void 1661Sema::CheckNonNullArguments(const NonNullAttr *NonNull, 1662 const Expr * const *ExprArgs, 1663 SourceLocation CallSiteLoc) { 1664 for (NonNullAttr::args_iterator i = NonNull->args_begin(), 1665 e = NonNull->args_end(); 1666 i != e; ++i) { 1667 const Expr *ArgExpr = ExprArgs[*i]; 1668 if (ArgExpr->isNullPointerConstant(Context, 1669 Expr::NPC_ValueDependentIsNotNull)) 1670 Diag(CallSiteLoc, diag::warn_null_arg) << ArgExpr->getSourceRange(); 1671 } 1672} 1673 1674Sema::FormatStringType Sema::GetFormatStringType(const FormatAttr *Format) { 1675 return llvm::StringSwitch<FormatStringType>(Format->getType()) 1676 .Case("scanf", FST_Scanf) 1677 .Cases("printf", "printf0", FST_Printf) 1678 .Cases("NSString", "CFString", FST_NSString) 1679 .Case("strftime", FST_Strftime) 1680 .Case("strfmon", FST_Strfmon) 1681 .Cases("kprintf", "cmn_err", "vcmn_err", "zcmn_err", FST_Kprintf) 1682 .Default(FST_Unknown); 1683} 1684 1685/// CheckPrintfScanfArguments - Check calls to printf and scanf (and similar 1686/// functions) for correct use of format strings. 1687void Sema::CheckFormatArguments(const FormatAttr *Format, CallExpr *TheCall) { 1688 bool IsCXXMember = false; 1689 // The way the format attribute works in GCC, the implicit this argument 1690 // of member functions is counted. However, it doesn't appear in our own 1691 // lists, so decrement format_idx in that case. 1692 IsCXXMember = isa<CXXMemberCallExpr>(TheCall); 1693 CheckFormatArguments(Format, TheCall->getArgs(), TheCall->getNumArgs(), 1694 IsCXXMember, TheCall->getRParenLoc(), 1695 TheCall->getCallee()->getSourceRange()); 1696} 1697 1698void Sema::CheckFormatArguments(const FormatAttr *Format, Expr **Args, 1699 unsigned NumArgs, bool IsCXXMember, 1700 SourceLocation Loc, SourceRange Range) { 1701 bool HasVAListArg = Format->getFirstArg() == 0; 1702 unsigned format_idx = Format->getFormatIdx() - 1; 1703 unsigned firstDataArg = HasVAListArg ? 0 : Format->getFirstArg() - 1; 1704 if (IsCXXMember) { 1705 if (format_idx == 0) 1706 return; 1707 --format_idx; 1708 if(firstDataArg != 0) 1709 --firstDataArg; 1710 } 1711 CheckFormatArguments(Args, NumArgs, HasVAListArg, format_idx, 1712 firstDataArg, GetFormatStringType(Format), Loc, Range); 1713} 1714 1715void Sema::CheckFormatArguments(Expr **Args, unsigned NumArgs, 1716 bool HasVAListArg, unsigned format_idx, 1717 unsigned firstDataArg, FormatStringType Type, 1718 SourceLocation Loc, SourceRange Range) { 1719 // CHECK: printf/scanf-like function is called with no format string. 1720 if (format_idx >= NumArgs) { 1721 Diag(Loc, diag::warn_missing_format_string) << Range; 1722 return; 1723 } 1724 1725 const Expr *OrigFormatExpr = Args[format_idx]->IgnoreParenCasts(); 1726 1727 // CHECK: format string is not a string literal. 1728 // 1729 // Dynamically generated format strings are difficult to 1730 // automatically vet at compile time. Requiring that format strings 1731 // are string literals: (1) permits the checking of format strings by 1732 // the compiler and thereby (2) can practically remove the source of 1733 // many format string exploits. 1734 1735 // Format string can be either ObjC string (e.g. @"%d") or 1736 // C string (e.g. "%d") 1737 // ObjC string uses the same format specifiers as C string, so we can use 1738 // the same format string checking logic for both ObjC and C strings. 1739 if (SemaCheckStringLiteral(OrigFormatExpr, Args, NumArgs, HasVAListArg, 1740 format_idx, firstDataArg, Type)) 1741 return; // Literal format string found, check done! 1742 1743 // Strftime is particular as it always uses a single 'time' argument, 1744 // so it is safe to pass a non-literal string. 1745 if (Type == FST_Strftime) 1746 return; 1747 1748 // Do not emit diag when the string param is a macro expansion and the 1749 // format is either NSString or CFString. This is a hack to prevent 1750 // diag when using the NSLocalizedString and CFCopyLocalizedString macros 1751 // which are usually used in place of NS and CF string literals. 1752 if (Type == FST_NSString && 1753 SourceMgr.isInSystemMacro(Args[format_idx]->getLocStart())) 1754 return; 1755 1756 // If there are no arguments specified, warn with -Wformat-security, otherwise 1757 // warn only with -Wformat-nonliteral. 1758 if (NumArgs == format_idx+1) 1759 Diag(Args[format_idx]->getLocStart(), 1760 diag::warn_format_nonliteral_noargs) 1761 << OrigFormatExpr->getSourceRange(); 1762 else 1763 Diag(Args[format_idx]->getLocStart(), 1764 diag::warn_format_nonliteral) 1765 << OrigFormatExpr->getSourceRange(); 1766} 1767 1768namespace { 1769class CheckFormatHandler : public analyze_format_string::FormatStringHandler { 1770protected: 1771 Sema &S; 1772 const StringLiteral *FExpr; 1773 const Expr *OrigFormatExpr; 1774 const unsigned FirstDataArg; 1775 const unsigned NumDataArgs; 1776 const bool IsObjCLiteral; 1777 const char *Beg; // Start of format string. 1778 const bool HasVAListArg; 1779 const Expr * const *Args; 1780 const unsigned NumArgs; 1781 unsigned FormatIdx; 1782 llvm::BitVector CoveredArgs; 1783 bool usesPositionalArgs; 1784 bool atFirstArg; 1785 bool inFunctionCall; 1786public: 1787 CheckFormatHandler(Sema &s, const StringLiteral *fexpr, 1788 const Expr *origFormatExpr, unsigned firstDataArg, 1789 unsigned numDataArgs, bool isObjCLiteral, 1790 const char *beg, bool hasVAListArg, 1791 Expr **args, unsigned numArgs, 1792 unsigned formatIdx, bool inFunctionCall) 1793 : S(s), FExpr(fexpr), OrigFormatExpr(origFormatExpr), 1794 FirstDataArg(firstDataArg), 1795 NumDataArgs(numDataArgs), 1796 IsObjCLiteral(isObjCLiteral), Beg(beg), 1797 HasVAListArg(hasVAListArg), 1798 Args(args), NumArgs(numArgs), FormatIdx(formatIdx), 1799 usesPositionalArgs(false), atFirstArg(true), 1800 inFunctionCall(inFunctionCall) { 1801 CoveredArgs.resize(numDataArgs); 1802 CoveredArgs.reset(); 1803 } 1804 1805 void DoneProcessing(); 1806 1807 void HandleIncompleteSpecifier(const char *startSpecifier, 1808 unsigned specifierLen); 1809 1810 void HandleNonStandardLengthModifier( 1811 const analyze_format_string::LengthModifier &LM, 1812 const char *startSpecifier, unsigned specifierLen); 1813 1814 void HandleNonStandardConversionSpecifier( 1815 const analyze_format_string::ConversionSpecifier &CS, 1816 const char *startSpecifier, unsigned specifierLen); 1817 1818 void HandleNonStandardConversionSpecification( 1819 const analyze_format_string::LengthModifier &LM, 1820 const analyze_format_string::ConversionSpecifier &CS, 1821 const char *startSpecifier, unsigned specifierLen); 1822 1823 virtual void HandlePosition(const char *startPos, unsigned posLen); 1824 1825 virtual void HandleInvalidPosition(const char *startSpecifier, 1826 unsigned specifierLen, 1827 analyze_format_string::PositionContext p); 1828 1829 virtual void HandleZeroPosition(const char *startPos, unsigned posLen); 1830 1831 void HandleNullChar(const char *nullCharacter); 1832 1833 template <typename Range> 1834 static void EmitFormatDiagnostic(Sema &S, bool inFunctionCall, 1835 const Expr *ArgumentExpr, 1836 PartialDiagnostic PDiag, 1837 SourceLocation StringLoc, 1838 bool IsStringLocation, Range StringRange, 1839 FixItHint Fixit = FixItHint()); 1840 1841protected: 1842 bool HandleInvalidConversionSpecifier(unsigned argIndex, SourceLocation Loc, 1843 const char *startSpec, 1844 unsigned specifierLen, 1845 const char *csStart, unsigned csLen); 1846 1847 void HandlePositionalNonpositionalArgs(SourceLocation Loc, 1848 const char *startSpec, 1849 unsigned specifierLen); 1850 1851 SourceRange getFormatStringRange(); 1852 CharSourceRange getSpecifierRange(const char *startSpecifier, 1853 unsigned specifierLen); 1854 SourceLocation getLocationOfByte(const char *x); 1855 1856 const Expr *getDataArg(unsigned i) const; 1857 1858 bool CheckNumArgs(const analyze_format_string::FormatSpecifier &FS, 1859 const analyze_format_string::ConversionSpecifier &CS, 1860 const char *startSpecifier, unsigned specifierLen, 1861 unsigned argIndex); 1862 1863 template <typename Range> 1864 void EmitFormatDiagnostic(PartialDiagnostic PDiag, SourceLocation StringLoc, 1865 bool IsStringLocation, Range StringRange, 1866 FixItHint Fixit = FixItHint()); 1867 1868 void CheckPositionalAndNonpositionalArgs( 1869 const analyze_format_string::FormatSpecifier *FS); 1870}; 1871} 1872 1873SourceRange CheckFormatHandler::getFormatStringRange() { 1874 return OrigFormatExpr->getSourceRange(); 1875} 1876 1877CharSourceRange CheckFormatHandler:: 1878getSpecifierRange(const char *startSpecifier, unsigned specifierLen) { 1879 SourceLocation Start = getLocationOfByte(startSpecifier); 1880 SourceLocation End = getLocationOfByte(startSpecifier + specifierLen - 1); 1881 1882 // Advance the end SourceLocation by one due to half-open ranges. 1883 End = End.getLocWithOffset(1); 1884 1885 return CharSourceRange::getCharRange(Start, End); 1886} 1887 1888SourceLocation CheckFormatHandler::getLocationOfByte(const char *x) { 1889 return S.getLocationOfStringLiteralByte(FExpr, x - Beg); 1890} 1891 1892void CheckFormatHandler::HandleIncompleteSpecifier(const char *startSpecifier, 1893 unsigned specifierLen){ 1894 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_incomplete_specifier), 1895 getLocationOfByte(startSpecifier), 1896 /*IsStringLocation*/true, 1897 getSpecifierRange(startSpecifier, specifierLen)); 1898} 1899 1900void CheckFormatHandler::HandleNonStandardLengthModifier( 1901 const analyze_format_string::LengthModifier &LM, 1902 const char *startSpecifier, unsigned specifierLen) { 1903 EmitFormatDiagnostic(S.PDiag(diag::warn_format_non_standard) << LM.toString() 1904 << 0, 1905 getLocationOfByte(LM.getStart()), 1906 /*IsStringLocation*/true, 1907 getSpecifierRange(startSpecifier, specifierLen)); 1908} 1909 1910void CheckFormatHandler::HandleNonStandardConversionSpecifier( 1911 const analyze_format_string::ConversionSpecifier &CS, 1912 const char *startSpecifier, unsigned specifierLen) { 1913 EmitFormatDiagnostic(S.PDiag(diag::warn_format_non_standard) << CS.toString() 1914 << 1, 1915 getLocationOfByte(CS.getStart()), 1916 /*IsStringLocation*/true, 1917 getSpecifierRange(startSpecifier, specifierLen)); 1918} 1919 1920void CheckFormatHandler::HandleNonStandardConversionSpecification( 1921 const analyze_format_string::LengthModifier &LM, 1922 const analyze_format_string::ConversionSpecifier &CS, 1923 const char *startSpecifier, unsigned specifierLen) { 1924 EmitFormatDiagnostic(S.PDiag(diag::warn_format_non_standard_conversion_spec) 1925 << LM.toString() << CS.toString(), 1926 getLocationOfByte(LM.getStart()), 1927 /*IsStringLocation*/true, 1928 getSpecifierRange(startSpecifier, specifierLen)); 1929} 1930 1931void CheckFormatHandler::HandlePosition(const char *startPos, 1932 unsigned posLen) { 1933 EmitFormatDiagnostic(S.PDiag(diag::warn_format_non_standard_positional_arg), 1934 getLocationOfByte(startPos), 1935 /*IsStringLocation*/true, 1936 getSpecifierRange(startPos, posLen)); 1937} 1938 1939void 1940CheckFormatHandler::HandleInvalidPosition(const char *startPos, unsigned posLen, 1941 analyze_format_string::PositionContext p) { 1942 EmitFormatDiagnostic(S.PDiag(diag::warn_format_invalid_positional_specifier) 1943 << (unsigned) p, 1944 getLocationOfByte(startPos), /*IsStringLocation*/true, 1945 getSpecifierRange(startPos, posLen)); 1946} 1947 1948void CheckFormatHandler::HandleZeroPosition(const char *startPos, 1949 unsigned posLen) { 1950 EmitFormatDiagnostic(S.PDiag(diag::warn_format_zero_positional_specifier), 1951 getLocationOfByte(startPos), 1952 /*IsStringLocation*/true, 1953 getSpecifierRange(startPos, posLen)); 1954} 1955 1956void CheckFormatHandler::HandleNullChar(const char *nullCharacter) { 1957 if (!IsObjCLiteral) { 1958 // The presence of a null character is likely an error. 1959 EmitFormatDiagnostic( 1960 S.PDiag(diag::warn_printf_format_string_contains_null_char), 1961 getLocationOfByte(nullCharacter), /*IsStringLocation*/true, 1962 getFormatStringRange()); 1963 } 1964} 1965 1966const Expr *CheckFormatHandler::getDataArg(unsigned i) const { 1967 return Args[FirstDataArg + i]; 1968} 1969 1970void CheckFormatHandler::DoneProcessing() { 1971 // Does the number of data arguments exceed the number of 1972 // format conversions in the format string? 1973 if (!HasVAListArg) { 1974 // Find any arguments that weren't covered. 1975 CoveredArgs.flip(); 1976 signed notCoveredArg = CoveredArgs.find_first(); 1977 if (notCoveredArg >= 0) { 1978 assert((unsigned)notCoveredArg < NumDataArgs); 1979 SourceLocation Loc = getDataArg((unsigned) notCoveredArg)->getLocStart(); 1980 if (!S.getSourceManager().isInSystemMacro(Loc)) { 1981 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_data_arg_not_used), 1982 Loc, 1983 /*IsStringLocation*/false, getFormatStringRange()); 1984 } 1985 } 1986 } 1987} 1988 1989bool 1990CheckFormatHandler::HandleInvalidConversionSpecifier(unsigned argIndex, 1991 SourceLocation Loc, 1992 const char *startSpec, 1993 unsigned specifierLen, 1994 const char *csStart, 1995 unsigned csLen) { 1996 1997 bool keepGoing = true; 1998 if (argIndex < NumDataArgs) { 1999 // Consider the argument coverered, even though the specifier doesn't 2000 // make sense. 2001 CoveredArgs.set(argIndex); 2002 } 2003 else { 2004 // If argIndex exceeds the number of data arguments we 2005 // don't issue a warning because that is just a cascade of warnings (and 2006 // they may have intended '%%' anyway). We don't want to continue processing 2007 // the format string after this point, however, as we will like just get 2008 // gibberish when trying to match arguments. 2009 keepGoing = false; 2010 } 2011 2012 EmitFormatDiagnostic(S.PDiag(diag::warn_format_invalid_conversion) 2013 << StringRef(csStart, csLen), 2014 Loc, /*IsStringLocation*/true, 2015 getSpecifierRange(startSpec, specifierLen)); 2016 2017 return keepGoing; 2018} 2019 2020void 2021CheckFormatHandler::HandlePositionalNonpositionalArgs(SourceLocation Loc, 2022 const char *startSpec, 2023 unsigned specifierLen) { 2024 EmitFormatDiagnostic( 2025 S.PDiag(diag::warn_format_mix_positional_nonpositional_args), 2026 Loc, /*isStringLoc*/true, getSpecifierRange(startSpec, specifierLen)); 2027} 2028 2029bool 2030CheckFormatHandler::CheckNumArgs( 2031 const analyze_format_string::FormatSpecifier &FS, 2032 const analyze_format_string::ConversionSpecifier &CS, 2033 const char *startSpecifier, unsigned specifierLen, unsigned argIndex) { 2034 2035 if (argIndex >= NumDataArgs) { 2036 PartialDiagnostic PDiag = FS.usesPositionalArg() 2037 ? (S.PDiag(diag::warn_printf_positional_arg_exceeds_data_args) 2038 << (argIndex+1) << NumDataArgs) 2039 : S.PDiag(diag::warn_printf_insufficient_data_args); 2040 EmitFormatDiagnostic( 2041 PDiag, getLocationOfByte(CS.getStart()), /*IsStringLocation*/true, 2042 getSpecifierRange(startSpecifier, specifierLen)); 2043 return false; 2044 } 2045 return true; 2046} 2047 2048template<typename Range> 2049void CheckFormatHandler::EmitFormatDiagnostic(PartialDiagnostic PDiag, 2050 SourceLocation Loc, 2051 bool IsStringLocation, 2052 Range StringRange, 2053 FixItHint FixIt) { 2054 EmitFormatDiagnostic(S, inFunctionCall, Args[FormatIdx], PDiag, 2055 Loc, IsStringLocation, StringRange, FixIt); 2056} 2057 2058/// \brief If the format string is not within the funcion call, emit a note 2059/// so that the function call and string are in diagnostic messages. 2060/// 2061/// \param inFunctionCall if true, the format string is within the function 2062/// call and only one diagnostic message will be produced. Otherwise, an 2063/// extra note will be emitted pointing to location of the format string. 2064/// 2065/// \param ArgumentExpr the expression that is passed as the format string 2066/// argument in the function call. Used for getting locations when two 2067/// diagnostics are emitted. 2068/// 2069/// \param PDiag the callee should already have provided any strings for the 2070/// diagnostic message. This function only adds locations and fixits 2071/// to diagnostics. 2072/// 2073/// \param Loc primary location for diagnostic. If two diagnostics are 2074/// required, one will be at Loc and a new SourceLocation will be created for 2075/// the other one. 2076/// 2077/// \param IsStringLocation if true, Loc points to the format string should be 2078/// used for the note. Otherwise, Loc points to the argument list and will 2079/// be used with PDiag. 2080/// 2081/// \param StringRange some or all of the string to highlight. This is 2082/// templated so it can accept either a CharSourceRange or a SourceRange. 2083/// 2084/// \param Fixit optional fix it hint for the format string. 2085template<typename Range> 2086void CheckFormatHandler::EmitFormatDiagnostic(Sema &S, bool InFunctionCall, 2087 const Expr *ArgumentExpr, 2088 PartialDiagnostic PDiag, 2089 SourceLocation Loc, 2090 bool IsStringLocation, 2091 Range StringRange, 2092 FixItHint FixIt) { 2093 if (InFunctionCall) 2094 S.Diag(Loc, PDiag) << StringRange << FixIt; 2095 else { 2096 S.Diag(IsStringLocation ? ArgumentExpr->getExprLoc() : Loc, PDiag) 2097 << ArgumentExpr->getSourceRange(); 2098 S.Diag(IsStringLocation ? Loc : StringRange.getBegin(), 2099 diag::note_format_string_defined) 2100 << StringRange << FixIt; 2101 } 2102} 2103 2104//===--- CHECK: Printf format string checking ------------------------------===// 2105 2106namespace { 2107class CheckPrintfHandler : public CheckFormatHandler { 2108public: 2109 CheckPrintfHandler(Sema &s, const StringLiteral *fexpr, 2110 const Expr *origFormatExpr, unsigned firstDataArg, 2111 unsigned numDataArgs, bool isObjCLiteral, 2112 const char *beg, bool hasVAListArg, 2113 Expr **Args, unsigned NumArgs, 2114 unsigned formatIdx, bool inFunctionCall) 2115 : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg, 2116 numDataArgs, isObjCLiteral, beg, hasVAListArg, 2117 Args, NumArgs, formatIdx, inFunctionCall) {} 2118 2119 2120 bool HandleInvalidPrintfConversionSpecifier( 2121 const analyze_printf::PrintfSpecifier &FS, 2122 const char *startSpecifier, 2123 unsigned specifierLen); 2124 2125 bool HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier &FS, 2126 const char *startSpecifier, 2127 unsigned specifierLen); 2128 2129 bool HandleAmount(const analyze_format_string::OptionalAmount &Amt, unsigned k, 2130 const char *startSpecifier, unsigned specifierLen); 2131 void HandleInvalidAmount(const analyze_printf::PrintfSpecifier &FS, 2132 const analyze_printf::OptionalAmount &Amt, 2133 unsigned type, 2134 const char *startSpecifier, unsigned specifierLen); 2135 void HandleFlag(const analyze_printf::PrintfSpecifier &FS, 2136 const analyze_printf::OptionalFlag &flag, 2137 const char *startSpecifier, unsigned specifierLen); 2138 void HandleIgnoredFlag(const analyze_printf::PrintfSpecifier &FS, 2139 const analyze_printf::OptionalFlag &ignoredFlag, 2140 const analyze_printf::OptionalFlag &flag, 2141 const char *startSpecifier, unsigned specifierLen); 2142}; 2143} 2144 2145bool CheckPrintfHandler::HandleInvalidPrintfConversionSpecifier( 2146 const analyze_printf::PrintfSpecifier &FS, 2147 const char *startSpecifier, 2148 unsigned specifierLen) { 2149 const analyze_printf::PrintfConversionSpecifier &CS = 2150 FS.getConversionSpecifier(); 2151 2152 return HandleInvalidConversionSpecifier(FS.getArgIndex(), 2153 getLocationOfByte(CS.getStart()), 2154 startSpecifier, specifierLen, 2155 CS.getStart(), CS.getLength()); 2156} 2157 2158bool CheckPrintfHandler::HandleAmount( 2159 const analyze_format_string::OptionalAmount &Amt, 2160 unsigned k, const char *startSpecifier, 2161 unsigned specifierLen) { 2162 2163 if (Amt.hasDataArgument()) { 2164 if (!HasVAListArg) { 2165 unsigned argIndex = Amt.getArgIndex(); 2166 if (argIndex >= NumDataArgs) { 2167 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_asterisk_missing_arg) 2168 << k, 2169 getLocationOfByte(Amt.getStart()), 2170 /*IsStringLocation*/true, 2171 getSpecifierRange(startSpecifier, specifierLen)); 2172 // Don't do any more checking. We will just emit 2173 // spurious errors. 2174 return false; 2175 } 2176 2177 // Type check the data argument. It should be an 'int'. 2178 // Although not in conformance with C99, we also allow the argument to be 2179 // an 'unsigned int' as that is a reasonably safe case. GCC also 2180 // doesn't emit a warning for that case. 2181 CoveredArgs.set(argIndex); 2182 const Expr *Arg = getDataArg(argIndex); 2183 QualType T = Arg->getType(); 2184 2185 const analyze_printf::ArgTypeResult &ATR = Amt.getArgType(S.Context); 2186 assert(ATR.isValid()); 2187 2188 if (!ATR.matchesType(S.Context, T)) { 2189 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_asterisk_wrong_type) 2190 << k << ATR.getRepresentativeTypeName(S.Context) 2191 << T << Arg->getSourceRange(), 2192 getLocationOfByte(Amt.getStart()), 2193 /*IsStringLocation*/true, 2194 getSpecifierRange(startSpecifier, specifierLen)); 2195 // Don't do any more checking. We will just emit 2196 // spurious errors. 2197 return false; 2198 } 2199 } 2200 } 2201 return true; 2202} 2203 2204void CheckPrintfHandler::HandleInvalidAmount( 2205 const analyze_printf::PrintfSpecifier &FS, 2206 const analyze_printf::OptionalAmount &Amt, 2207 unsigned type, 2208 const char *startSpecifier, 2209 unsigned specifierLen) { 2210 const analyze_printf::PrintfConversionSpecifier &CS = 2211 FS.getConversionSpecifier(); 2212 2213 FixItHint fixit = 2214 Amt.getHowSpecified() == analyze_printf::OptionalAmount::Constant 2215 ? FixItHint::CreateRemoval(getSpecifierRange(Amt.getStart(), 2216 Amt.getConstantLength())) 2217 : FixItHint(); 2218 2219 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_nonsensical_optional_amount) 2220 << type << CS.toString(), 2221 getLocationOfByte(Amt.getStart()), 2222 /*IsStringLocation*/true, 2223 getSpecifierRange(startSpecifier, specifierLen), 2224 fixit); 2225} 2226 2227void CheckPrintfHandler::HandleFlag(const analyze_printf::PrintfSpecifier &FS, 2228 const analyze_printf::OptionalFlag &flag, 2229 const char *startSpecifier, 2230 unsigned specifierLen) { 2231 // Warn about pointless flag with a fixit removal. 2232 const analyze_printf::PrintfConversionSpecifier &CS = 2233 FS.getConversionSpecifier(); 2234 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_nonsensical_flag) 2235 << flag.toString() << CS.toString(), 2236 getLocationOfByte(flag.getPosition()), 2237 /*IsStringLocation*/true, 2238 getSpecifierRange(startSpecifier, specifierLen), 2239 FixItHint::CreateRemoval( 2240 getSpecifierRange(flag.getPosition(), 1))); 2241} 2242 2243void CheckPrintfHandler::HandleIgnoredFlag( 2244 const analyze_printf::PrintfSpecifier &FS, 2245 const analyze_printf::OptionalFlag &ignoredFlag, 2246 const analyze_printf::OptionalFlag &flag, 2247 const char *startSpecifier, 2248 unsigned specifierLen) { 2249 // Warn about ignored flag with a fixit removal. 2250 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_ignored_flag) 2251 << ignoredFlag.toString() << flag.toString(), 2252 getLocationOfByte(ignoredFlag.getPosition()), 2253 /*IsStringLocation*/true, 2254 getSpecifierRange(startSpecifier, specifierLen), 2255 FixItHint::CreateRemoval( 2256 getSpecifierRange(ignoredFlag.getPosition(), 1))); 2257} 2258 2259bool 2260CheckPrintfHandler::HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier 2261 &FS, 2262 const char *startSpecifier, 2263 unsigned specifierLen) { 2264 2265 using namespace analyze_format_string; 2266 using namespace analyze_printf; 2267 const PrintfConversionSpecifier &CS = FS.getConversionSpecifier(); 2268 2269 if (FS.consumesDataArgument()) { 2270 if (atFirstArg) { 2271 atFirstArg = false; 2272 usesPositionalArgs = FS.usesPositionalArg(); 2273 } 2274 else if (usesPositionalArgs != FS.usesPositionalArg()) { 2275 HandlePositionalNonpositionalArgs(getLocationOfByte(CS.getStart()), 2276 startSpecifier, specifierLen); 2277 return false; 2278 } 2279 } 2280 2281 // First check if the field width, precision, and conversion specifier 2282 // have matching data arguments. 2283 if (!HandleAmount(FS.getFieldWidth(), /* field width */ 0, 2284 startSpecifier, specifierLen)) { 2285 return false; 2286 } 2287 2288 if (!HandleAmount(FS.getPrecision(), /* precision */ 1, 2289 startSpecifier, specifierLen)) { 2290 return false; 2291 } 2292 2293 if (!CS.consumesDataArgument()) { 2294 // FIXME: Technically specifying a precision or field width here 2295 // makes no sense. Worth issuing a warning at some point. 2296 return true; 2297 } 2298 2299 // Consume the argument. 2300 unsigned argIndex = FS.getArgIndex(); 2301 if (argIndex < NumDataArgs) { 2302 // The check to see if the argIndex is valid will come later. 2303 // We set the bit here because we may exit early from this 2304 // function if we encounter some other error. 2305 CoveredArgs.set(argIndex); 2306 } 2307 2308 // Check for using an Objective-C specific conversion specifier 2309 // in a non-ObjC literal. 2310 if (!IsObjCLiteral && CS.isObjCArg()) { 2311 return HandleInvalidPrintfConversionSpecifier(FS, startSpecifier, 2312 specifierLen); 2313 } 2314 2315 // Check for invalid use of field width 2316 if (!FS.hasValidFieldWidth()) { 2317 HandleInvalidAmount(FS, FS.getFieldWidth(), /* field width */ 0, 2318 startSpecifier, specifierLen); 2319 } 2320 2321 // Check for invalid use of precision 2322 if (!FS.hasValidPrecision()) { 2323 HandleInvalidAmount(FS, FS.getPrecision(), /* precision */ 1, 2324 startSpecifier, specifierLen); 2325 } 2326 2327 // Check each flag does not conflict with any other component. 2328 if (!FS.hasValidThousandsGroupingPrefix()) 2329 HandleFlag(FS, FS.hasThousandsGrouping(), startSpecifier, specifierLen); 2330 if (!FS.hasValidLeadingZeros()) 2331 HandleFlag(FS, FS.hasLeadingZeros(), startSpecifier, specifierLen); 2332 if (!FS.hasValidPlusPrefix()) 2333 HandleFlag(FS, FS.hasPlusPrefix(), startSpecifier, specifierLen); 2334 if (!FS.hasValidSpacePrefix()) 2335 HandleFlag(FS, FS.hasSpacePrefix(), startSpecifier, specifierLen); 2336 if (!FS.hasValidAlternativeForm()) 2337 HandleFlag(FS, FS.hasAlternativeForm(), startSpecifier, specifierLen); 2338 if (!FS.hasValidLeftJustified()) 2339 HandleFlag(FS, FS.isLeftJustified(), startSpecifier, specifierLen); 2340 2341 // Check that flags are not ignored by another flag 2342 if (FS.hasSpacePrefix() && FS.hasPlusPrefix()) // ' ' ignored by '+' 2343 HandleIgnoredFlag(FS, FS.hasSpacePrefix(), FS.hasPlusPrefix(), 2344 startSpecifier, specifierLen); 2345 if (FS.hasLeadingZeros() && FS.isLeftJustified()) // '0' ignored by '-' 2346 HandleIgnoredFlag(FS, FS.hasLeadingZeros(), FS.isLeftJustified(), 2347 startSpecifier, specifierLen); 2348 2349 // Check the length modifier is valid with the given conversion specifier. 2350 const LengthModifier &LM = FS.getLengthModifier(); 2351 if (!FS.hasValidLengthModifier()) 2352 EmitFormatDiagnostic(S.PDiag(diag::warn_format_nonsensical_length) 2353 << LM.toString() << CS.toString(), 2354 getLocationOfByte(LM.getStart()), 2355 /*IsStringLocation*/true, 2356 getSpecifierRange(startSpecifier, specifierLen), 2357 FixItHint::CreateRemoval( 2358 getSpecifierRange(LM.getStart(), 2359 LM.getLength()))); 2360 if (!FS.hasStandardLengthModifier()) 2361 HandleNonStandardLengthModifier(LM, startSpecifier, specifierLen); 2362 if (!FS.hasStandardConversionSpecifier(S.getLangOpts())) 2363 HandleNonStandardConversionSpecifier(CS, startSpecifier, specifierLen); 2364 if (!FS.hasStandardLengthConversionCombination()) 2365 HandleNonStandardConversionSpecification(LM, CS, startSpecifier, 2366 specifierLen); 2367 2368 // Are we using '%n'? 2369 if (CS.getKind() == ConversionSpecifier::nArg) { 2370 // Issue a warning about this being a possible security issue. 2371 EmitFormatDiagnostic(S.PDiag(diag::warn_printf_write_back), 2372 getLocationOfByte(CS.getStart()), 2373 /*IsStringLocation*/true, 2374 getSpecifierRange(startSpecifier, specifierLen)); 2375 // Continue checking the other format specifiers. 2376 return true; 2377 } 2378 2379 // The remaining checks depend on the data arguments. 2380 if (HasVAListArg) 2381 return true; 2382 2383 if (!CheckNumArgs(FS, CS, startSpecifier, specifierLen, argIndex)) 2384 return false; 2385 2386 // Now type check the data expression that matches the 2387 // format specifier. 2388 const Expr *Ex = getDataArg(argIndex); 2389 const analyze_printf::ArgTypeResult &ATR = FS.getArgType(S.Context, 2390 IsObjCLiteral); 2391 if (ATR.isValid() && !ATR.matchesType(S.Context, Ex->getType())) { 2392 // Check if we didn't match because of an implicit cast from a 'char' 2393 // or 'short' to an 'int'. This is done because printf is a varargs 2394 // function. 2395 if (const ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(Ex)) 2396 if (ICE->getType() == S.Context.IntTy || 2397 ICE->getType() == S.Context.UnsignedIntTy) { 2398 // All further checking is done on the subexpression. 2399 Ex = ICE->getSubExpr(); 2400 if (ATR.matchesType(S.Context, Ex->getType())) 2401 return true; 2402 } 2403 2404 // We may be able to offer a FixItHint if it is a supported type. 2405 PrintfSpecifier fixedFS = FS; 2406 bool success = fixedFS.fixType(Ex->getType(), S.getLangOpts(), 2407 S.Context, IsObjCLiteral); 2408 2409 if (success) { 2410 // Get the fix string from the fixed format specifier 2411 SmallString<128> buf; 2412 llvm::raw_svector_ostream os(buf); 2413 fixedFS.toString(os); 2414 2415 EmitFormatDiagnostic( 2416 S.PDiag(diag::warn_printf_conversion_argument_type_mismatch) 2417 << ATR.getRepresentativeTypeName(S.Context) << Ex->getType() 2418 << Ex->getSourceRange(), 2419 getLocationOfByte(CS.getStart()), 2420 /*IsStringLocation*/true, 2421 getSpecifierRange(startSpecifier, specifierLen), 2422 FixItHint::CreateReplacement( 2423 getSpecifierRange(startSpecifier, specifierLen), 2424 os.str())); 2425 } 2426 else { 2427 EmitFormatDiagnostic( 2428 S.PDiag(diag::warn_printf_conversion_argument_type_mismatch) 2429 << ATR.getRepresentativeTypeName(S.Context) << Ex->getType() 2430 << getSpecifierRange(startSpecifier, specifierLen) 2431 << Ex->getSourceRange(), 2432 getLocationOfByte(CS.getStart()), 2433 true, 2434 getSpecifierRange(startSpecifier, specifierLen)); 2435 } 2436 } 2437 2438 return true; 2439} 2440 2441//===--- CHECK: Scanf format string checking ------------------------------===// 2442 2443namespace { 2444class CheckScanfHandler : public CheckFormatHandler { 2445public: 2446 CheckScanfHandler(Sema &s, const StringLiteral *fexpr, 2447 const Expr *origFormatExpr, unsigned firstDataArg, 2448 unsigned numDataArgs, bool isObjCLiteral, 2449 const char *beg, bool hasVAListArg, 2450 Expr **Args, unsigned NumArgs, 2451 unsigned formatIdx, bool inFunctionCall) 2452 : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg, 2453 numDataArgs, isObjCLiteral, beg, hasVAListArg, 2454 Args, NumArgs, formatIdx, inFunctionCall) {} 2455 2456 bool HandleScanfSpecifier(const analyze_scanf::ScanfSpecifier &FS, 2457 const char *startSpecifier, 2458 unsigned specifierLen); 2459 2460 bool HandleInvalidScanfConversionSpecifier( 2461 const analyze_scanf::ScanfSpecifier &FS, 2462 const char *startSpecifier, 2463 unsigned specifierLen); 2464 2465 void HandleIncompleteScanList(const char *start, const char *end); 2466}; 2467} 2468 2469void CheckScanfHandler::HandleIncompleteScanList(const char *start, 2470 const char *end) { 2471 EmitFormatDiagnostic(S.PDiag(diag::warn_scanf_scanlist_incomplete), 2472 getLocationOfByte(end), /*IsStringLocation*/true, 2473 getSpecifierRange(start, end - start)); 2474} 2475 2476bool CheckScanfHandler::HandleInvalidScanfConversionSpecifier( 2477 const analyze_scanf::ScanfSpecifier &FS, 2478 const char *startSpecifier, 2479 unsigned specifierLen) { 2480 2481 const analyze_scanf::ScanfConversionSpecifier &CS = 2482 FS.getConversionSpecifier(); 2483 2484 return HandleInvalidConversionSpecifier(FS.getArgIndex(), 2485 getLocationOfByte(CS.getStart()), 2486 startSpecifier, specifierLen, 2487 CS.getStart(), CS.getLength()); 2488} 2489 2490bool CheckScanfHandler::HandleScanfSpecifier( 2491 const analyze_scanf::ScanfSpecifier &FS, 2492 const char *startSpecifier, 2493 unsigned specifierLen) { 2494 2495 using namespace analyze_scanf; 2496 using namespace analyze_format_string; 2497 2498 const ScanfConversionSpecifier &CS = FS.getConversionSpecifier(); 2499 2500 // Handle case where '%' and '*' don't consume an argument. These shouldn't 2501 // be used to decide if we are using positional arguments consistently. 2502 if (FS.consumesDataArgument()) { 2503 if (atFirstArg) { 2504 atFirstArg = false; 2505 usesPositionalArgs = FS.usesPositionalArg(); 2506 } 2507 else if (usesPositionalArgs != FS.usesPositionalArg()) { 2508 HandlePositionalNonpositionalArgs(getLocationOfByte(CS.getStart()), 2509 startSpecifier, specifierLen); 2510 return false; 2511 } 2512 } 2513 2514 // Check if the field with is non-zero. 2515 const OptionalAmount &Amt = FS.getFieldWidth(); 2516 if (Amt.getHowSpecified() == OptionalAmount::Constant) { 2517 if (Amt.getConstantAmount() == 0) { 2518 const CharSourceRange &R = getSpecifierRange(Amt.getStart(), 2519 Amt.getConstantLength()); 2520 EmitFormatDiagnostic(S.PDiag(diag::warn_scanf_nonzero_width), 2521 getLocationOfByte(Amt.getStart()), 2522 /*IsStringLocation*/true, R, 2523 FixItHint::CreateRemoval(R)); 2524 } 2525 } 2526 2527 if (!FS.consumesDataArgument()) { 2528 // FIXME: Technically specifying a precision or field width here 2529 // makes no sense. Worth issuing a warning at some point. 2530 return true; 2531 } 2532 2533 // Consume the argument. 2534 unsigned argIndex = FS.getArgIndex(); 2535 if (argIndex < NumDataArgs) { 2536 // The check to see if the argIndex is valid will come later. 2537 // We set the bit here because we may exit early from this 2538 // function if we encounter some other error. 2539 CoveredArgs.set(argIndex); 2540 } 2541 2542 // Check the length modifier is valid with the given conversion specifier. 2543 const LengthModifier &LM = FS.getLengthModifier(); 2544 if (!FS.hasValidLengthModifier()) { 2545 const CharSourceRange &R = getSpecifierRange(LM.getStart(), LM.getLength()); 2546 EmitFormatDiagnostic(S.PDiag(diag::warn_format_nonsensical_length) 2547 << LM.toString() << CS.toString() 2548 << getSpecifierRange(startSpecifier, specifierLen), 2549 getLocationOfByte(LM.getStart()), 2550 /*IsStringLocation*/true, R, 2551 FixItHint::CreateRemoval(R)); 2552 } 2553 2554 if (!FS.hasStandardLengthModifier()) 2555 HandleNonStandardLengthModifier(LM, startSpecifier, specifierLen); 2556 if (!FS.hasStandardConversionSpecifier(S.getLangOpts())) 2557 HandleNonStandardConversionSpecifier(CS, startSpecifier, specifierLen); 2558 if (!FS.hasStandardLengthConversionCombination()) 2559 HandleNonStandardConversionSpecification(LM, CS, startSpecifier, 2560 specifierLen); 2561 2562 // The remaining checks depend on the data arguments. 2563 if (HasVAListArg) 2564 return true; 2565 2566 if (!CheckNumArgs(FS, CS, startSpecifier, specifierLen, argIndex)) 2567 return false; 2568 2569 // Check that the argument type matches the format specifier. 2570 const Expr *Ex = getDataArg(argIndex); 2571 const analyze_scanf::ScanfArgTypeResult &ATR = FS.getArgType(S.Context); 2572 if (ATR.isValid() && !ATR.matchesType(S.Context, Ex->getType())) { 2573 ScanfSpecifier fixedFS = FS; 2574 bool success = fixedFS.fixType(Ex->getType(), S.getLangOpts(), 2575 S.Context); 2576 2577 if (success) { 2578 // Get the fix string from the fixed format specifier. 2579 SmallString<128> buf; 2580 llvm::raw_svector_ostream os(buf); 2581 fixedFS.toString(os); 2582 2583 EmitFormatDiagnostic( 2584 S.PDiag(diag::warn_printf_conversion_argument_type_mismatch) 2585 << ATR.getRepresentativeTypeName(S.Context) << Ex->getType() 2586 << Ex->getSourceRange(), 2587 getLocationOfByte(CS.getStart()), 2588 /*IsStringLocation*/true, 2589 getSpecifierRange(startSpecifier, specifierLen), 2590 FixItHint::CreateReplacement( 2591 getSpecifierRange(startSpecifier, specifierLen), 2592 os.str())); 2593 } else { 2594 EmitFormatDiagnostic( 2595 S.PDiag(diag::warn_printf_conversion_argument_type_mismatch) 2596 << ATR.getRepresentativeTypeName(S.Context) << Ex->getType() 2597 << Ex->getSourceRange(), 2598 getLocationOfByte(CS.getStart()), 2599 /*IsStringLocation*/true, 2600 getSpecifierRange(startSpecifier, specifierLen)); 2601 } 2602 } 2603 2604 return true; 2605} 2606 2607void Sema::CheckFormatString(const StringLiteral *FExpr, 2608 const Expr *OrigFormatExpr, 2609 Expr **Args, unsigned NumArgs, 2610 bool HasVAListArg, unsigned format_idx, 2611 unsigned firstDataArg, FormatStringType Type, 2612 bool inFunctionCall) { 2613 2614 // CHECK: is the format string a wide literal? 2615 if (!FExpr->isAscii()) { 2616 CheckFormatHandler::EmitFormatDiagnostic( 2617 *this, inFunctionCall, Args[format_idx], 2618 PDiag(diag::warn_format_string_is_wide_literal), FExpr->getLocStart(), 2619 /*IsStringLocation*/true, OrigFormatExpr->getSourceRange()); 2620 return; 2621 } 2622 2623 // Str - The format string. NOTE: this is NOT null-terminated! 2624 StringRef StrRef = FExpr->getString(); 2625 const char *Str = StrRef.data(); 2626 unsigned StrLen = StrRef.size(); 2627 const unsigned numDataArgs = NumArgs - firstDataArg; 2628 2629 // CHECK: empty format string? 2630 if (StrLen == 0 && numDataArgs > 0) { 2631 CheckFormatHandler::EmitFormatDiagnostic( 2632 *this, inFunctionCall, Args[format_idx], 2633 PDiag(diag::warn_empty_format_string), FExpr->getLocStart(), 2634 /*IsStringLocation*/true, OrigFormatExpr->getSourceRange()); 2635 return; 2636 } 2637 2638 if (Type == FST_Printf || Type == FST_NSString) { 2639 CheckPrintfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg, 2640 numDataArgs, isa<ObjCStringLiteral>(OrigFormatExpr), 2641 Str, HasVAListArg, Args, NumArgs, format_idx, 2642 inFunctionCall); 2643 2644 if (!analyze_format_string::ParsePrintfString(H, Str, Str + StrLen, 2645 getLangOpts())) 2646 H.DoneProcessing(); 2647 } else if (Type == FST_Scanf) { 2648 CheckScanfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg, 2649 numDataArgs, isa<ObjCStringLiteral>(OrigFormatExpr), 2650 Str, HasVAListArg, Args, NumArgs, format_idx, 2651 inFunctionCall); 2652 2653 if (!analyze_format_string::ParseScanfString(H, Str, Str + StrLen, 2654 getLangOpts())) 2655 H.DoneProcessing(); 2656 } // TODO: handle other formats 2657} 2658 2659//===--- CHECK: Standard memory functions ---------------------------------===// 2660 2661/// \brief Determine whether the given type is a dynamic class type (e.g., 2662/// whether it has a vtable). 2663static bool isDynamicClassType(QualType T) { 2664 if (CXXRecordDecl *Record = T->getAsCXXRecordDecl()) 2665 if (CXXRecordDecl *Definition = Record->getDefinition()) 2666 if (Definition->isDynamicClass()) 2667 return true; 2668 2669 return false; 2670} 2671 2672/// \brief If E is a sizeof expression, returns its argument expression, 2673/// otherwise returns NULL. 2674static const Expr *getSizeOfExprArg(const Expr* E) { 2675 if (const UnaryExprOrTypeTraitExpr *SizeOf = 2676 dyn_cast<UnaryExprOrTypeTraitExpr>(E)) 2677 if (SizeOf->getKind() == clang::UETT_SizeOf && !SizeOf->isArgumentType()) 2678 return SizeOf->getArgumentExpr()->IgnoreParenImpCasts(); 2679 2680 return 0; 2681} 2682 2683/// \brief If E is a sizeof expression, returns its argument type. 2684static QualType getSizeOfArgType(const Expr* E) { 2685 if (const UnaryExprOrTypeTraitExpr *SizeOf = 2686 dyn_cast<UnaryExprOrTypeTraitExpr>(E)) 2687 if (SizeOf->getKind() == clang::UETT_SizeOf) 2688 return SizeOf->getTypeOfArgument(); 2689 2690 return QualType(); 2691} 2692 2693/// \brief Check for dangerous or invalid arguments to memset(). 2694/// 2695/// This issues warnings on known problematic, dangerous or unspecified 2696/// arguments to the standard 'memset', 'memcpy', 'memmove', and 'memcmp' 2697/// function calls. 2698/// 2699/// \param Call The call expression to diagnose. 2700void Sema::CheckMemaccessArguments(const CallExpr *Call, 2701 unsigned BId, 2702 IdentifierInfo *FnName) { 2703 assert(BId != 0); 2704 2705 // It is possible to have a non-standard definition of memset. Validate 2706 // we have enough arguments, and if not, abort further checking. 2707 unsigned ExpectedNumArgs = (BId == Builtin::BIstrndup ? 2 : 3); 2708 if (Call->getNumArgs() < ExpectedNumArgs) 2709 return; 2710 2711 unsigned LastArg = (BId == Builtin::BImemset || 2712 BId == Builtin::BIstrndup ? 1 : 2); 2713 unsigned LenArg = (BId == Builtin::BIstrndup ? 1 : 2); 2714 const Expr *LenExpr = Call->getArg(LenArg)->IgnoreParenImpCasts(); 2715 2716 // We have special checking when the length is a sizeof expression. 2717 QualType SizeOfArgTy = getSizeOfArgType(LenExpr); 2718 const Expr *SizeOfArg = getSizeOfExprArg(LenExpr); 2719 llvm::FoldingSetNodeID SizeOfArgID; 2720 2721 for (unsigned ArgIdx = 0; ArgIdx != LastArg; ++ArgIdx) { 2722 const Expr *Dest = Call->getArg(ArgIdx)->IgnoreParenImpCasts(); 2723 SourceRange ArgRange = Call->getArg(ArgIdx)->getSourceRange(); 2724 2725 QualType DestTy = Dest->getType(); 2726 if (const PointerType *DestPtrTy = DestTy->getAs<PointerType>()) { 2727 QualType PointeeTy = DestPtrTy->getPointeeType(); 2728 2729 // Never warn about void type pointers. This can be used to suppress 2730 // false positives. 2731 if (PointeeTy->isVoidType()) 2732 continue; 2733 2734 // Catch "memset(p, 0, sizeof(p))" -- needs to be sizeof(*p). Do this by 2735 // actually comparing the expressions for equality. Because computing the 2736 // expression IDs can be expensive, we only do this if the diagnostic is 2737 // enabled. 2738 if (SizeOfArg && 2739 Diags.getDiagnosticLevel(diag::warn_sizeof_pointer_expr_memaccess, 2740 SizeOfArg->getExprLoc())) { 2741 // We only compute IDs for expressions if the warning is enabled, and 2742 // cache the sizeof arg's ID. 2743 if (SizeOfArgID == llvm::FoldingSetNodeID()) 2744 SizeOfArg->Profile(SizeOfArgID, Context, true); 2745 llvm::FoldingSetNodeID DestID; 2746 Dest->Profile(DestID, Context, true); 2747 if (DestID == SizeOfArgID) { 2748 // TODO: For strncpy() and friends, this could suggest sizeof(dst) 2749 // over sizeof(src) as well. 2750 unsigned ActionIdx = 0; // Default is to suggest dereferencing. 2751 if (const UnaryOperator *UnaryOp = dyn_cast<UnaryOperator>(Dest)) 2752 if (UnaryOp->getOpcode() == UO_AddrOf) 2753 ActionIdx = 1; // If its an address-of operator, just remove it. 2754 if (Context.getTypeSize(PointeeTy) == Context.getCharWidth()) 2755 ActionIdx = 2; // If the pointee's size is sizeof(char), 2756 // suggest an explicit length. 2757 unsigned DestSrcSelect = 2758 (BId == Builtin::BIstrndup ? 1 : ArgIdx); 2759 DiagRuntimeBehavior(SizeOfArg->getExprLoc(), Dest, 2760 PDiag(diag::warn_sizeof_pointer_expr_memaccess) 2761 << FnName << DestSrcSelect << ActionIdx 2762 << Dest->getSourceRange() 2763 << SizeOfArg->getSourceRange()); 2764 break; 2765 } 2766 } 2767 2768 // Also check for cases where the sizeof argument is the exact same 2769 // type as the memory argument, and where it points to a user-defined 2770 // record type. 2771 if (SizeOfArgTy != QualType()) { 2772 if (PointeeTy->isRecordType() && 2773 Context.typesAreCompatible(SizeOfArgTy, DestTy)) { 2774 DiagRuntimeBehavior(LenExpr->getExprLoc(), Dest, 2775 PDiag(diag::warn_sizeof_pointer_type_memaccess) 2776 << FnName << SizeOfArgTy << ArgIdx 2777 << PointeeTy << Dest->getSourceRange() 2778 << LenExpr->getSourceRange()); 2779 break; 2780 } 2781 } 2782 2783 // Always complain about dynamic classes. 2784 if (isDynamicClassType(PointeeTy)) { 2785 2786 unsigned OperationType = 0; 2787 // "overwritten" if we're warning about the destination for any call 2788 // but memcmp; otherwise a verb appropriate to the call. 2789 if (ArgIdx != 0 || BId == Builtin::BImemcmp) { 2790 if (BId == Builtin::BImemcpy) 2791 OperationType = 1; 2792 else if(BId == Builtin::BImemmove) 2793 OperationType = 2; 2794 else if (BId == Builtin::BImemcmp) 2795 OperationType = 3; 2796 } 2797 2798 DiagRuntimeBehavior( 2799 Dest->getExprLoc(), Dest, 2800 PDiag(diag::warn_dyn_class_memaccess) 2801 << (BId == Builtin::BImemcmp ? ArgIdx + 2 : ArgIdx) 2802 << FnName << PointeeTy 2803 << OperationType 2804 << Call->getCallee()->getSourceRange()); 2805 } else if (PointeeTy.hasNonTrivialObjCLifetime() && 2806 BId != Builtin::BImemset) 2807 DiagRuntimeBehavior( 2808 Dest->getExprLoc(), Dest, 2809 PDiag(diag::warn_arc_object_memaccess) 2810 << ArgIdx << FnName << PointeeTy 2811 << Call->getCallee()->getSourceRange()); 2812 else 2813 continue; 2814 2815 DiagRuntimeBehavior( 2816 Dest->getExprLoc(), Dest, 2817 PDiag(diag::note_bad_memaccess_silence) 2818 << FixItHint::CreateInsertion(ArgRange.getBegin(), "(void*)")); 2819 break; 2820 } 2821 } 2822} 2823 2824// A little helper routine: ignore addition and subtraction of integer literals. 2825// This intentionally does not ignore all integer constant expressions because 2826// we don't want to remove sizeof(). 2827static const Expr *ignoreLiteralAdditions(const Expr *Ex, ASTContext &Ctx) { 2828 Ex = Ex->IgnoreParenCasts(); 2829 2830 for (;;) { 2831 const BinaryOperator * BO = dyn_cast<BinaryOperator>(Ex); 2832 if (!BO || !BO->isAdditiveOp()) 2833 break; 2834 2835 const Expr *RHS = BO->getRHS()->IgnoreParenCasts(); 2836 const Expr *LHS = BO->getLHS()->IgnoreParenCasts(); 2837 2838 if (isa<IntegerLiteral>(RHS)) 2839 Ex = LHS; 2840 else if (isa<IntegerLiteral>(LHS)) 2841 Ex = RHS; 2842 else 2843 break; 2844 } 2845 2846 return Ex; 2847} 2848 2849// Warn if the user has made the 'size' argument to strlcpy or strlcat 2850// be the size of the source, instead of the destination. 2851void Sema::CheckStrlcpycatArguments(const CallExpr *Call, 2852 IdentifierInfo *FnName) { 2853 2854 // Don't crash if the user has the wrong number of arguments 2855 if (Call->getNumArgs() != 3) 2856 return; 2857 2858 const Expr *SrcArg = ignoreLiteralAdditions(Call->getArg(1), Context); 2859 const Expr *SizeArg = ignoreLiteralAdditions(Call->getArg(2), Context); 2860 const Expr *CompareWithSrc = NULL; 2861 2862 // Look for 'strlcpy(dst, x, sizeof(x))' 2863 if (const Expr *Ex = getSizeOfExprArg(SizeArg)) 2864 CompareWithSrc = Ex; 2865 else { 2866 // Look for 'strlcpy(dst, x, strlen(x))' 2867 if (const CallExpr *SizeCall = dyn_cast<CallExpr>(SizeArg)) { 2868 if (SizeCall->isBuiltinCall() == Builtin::BIstrlen 2869 && SizeCall->getNumArgs() == 1) 2870 CompareWithSrc = ignoreLiteralAdditions(SizeCall->getArg(0), Context); 2871 } 2872 } 2873 2874 if (!CompareWithSrc) 2875 return; 2876 2877 // Determine if the argument to sizeof/strlen is equal to the source 2878 // argument. In principle there's all kinds of things you could do 2879 // here, for instance creating an == expression and evaluating it with 2880 // EvaluateAsBooleanCondition, but this uses a more direct technique: 2881 const DeclRefExpr *SrcArgDRE = dyn_cast<DeclRefExpr>(SrcArg); 2882 if (!SrcArgDRE) 2883 return; 2884 2885 const DeclRefExpr *CompareWithSrcDRE = dyn_cast<DeclRefExpr>(CompareWithSrc); 2886 if (!CompareWithSrcDRE || 2887 SrcArgDRE->getDecl() != CompareWithSrcDRE->getDecl()) 2888 return; 2889 2890 const Expr *OriginalSizeArg = Call->getArg(2); 2891 Diag(CompareWithSrcDRE->getLocStart(), diag::warn_strlcpycat_wrong_size) 2892 << OriginalSizeArg->getSourceRange() << FnName; 2893 2894 // Output a FIXIT hint if the destination is an array (rather than a 2895 // pointer to an array). This could be enhanced to handle some 2896 // pointers if we know the actual size, like if DstArg is 'array+2' 2897 // we could say 'sizeof(array)-2'. 2898 const Expr *DstArg = Call->getArg(0)->IgnoreParenImpCasts(); 2899 QualType DstArgTy = DstArg->getType(); 2900 2901 // Only handle constant-sized or VLAs, but not flexible members. 2902 if (const ConstantArrayType *CAT = Context.getAsConstantArrayType(DstArgTy)) { 2903 // Only issue the FIXIT for arrays of size > 1. 2904 if (CAT->getSize().getSExtValue() <= 1) 2905 return; 2906 } else if (!DstArgTy->isVariableArrayType()) { 2907 return; 2908 } 2909 2910 SmallString<128> sizeString; 2911 llvm::raw_svector_ostream OS(sizeString); 2912 OS << "sizeof("; 2913 DstArg->printPretty(OS, Context, 0, getPrintingPolicy()); 2914 OS << ")"; 2915 2916 Diag(OriginalSizeArg->getLocStart(), diag::note_strlcpycat_wrong_size) 2917 << FixItHint::CreateReplacement(OriginalSizeArg->getSourceRange(), 2918 OS.str()); 2919} 2920 2921/// Check if two expressions refer to the same declaration. 2922static bool referToTheSameDecl(const Expr *E1, const Expr *E2) { 2923 if (const DeclRefExpr *D1 = dyn_cast_or_null<DeclRefExpr>(E1)) 2924 if (const DeclRefExpr *D2 = dyn_cast_or_null<DeclRefExpr>(E2)) 2925 return D1->getDecl() == D2->getDecl(); 2926 return false; 2927} 2928 2929static const Expr *getStrlenExprArg(const Expr *E) { 2930 if (const CallExpr *CE = dyn_cast<CallExpr>(E)) { 2931 const FunctionDecl *FD = CE->getDirectCallee(); 2932 if (!FD || FD->getMemoryFunctionKind() != Builtin::BIstrlen) 2933 return 0; 2934 return CE->getArg(0)->IgnoreParenCasts(); 2935 } 2936 return 0; 2937} 2938 2939// Warn on anti-patterns as the 'size' argument to strncat. 2940// The correct size argument should look like following: 2941// strncat(dst, src, sizeof(dst) - strlen(dest) - 1); 2942void Sema::CheckStrncatArguments(const CallExpr *CE, 2943 IdentifierInfo *FnName) { 2944 // Don't crash if the user has the wrong number of arguments. 2945 if (CE->getNumArgs() < 3) 2946 return; 2947 const Expr *DstArg = CE->getArg(0)->IgnoreParenCasts(); 2948 const Expr *SrcArg = CE->getArg(1)->IgnoreParenCasts(); 2949 const Expr *LenArg = CE->getArg(2)->IgnoreParenCasts(); 2950 2951 // Identify common expressions, which are wrongly used as the size argument 2952 // to strncat and may lead to buffer overflows. 2953 unsigned PatternType = 0; 2954 if (const Expr *SizeOfArg = getSizeOfExprArg(LenArg)) { 2955 // - sizeof(dst) 2956 if (referToTheSameDecl(SizeOfArg, DstArg)) 2957 PatternType = 1; 2958 // - sizeof(src) 2959 else if (referToTheSameDecl(SizeOfArg, SrcArg)) 2960 PatternType = 2; 2961 } else if (const BinaryOperator *BE = dyn_cast<BinaryOperator>(LenArg)) { 2962 if (BE->getOpcode() == BO_Sub) { 2963 const Expr *L = BE->getLHS()->IgnoreParenCasts(); 2964 const Expr *R = BE->getRHS()->IgnoreParenCasts(); 2965 // - sizeof(dst) - strlen(dst) 2966 if (referToTheSameDecl(DstArg, getSizeOfExprArg(L)) && 2967 referToTheSameDecl(DstArg, getStrlenExprArg(R))) 2968 PatternType = 1; 2969 // - sizeof(src) - (anything) 2970 else if (referToTheSameDecl(SrcArg, getSizeOfExprArg(L))) 2971 PatternType = 2; 2972 } 2973 } 2974 2975 if (PatternType == 0) 2976 return; 2977 2978 // Generate the diagnostic. 2979 SourceLocation SL = LenArg->getLocStart(); 2980 SourceRange SR = LenArg->getSourceRange(); 2981 SourceManager &SM = PP.getSourceManager(); 2982 2983 // If the function is defined as a builtin macro, do not show macro expansion. 2984 if (SM.isMacroArgExpansion(SL)) { 2985 SL = SM.getSpellingLoc(SL); 2986 SR = SourceRange(SM.getSpellingLoc(SR.getBegin()), 2987 SM.getSpellingLoc(SR.getEnd())); 2988 } 2989 2990 if (PatternType == 1) 2991 Diag(SL, diag::warn_strncat_large_size) << SR; 2992 else 2993 Diag(SL, diag::warn_strncat_src_size) << SR; 2994 2995 // Output a FIXIT hint if the destination is an array (rather than a 2996 // pointer to an array). This could be enhanced to handle some 2997 // pointers if we know the actual size, like if DstArg is 'array+2' 2998 // we could say 'sizeof(array)-2'. 2999 QualType DstArgTy = DstArg->getType(); 3000 3001 // Only handle constant-sized or VLAs, but not flexible members. 3002 if (const ConstantArrayType *CAT = Context.getAsConstantArrayType(DstArgTy)) { 3003 // Only issue the FIXIT for arrays of size > 1. 3004 if (CAT->getSize().getSExtValue() <= 1) 3005 return; 3006 } else if (!DstArgTy->isVariableArrayType()) { 3007 return; 3008 } 3009 3010 SmallString<128> sizeString; 3011 llvm::raw_svector_ostream OS(sizeString); 3012 OS << "sizeof("; 3013 DstArg->printPretty(OS, Context, 0, getPrintingPolicy()); 3014 OS << ") - "; 3015 OS << "strlen("; 3016 DstArg->printPretty(OS, Context, 0, getPrintingPolicy()); 3017 OS << ") - 1"; 3018 3019 Diag(SL, diag::note_strncat_wrong_size) 3020 << FixItHint::CreateReplacement(SR, OS.str()); 3021} 3022 3023//===--- CHECK: Return Address of Stack Variable --------------------------===// 3024 3025static Expr *EvalVal(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars, 3026 Decl *ParentDecl); 3027static Expr *EvalAddr(Expr* E, SmallVectorImpl<DeclRefExpr *> &refVars, 3028 Decl *ParentDecl); 3029 3030/// CheckReturnStackAddr - Check if a return statement returns the address 3031/// of a stack variable. 3032void 3033Sema::CheckReturnStackAddr(Expr *RetValExp, QualType lhsType, 3034 SourceLocation ReturnLoc) { 3035 3036 Expr *stackE = 0; 3037 SmallVector<DeclRefExpr *, 8> refVars; 3038 3039 // Perform checking for returned stack addresses, local blocks, 3040 // label addresses or references to temporaries. 3041 if (lhsType->isPointerType() || 3042 (!getLangOpts().ObjCAutoRefCount && lhsType->isBlockPointerType())) { 3043 stackE = EvalAddr(RetValExp, refVars, /*ParentDecl=*/0); 3044 } else if (lhsType->isReferenceType()) { 3045 stackE = EvalVal(RetValExp, refVars, /*ParentDecl=*/0); 3046 } 3047 3048 if (stackE == 0) 3049 return; // Nothing suspicious was found. 3050 3051 SourceLocation diagLoc; 3052 SourceRange diagRange; 3053 if (refVars.empty()) { 3054 diagLoc = stackE->getLocStart(); 3055 diagRange = stackE->getSourceRange(); 3056 } else { 3057 // We followed through a reference variable. 'stackE' contains the 3058 // problematic expression but we will warn at the return statement pointing 3059 // at the reference variable. We will later display the "trail" of 3060 // reference variables using notes. 3061 diagLoc = refVars[0]->getLocStart(); 3062 diagRange = refVars[0]->getSourceRange(); 3063 } 3064 3065 if (DeclRefExpr *DR = dyn_cast<DeclRefExpr>(stackE)) { //address of local var. 3066 Diag(diagLoc, lhsType->isReferenceType() ? diag::warn_ret_stack_ref 3067 : diag::warn_ret_stack_addr) 3068 << DR->getDecl()->getDeclName() << diagRange; 3069 } else if (isa<BlockExpr>(stackE)) { // local block. 3070 Diag(diagLoc, diag::err_ret_local_block) << diagRange; 3071 } else if (isa<AddrLabelExpr>(stackE)) { // address of label. 3072 Diag(diagLoc, diag::warn_ret_addr_label) << diagRange; 3073 } else { // local temporary. 3074 Diag(diagLoc, lhsType->isReferenceType() ? diag::warn_ret_local_temp_ref 3075 : diag::warn_ret_local_temp_addr) 3076 << diagRange; 3077 } 3078 3079 // Display the "trail" of reference variables that we followed until we 3080 // found the problematic expression using notes. 3081 for (unsigned i = 0, e = refVars.size(); i != e; ++i) { 3082 VarDecl *VD = cast<VarDecl>(refVars[i]->getDecl()); 3083 // If this var binds to another reference var, show the range of the next 3084 // var, otherwise the var binds to the problematic expression, in which case 3085 // show the range of the expression. 3086 SourceRange range = (i < e-1) ? refVars[i+1]->getSourceRange() 3087 : stackE->getSourceRange(); 3088 Diag(VD->getLocation(), diag::note_ref_var_local_bind) 3089 << VD->getDeclName() << range; 3090 } 3091} 3092 3093/// EvalAddr - EvalAddr and EvalVal are mutually recursive functions that 3094/// check if the expression in a return statement evaluates to an address 3095/// to a location on the stack, a local block, an address of a label, or a 3096/// reference to local temporary. The recursion is used to traverse the 3097/// AST of the return expression, with recursion backtracking when we 3098/// encounter a subexpression that (1) clearly does not lead to one of the 3099/// above problematic expressions (2) is something we cannot determine leads to 3100/// a problematic expression based on such local checking. 3101/// 3102/// Both EvalAddr and EvalVal follow through reference variables to evaluate 3103/// the expression that they point to. Such variables are added to the 3104/// 'refVars' vector so that we know what the reference variable "trail" was. 3105/// 3106/// EvalAddr processes expressions that are pointers that are used as 3107/// references (and not L-values). EvalVal handles all other values. 3108/// At the base case of the recursion is a check for the above problematic 3109/// expressions. 3110/// 3111/// This implementation handles: 3112/// 3113/// * pointer-to-pointer casts 3114/// * implicit conversions from array references to pointers 3115/// * taking the address of fields 3116/// * arbitrary interplay between "&" and "*" operators 3117/// * pointer arithmetic from an address of a stack variable 3118/// * taking the address of an array element where the array is on the stack 3119static Expr *EvalAddr(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars, 3120 Decl *ParentDecl) { 3121 if (E->isTypeDependent()) 3122 return NULL; 3123 3124 // We should only be called for evaluating pointer expressions. 3125 assert((E->getType()->isAnyPointerType() || 3126 E->getType()->isBlockPointerType() || 3127 E->getType()->isObjCQualifiedIdType()) && 3128 "EvalAddr only works on pointers"); 3129 3130 E = E->IgnoreParens(); 3131 3132 // Our "symbolic interpreter" is just a dispatch off the currently 3133 // viewed AST node. We then recursively traverse the AST by calling 3134 // EvalAddr and EvalVal appropriately. 3135 switch (E->getStmtClass()) { 3136 case Stmt::DeclRefExprClass: { 3137 DeclRefExpr *DR = cast<DeclRefExpr>(E); 3138 3139 if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl())) 3140 // If this is a reference variable, follow through to the expression that 3141 // it points to. 3142 if (V->hasLocalStorage() && 3143 V->getType()->isReferenceType() && V->hasInit()) { 3144 // Add the reference variable to the "trail". 3145 refVars.push_back(DR); 3146 return EvalAddr(V->getInit(), refVars, ParentDecl); 3147 } 3148 3149 return NULL; 3150 } 3151 3152 case Stmt::UnaryOperatorClass: { 3153 // The only unary operator that make sense to handle here 3154 // is AddrOf. All others don't make sense as pointers. 3155 UnaryOperator *U = cast<UnaryOperator>(E); 3156 3157 if (U->getOpcode() == UO_AddrOf) 3158 return EvalVal(U->getSubExpr(), refVars, ParentDecl); 3159 else 3160 return NULL; 3161 } 3162 3163 case Stmt::BinaryOperatorClass: { 3164 // Handle pointer arithmetic. All other binary operators are not valid 3165 // in this context. 3166 BinaryOperator *B = cast<BinaryOperator>(E); 3167 BinaryOperatorKind op = B->getOpcode(); 3168 3169 if (op != BO_Add && op != BO_Sub) 3170 return NULL; 3171 3172 Expr *Base = B->getLHS(); 3173 3174 // Determine which argument is the real pointer base. It could be 3175 // the RHS argument instead of the LHS. 3176 if (!Base->getType()->isPointerType()) Base = B->getRHS(); 3177 3178 assert (Base->getType()->isPointerType()); 3179 return EvalAddr(Base, refVars, ParentDecl); 3180 } 3181 3182 // For conditional operators we need to see if either the LHS or RHS are 3183 // valid DeclRefExpr*s. If one of them is valid, we return it. 3184 case Stmt::ConditionalOperatorClass: { 3185 ConditionalOperator *C = cast<ConditionalOperator>(E); 3186 3187 // Handle the GNU extension for missing LHS. 3188 if (Expr *lhsExpr = C->getLHS()) { 3189 // In C++, we can have a throw-expression, which has 'void' type. 3190 if (!lhsExpr->getType()->isVoidType()) 3191 if (Expr* LHS = EvalAddr(lhsExpr, refVars, ParentDecl)) 3192 return LHS; 3193 } 3194 3195 // In C++, we can have a throw-expression, which has 'void' type. 3196 if (C->getRHS()->getType()->isVoidType()) 3197 return NULL; 3198 3199 return EvalAddr(C->getRHS(), refVars, ParentDecl); 3200 } 3201 3202 case Stmt::BlockExprClass: 3203 if (cast<BlockExpr>(E)->getBlockDecl()->hasCaptures()) 3204 return E; // local block. 3205 return NULL; 3206 3207 case Stmt::AddrLabelExprClass: 3208 return E; // address of label. 3209 3210 case Stmt::ExprWithCleanupsClass: 3211 return EvalAddr(cast<ExprWithCleanups>(E)->getSubExpr(), refVars, 3212 ParentDecl); 3213 3214 // For casts, we need to handle conversions from arrays to 3215 // pointer values, and pointer-to-pointer conversions. 3216 case Stmt::ImplicitCastExprClass: 3217 case Stmt::CStyleCastExprClass: 3218 case Stmt::CXXFunctionalCastExprClass: 3219 case Stmt::ObjCBridgedCastExprClass: 3220 case Stmt::CXXStaticCastExprClass: 3221 case Stmt::CXXDynamicCastExprClass: 3222 case Stmt::CXXConstCastExprClass: 3223 case Stmt::CXXReinterpretCastExprClass: { 3224 Expr* SubExpr = cast<CastExpr>(E)->getSubExpr(); 3225 switch (cast<CastExpr>(E)->getCastKind()) { 3226 case CK_BitCast: 3227 case CK_LValueToRValue: 3228 case CK_NoOp: 3229 case CK_BaseToDerived: 3230 case CK_DerivedToBase: 3231 case CK_UncheckedDerivedToBase: 3232 case CK_Dynamic: 3233 case CK_CPointerToObjCPointerCast: 3234 case CK_BlockPointerToObjCPointerCast: 3235 case CK_AnyPointerToBlockPointerCast: 3236 return EvalAddr(SubExpr, refVars, ParentDecl); 3237 3238 case CK_ArrayToPointerDecay: 3239 return EvalVal(SubExpr, refVars, ParentDecl); 3240 3241 default: 3242 return 0; 3243 } 3244 } 3245 3246 case Stmt::MaterializeTemporaryExprClass: 3247 if (Expr *Result = EvalAddr( 3248 cast<MaterializeTemporaryExpr>(E)->GetTemporaryExpr(), 3249 refVars, ParentDecl)) 3250 return Result; 3251 3252 return E; 3253 3254 // Everything else: we simply don't reason about them. 3255 default: 3256 return NULL; 3257 } 3258} 3259 3260 3261/// EvalVal - This function is complements EvalAddr in the mutual recursion. 3262/// See the comments for EvalAddr for more details. 3263static Expr *EvalVal(Expr *E, SmallVectorImpl<DeclRefExpr *> &refVars, 3264 Decl *ParentDecl) { 3265do { 3266 // We should only be called for evaluating non-pointer expressions, or 3267 // expressions with a pointer type that are not used as references but instead 3268 // are l-values (e.g., DeclRefExpr with a pointer type). 3269 3270 // Our "symbolic interpreter" is just a dispatch off the currently 3271 // viewed AST node. We then recursively traverse the AST by calling 3272 // EvalAddr and EvalVal appropriately. 3273 3274 E = E->IgnoreParens(); 3275 switch (E->getStmtClass()) { 3276 case Stmt::ImplicitCastExprClass: { 3277 ImplicitCastExpr *IE = cast<ImplicitCastExpr>(E); 3278 if (IE->getValueKind() == VK_LValue) { 3279 E = IE->getSubExpr(); 3280 continue; 3281 } 3282 return NULL; 3283 } 3284 3285 case Stmt::ExprWithCleanupsClass: 3286 return EvalVal(cast<ExprWithCleanups>(E)->getSubExpr(), refVars,ParentDecl); 3287 3288 case Stmt::DeclRefExprClass: { 3289 // When we hit a DeclRefExpr we are looking at code that refers to a 3290 // variable's name. If it's not a reference variable we check if it has 3291 // local storage within the function, and if so, return the expression. 3292 DeclRefExpr *DR = cast<DeclRefExpr>(E); 3293 3294 if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl())) { 3295 // Check if it refers to itself, e.g. "int& i = i;". 3296 if (V == ParentDecl) 3297 return DR; 3298 3299 if (V->hasLocalStorage()) { 3300 if (!V->getType()->isReferenceType()) 3301 return DR; 3302 3303 // Reference variable, follow through to the expression that 3304 // it points to. 3305 if (V->hasInit()) { 3306 // Add the reference variable to the "trail". 3307 refVars.push_back(DR); 3308 return EvalVal(V->getInit(), refVars, V); 3309 } 3310 } 3311 } 3312 3313 return NULL; 3314 } 3315 3316 case Stmt::UnaryOperatorClass: { 3317 // The only unary operator that make sense to handle here 3318 // is Deref. All others don't resolve to a "name." This includes 3319 // handling all sorts of rvalues passed to a unary operator. 3320 UnaryOperator *U = cast<UnaryOperator>(E); 3321 3322 if (U->getOpcode() == UO_Deref) 3323 return EvalAddr(U->getSubExpr(), refVars, ParentDecl); 3324 3325 return NULL; 3326 } 3327 3328 case Stmt::ArraySubscriptExprClass: { 3329 // Array subscripts are potential references to data on the stack. We 3330 // retrieve the DeclRefExpr* for the array variable if it indeed 3331 // has local storage. 3332 return EvalAddr(cast<ArraySubscriptExpr>(E)->getBase(), refVars,ParentDecl); 3333 } 3334 3335 case Stmt::ConditionalOperatorClass: { 3336 // For conditional operators we need to see if either the LHS or RHS are 3337 // non-NULL Expr's. If one is non-NULL, we return it. 3338 ConditionalOperator *C = cast<ConditionalOperator>(E); 3339 3340 // Handle the GNU extension for missing LHS. 3341 if (Expr *lhsExpr = C->getLHS()) 3342 if (Expr *LHS = EvalVal(lhsExpr, refVars, ParentDecl)) 3343 return LHS; 3344 3345 return EvalVal(C->getRHS(), refVars, ParentDecl); 3346 } 3347 3348 // Accesses to members are potential references to data on the stack. 3349 case Stmt::MemberExprClass: { 3350 MemberExpr *M = cast<MemberExpr>(E); 3351 3352 // Check for indirect access. We only want direct field accesses. 3353 if (M->isArrow()) 3354 return NULL; 3355 3356 // Check whether the member type is itself a reference, in which case 3357 // we're not going to refer to the member, but to what the member refers to. 3358 if (M->getMemberDecl()->getType()->isReferenceType()) 3359 return NULL; 3360 3361 return EvalVal(M->getBase(), refVars, ParentDecl); 3362 } 3363 3364 case Stmt::MaterializeTemporaryExprClass: 3365 if (Expr *Result = EvalVal( 3366 cast<MaterializeTemporaryExpr>(E)->GetTemporaryExpr(), 3367 refVars, ParentDecl)) 3368 return Result; 3369 3370 return E; 3371 3372 default: 3373 // Check that we don't return or take the address of a reference to a 3374 // temporary. This is only useful in C++. 3375 if (!E->isTypeDependent() && E->isRValue()) 3376 return E; 3377 3378 // Everything else: we simply don't reason about them. 3379 return NULL; 3380 } 3381} while (true); 3382} 3383 3384//===--- CHECK: Floating-Point comparisons (-Wfloat-equal) ---------------===// 3385 3386/// Check for comparisons of floating point operands using != and ==. 3387/// Issue a warning if these are no self-comparisons, as they are not likely 3388/// to do what the programmer intended. 3389void Sema::CheckFloatComparison(SourceLocation Loc, Expr* LHS, Expr *RHS) { 3390 bool EmitWarning = true; 3391 3392 Expr* LeftExprSansParen = LHS->IgnoreParenImpCasts(); 3393 Expr* RightExprSansParen = RHS->IgnoreParenImpCasts(); 3394 3395 // Special case: check for x == x (which is OK). 3396 // Do not emit warnings for such cases. 3397 if (DeclRefExpr* DRL = dyn_cast<DeclRefExpr>(LeftExprSansParen)) 3398 if (DeclRefExpr* DRR = dyn_cast<DeclRefExpr>(RightExprSansParen)) 3399 if (DRL->getDecl() == DRR->getDecl()) 3400 EmitWarning = false; 3401 3402 3403 // Special case: check for comparisons against literals that can be exactly 3404 // represented by APFloat. In such cases, do not emit a warning. This 3405 // is a heuristic: often comparison against such literals are used to 3406 // detect if a value in a variable has not changed. This clearly can 3407 // lead to false negatives. 3408 if (EmitWarning) { 3409 if (FloatingLiteral* FLL = dyn_cast<FloatingLiteral>(LeftExprSansParen)) { 3410 if (FLL->isExact()) 3411 EmitWarning = false; 3412 } else 3413 if (FloatingLiteral* FLR = dyn_cast<FloatingLiteral>(RightExprSansParen)){ 3414 if (FLR->isExact()) 3415 EmitWarning = false; 3416 } 3417 } 3418 3419 // Check for comparisons with builtin types. 3420 if (EmitWarning) 3421 if (CallExpr* CL = dyn_cast<CallExpr>(LeftExprSansParen)) 3422 if (CL->isBuiltinCall()) 3423 EmitWarning = false; 3424 3425 if (EmitWarning) 3426 if (CallExpr* CR = dyn_cast<CallExpr>(RightExprSansParen)) 3427 if (CR->isBuiltinCall()) 3428 EmitWarning = false; 3429 3430 // Emit the diagnostic. 3431 if (EmitWarning) 3432 Diag(Loc, diag::warn_floatingpoint_eq) 3433 << LHS->getSourceRange() << RHS->getSourceRange(); 3434} 3435 3436//===--- CHECK: Integer mixed-sign comparisons (-Wsign-compare) --------===// 3437//===--- CHECK: Lossy implicit conversions (-Wconversion) --------------===// 3438 3439namespace { 3440 3441/// Structure recording the 'active' range of an integer-valued 3442/// expression. 3443struct IntRange { 3444 /// The number of bits active in the int. 3445 unsigned Width; 3446 3447 /// True if the int is known not to have negative values. 3448 bool NonNegative; 3449 3450 IntRange(unsigned Width, bool NonNegative) 3451 : Width(Width), NonNegative(NonNegative) 3452 {} 3453 3454 /// Returns the range of the bool type. 3455 static IntRange forBoolType() { 3456 return IntRange(1, true); 3457 } 3458 3459 /// Returns the range of an opaque value of the given integral type. 3460 static IntRange forValueOfType(ASTContext &C, QualType T) { 3461 return forValueOfCanonicalType(C, 3462 T->getCanonicalTypeInternal().getTypePtr()); 3463 } 3464 3465 /// Returns the range of an opaque value of a canonical integral type. 3466 static IntRange forValueOfCanonicalType(ASTContext &C, const Type *T) { 3467 assert(T->isCanonicalUnqualified()); 3468 3469 if (const VectorType *VT = dyn_cast<VectorType>(T)) 3470 T = VT->getElementType().getTypePtr(); 3471 if (const ComplexType *CT = dyn_cast<ComplexType>(T)) 3472 T = CT->getElementType().getTypePtr(); 3473 3474 // For enum types, use the known bit width of the enumerators. 3475 if (const EnumType *ET = dyn_cast<EnumType>(T)) { 3476 EnumDecl *Enum = ET->getDecl(); 3477 if (!Enum->isCompleteDefinition()) 3478 return IntRange(C.getIntWidth(QualType(T, 0)), false); 3479 3480 unsigned NumPositive = Enum->getNumPositiveBits(); 3481 unsigned NumNegative = Enum->getNumNegativeBits(); 3482 3483 return IntRange(std::max(NumPositive, NumNegative), NumNegative == 0); 3484 } 3485 3486 const BuiltinType *BT = cast<BuiltinType>(T); 3487 assert(BT->isInteger()); 3488 3489 return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger()); 3490 } 3491 3492 /// Returns the "target" range of a canonical integral type, i.e. 3493 /// the range of values expressible in the type. 3494 /// 3495 /// This matches forValueOfCanonicalType except that enums have the 3496 /// full range of their type, not the range of their enumerators. 3497 static IntRange forTargetOfCanonicalType(ASTContext &C, const Type *T) { 3498 assert(T->isCanonicalUnqualified()); 3499 3500 if (const VectorType *VT = dyn_cast<VectorType>(T)) 3501 T = VT->getElementType().getTypePtr(); 3502 if (const ComplexType *CT = dyn_cast<ComplexType>(T)) 3503 T = CT->getElementType().getTypePtr(); 3504 if (const EnumType *ET = dyn_cast<EnumType>(T)) 3505 T = C.getCanonicalType(ET->getDecl()->getIntegerType()).getTypePtr(); 3506 3507 const BuiltinType *BT = cast<BuiltinType>(T); 3508 assert(BT->isInteger()); 3509 3510 return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger()); 3511 } 3512 3513 /// Returns the supremum of two ranges: i.e. their conservative merge. 3514 static IntRange join(IntRange L, IntRange R) { 3515 return IntRange(std::max(L.Width, R.Width), 3516 L.NonNegative && R.NonNegative); 3517 } 3518 3519 /// Returns the infinum of two ranges: i.e. their aggressive merge. 3520 static IntRange meet(IntRange L, IntRange R) { 3521 return IntRange(std::min(L.Width, R.Width), 3522 L.NonNegative || R.NonNegative); 3523 } 3524}; 3525 3526static IntRange GetValueRange(ASTContext &C, llvm::APSInt &value, 3527 unsigned MaxWidth) { 3528 if (value.isSigned() && value.isNegative()) 3529 return IntRange(value.getMinSignedBits(), false); 3530 3531 if (value.getBitWidth() > MaxWidth) 3532 value = value.trunc(MaxWidth); 3533 3534 // isNonNegative() just checks the sign bit without considering 3535 // signedness. 3536 return IntRange(value.getActiveBits(), true); 3537} 3538 3539static IntRange GetValueRange(ASTContext &C, APValue &result, QualType Ty, 3540 unsigned MaxWidth) { 3541 if (result.isInt()) 3542 return GetValueRange(C, result.getInt(), MaxWidth); 3543 3544 if (result.isVector()) { 3545 IntRange R = GetValueRange(C, result.getVectorElt(0), Ty, MaxWidth); 3546 for (unsigned i = 1, e = result.getVectorLength(); i != e; ++i) { 3547 IntRange El = GetValueRange(C, result.getVectorElt(i), Ty, MaxWidth); 3548 R = IntRange::join(R, El); 3549 } 3550 return R; 3551 } 3552 3553 if (result.isComplexInt()) { 3554 IntRange R = GetValueRange(C, result.getComplexIntReal(), MaxWidth); 3555 IntRange I = GetValueRange(C, result.getComplexIntImag(), MaxWidth); 3556 return IntRange::join(R, I); 3557 } 3558 3559 // This can happen with lossless casts to intptr_t of "based" lvalues. 3560 // Assume it might use arbitrary bits. 3561 // FIXME: The only reason we need to pass the type in here is to get 3562 // the sign right on this one case. It would be nice if APValue 3563 // preserved this. 3564 assert(result.isLValue() || result.isAddrLabelDiff()); 3565 return IntRange(MaxWidth, Ty->isUnsignedIntegerOrEnumerationType()); 3566} 3567 3568/// Pseudo-evaluate the given integer expression, estimating the 3569/// range of values it might take. 3570/// 3571/// \param MaxWidth - the width to which the value will be truncated 3572static IntRange GetExprRange(ASTContext &C, Expr *E, unsigned MaxWidth) { 3573 E = E->IgnoreParens(); 3574 3575 // Try a full evaluation first. 3576 Expr::EvalResult result; 3577 if (E->EvaluateAsRValue(result, C)) 3578 return GetValueRange(C, result.Val, E->getType(), MaxWidth); 3579 3580 // I think we only want to look through implicit casts here; if the 3581 // user has an explicit widening cast, we should treat the value as 3582 // being of the new, wider type. 3583 if (ImplicitCastExpr *CE = dyn_cast<ImplicitCastExpr>(E)) { 3584 if (CE->getCastKind() == CK_NoOp || CE->getCastKind() == CK_LValueToRValue) 3585 return GetExprRange(C, CE->getSubExpr(), MaxWidth); 3586 3587 IntRange OutputTypeRange = IntRange::forValueOfType(C, CE->getType()); 3588 3589 bool isIntegerCast = (CE->getCastKind() == CK_IntegralCast); 3590 3591 // Assume that non-integer casts can span the full range of the type. 3592 if (!isIntegerCast) 3593 return OutputTypeRange; 3594 3595 IntRange SubRange 3596 = GetExprRange(C, CE->getSubExpr(), 3597 std::min(MaxWidth, OutputTypeRange.Width)); 3598 3599 // Bail out if the subexpr's range is as wide as the cast type. 3600 if (SubRange.Width >= OutputTypeRange.Width) 3601 return OutputTypeRange; 3602 3603 // Otherwise, we take the smaller width, and we're non-negative if 3604 // either the output type or the subexpr is. 3605 return IntRange(SubRange.Width, 3606 SubRange.NonNegative || OutputTypeRange.NonNegative); 3607 } 3608 3609 if (ConditionalOperator *CO = dyn_cast<ConditionalOperator>(E)) { 3610 // If we can fold the condition, just take that operand. 3611 bool CondResult; 3612 if (CO->getCond()->EvaluateAsBooleanCondition(CondResult, C)) 3613 return GetExprRange(C, CondResult ? CO->getTrueExpr() 3614 : CO->getFalseExpr(), 3615 MaxWidth); 3616 3617 // Otherwise, conservatively merge. 3618 IntRange L = GetExprRange(C, CO->getTrueExpr(), MaxWidth); 3619 IntRange R = GetExprRange(C, CO->getFalseExpr(), MaxWidth); 3620 return IntRange::join(L, R); 3621 } 3622 3623 if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) { 3624 switch (BO->getOpcode()) { 3625 3626 // Boolean-valued operations are single-bit and positive. 3627 case BO_LAnd: 3628 case BO_LOr: 3629 case BO_LT: 3630 case BO_GT: 3631 case BO_LE: 3632 case BO_GE: 3633 case BO_EQ: 3634 case BO_NE: 3635 return IntRange::forBoolType(); 3636 3637 // The type of the assignments is the type of the LHS, so the RHS 3638 // is not necessarily the same type. 3639 case BO_MulAssign: 3640 case BO_DivAssign: 3641 case BO_RemAssign: 3642 case BO_AddAssign: 3643 case BO_SubAssign: 3644 case BO_XorAssign: 3645 case BO_OrAssign: 3646 // TODO: bitfields? 3647 return IntRange::forValueOfType(C, E->getType()); 3648 3649 // Simple assignments just pass through the RHS, which will have 3650 // been coerced to the LHS type. 3651 case BO_Assign: 3652 // TODO: bitfields? 3653 return GetExprRange(C, BO->getRHS(), MaxWidth); 3654 3655 // Operations with opaque sources are black-listed. 3656 case BO_PtrMemD: 3657 case BO_PtrMemI: 3658 return IntRange::forValueOfType(C, E->getType()); 3659 3660 // Bitwise-and uses the *infinum* of the two source ranges. 3661 case BO_And: 3662 case BO_AndAssign: 3663 return IntRange::meet(GetExprRange(C, BO->getLHS(), MaxWidth), 3664 GetExprRange(C, BO->getRHS(), MaxWidth)); 3665 3666 // Left shift gets black-listed based on a judgement call. 3667 case BO_Shl: 3668 // ...except that we want to treat '1 << (blah)' as logically 3669 // positive. It's an important idiom. 3670 if (IntegerLiteral *I 3671 = dyn_cast<IntegerLiteral>(BO->getLHS()->IgnoreParenCasts())) { 3672 if (I->getValue() == 1) { 3673 IntRange R = IntRange::forValueOfType(C, E->getType()); 3674 return IntRange(R.Width, /*NonNegative*/ true); 3675 } 3676 } 3677 // fallthrough 3678 3679 case BO_ShlAssign: 3680 return IntRange::forValueOfType(C, E->getType()); 3681 3682 // Right shift by a constant can narrow its left argument. 3683 case BO_Shr: 3684 case BO_ShrAssign: { 3685 IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth); 3686 3687 // If the shift amount is a positive constant, drop the width by 3688 // that much. 3689 llvm::APSInt shift; 3690 if (BO->getRHS()->isIntegerConstantExpr(shift, C) && 3691 shift.isNonNegative()) { 3692 unsigned zext = shift.getZExtValue(); 3693 if (zext >= L.Width) 3694 L.Width = (L.NonNegative ? 0 : 1); 3695 else 3696 L.Width -= zext; 3697 } 3698 3699 return L; 3700 } 3701 3702 // Comma acts as its right operand. 3703 case BO_Comma: 3704 return GetExprRange(C, BO->getRHS(), MaxWidth); 3705 3706 // Black-list pointer subtractions. 3707 case BO_Sub: 3708 if (BO->getLHS()->getType()->isPointerType()) 3709 return IntRange::forValueOfType(C, E->getType()); 3710 break; 3711 3712 // The width of a division result is mostly determined by the size 3713 // of the LHS. 3714 case BO_Div: { 3715 // Don't 'pre-truncate' the operands. 3716 unsigned opWidth = C.getIntWidth(E->getType()); 3717 IntRange L = GetExprRange(C, BO->getLHS(), opWidth); 3718 3719 // If the divisor is constant, use that. 3720 llvm::APSInt divisor; 3721 if (BO->getRHS()->isIntegerConstantExpr(divisor, C)) { 3722 unsigned log2 = divisor.logBase2(); // floor(log_2(divisor)) 3723 if (log2 >= L.Width) 3724 L.Width = (L.NonNegative ? 0 : 1); 3725 else 3726 L.Width = std::min(L.Width - log2, MaxWidth); 3727 return L; 3728 } 3729 3730 // Otherwise, just use the LHS's width. 3731 IntRange R = GetExprRange(C, BO->getRHS(), opWidth); 3732 return IntRange(L.Width, L.NonNegative && R.NonNegative); 3733 } 3734 3735 // The result of a remainder can't be larger than the result of 3736 // either side. 3737 case BO_Rem: { 3738 // Don't 'pre-truncate' the operands. 3739 unsigned opWidth = C.getIntWidth(E->getType()); 3740 IntRange L = GetExprRange(C, BO->getLHS(), opWidth); 3741 IntRange R = GetExprRange(C, BO->getRHS(), opWidth); 3742 3743 IntRange meet = IntRange::meet(L, R); 3744 meet.Width = std::min(meet.Width, MaxWidth); 3745 return meet; 3746 } 3747 3748 // The default behavior is okay for these. 3749 case BO_Mul: 3750 case BO_Add: 3751 case BO_Xor: 3752 case BO_Or: 3753 break; 3754 } 3755 3756 // The default case is to treat the operation as if it were closed 3757 // on the narrowest type that encompasses both operands. 3758 IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth); 3759 IntRange R = GetExprRange(C, BO->getRHS(), MaxWidth); 3760 return IntRange::join(L, R); 3761 } 3762 3763 if (UnaryOperator *UO = dyn_cast<UnaryOperator>(E)) { 3764 switch (UO->getOpcode()) { 3765 // Boolean-valued operations are white-listed. 3766 case UO_LNot: 3767 return IntRange::forBoolType(); 3768 3769 // Operations with opaque sources are black-listed. 3770 case UO_Deref: 3771 case UO_AddrOf: // should be impossible 3772 return IntRange::forValueOfType(C, E->getType()); 3773 3774 default: 3775 return GetExprRange(C, UO->getSubExpr(), MaxWidth); 3776 } 3777 } 3778 3779 if (dyn_cast<OffsetOfExpr>(E)) { 3780 IntRange::forValueOfType(C, E->getType()); 3781 } 3782 3783 if (FieldDecl *BitField = E->getBitField()) 3784 return IntRange(BitField->getBitWidthValue(C), 3785 BitField->getType()->isUnsignedIntegerOrEnumerationType()); 3786 3787 return IntRange::forValueOfType(C, E->getType()); 3788} 3789 3790static IntRange GetExprRange(ASTContext &C, Expr *E) { 3791 return GetExprRange(C, E, C.getIntWidth(E->getType())); 3792} 3793 3794/// Checks whether the given value, which currently has the given 3795/// source semantics, has the same value when coerced through the 3796/// target semantics. 3797static bool IsSameFloatAfterCast(const llvm::APFloat &value, 3798 const llvm::fltSemantics &Src, 3799 const llvm::fltSemantics &Tgt) { 3800 llvm::APFloat truncated = value; 3801 3802 bool ignored; 3803 truncated.convert(Src, llvm::APFloat::rmNearestTiesToEven, &ignored); 3804 truncated.convert(Tgt, llvm::APFloat::rmNearestTiesToEven, &ignored); 3805 3806 return truncated.bitwiseIsEqual(value); 3807} 3808 3809/// Checks whether the given value, which currently has the given 3810/// source semantics, has the same value when coerced through the 3811/// target semantics. 3812/// 3813/// The value might be a vector of floats (or a complex number). 3814static bool IsSameFloatAfterCast(const APValue &value, 3815 const llvm::fltSemantics &Src, 3816 const llvm::fltSemantics &Tgt) { 3817 if (value.isFloat()) 3818 return IsSameFloatAfterCast(value.getFloat(), Src, Tgt); 3819 3820 if (value.isVector()) { 3821 for (unsigned i = 0, e = value.getVectorLength(); i != e; ++i) 3822 if (!IsSameFloatAfterCast(value.getVectorElt(i), Src, Tgt)) 3823 return false; 3824 return true; 3825 } 3826 3827 assert(value.isComplexFloat()); 3828 return (IsSameFloatAfterCast(value.getComplexFloatReal(), Src, Tgt) && 3829 IsSameFloatAfterCast(value.getComplexFloatImag(), Src, Tgt)); 3830} 3831 3832static void AnalyzeImplicitConversions(Sema &S, Expr *E, SourceLocation CC); 3833 3834static bool IsZero(Sema &S, Expr *E) { 3835 // Suppress cases where we are comparing against an enum constant. 3836 if (const DeclRefExpr *DR = 3837 dyn_cast<DeclRefExpr>(E->IgnoreParenImpCasts())) 3838 if (isa<EnumConstantDecl>(DR->getDecl())) 3839 return false; 3840 3841 // Suppress cases where the '0' value is expanded from a macro. 3842 if (E->getLocStart().isMacroID()) 3843 return false; 3844 3845 llvm::APSInt Value; 3846 return E->isIntegerConstantExpr(Value, S.Context) && Value == 0; 3847} 3848 3849static bool HasEnumType(Expr *E) { 3850 // Strip off implicit integral promotions. 3851 while (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(E)) { 3852 if (ICE->getCastKind() != CK_IntegralCast && 3853 ICE->getCastKind() != CK_NoOp) 3854 break; 3855 E = ICE->getSubExpr(); 3856 } 3857 3858 return E->getType()->isEnumeralType(); 3859} 3860 3861static void CheckTrivialUnsignedComparison(Sema &S, BinaryOperator *E) { 3862 BinaryOperatorKind op = E->getOpcode(); 3863 if (E->isValueDependent()) 3864 return; 3865 3866 if (op == BO_LT && IsZero(S, E->getRHS())) { 3867 S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison) 3868 << "< 0" << "false" << HasEnumType(E->getLHS()) 3869 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 3870 } else if (op == BO_GE && IsZero(S, E->getRHS())) { 3871 S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison) 3872 << ">= 0" << "true" << HasEnumType(E->getLHS()) 3873 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 3874 } else if (op == BO_GT && IsZero(S, E->getLHS())) { 3875 S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison) 3876 << "0 >" << "false" << HasEnumType(E->getRHS()) 3877 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 3878 } else if (op == BO_LE && IsZero(S, E->getLHS())) { 3879 S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison) 3880 << "0 <=" << "true" << HasEnumType(E->getRHS()) 3881 << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange(); 3882 } 3883} 3884 3885/// Analyze the operands of the given comparison. Implements the 3886/// fallback case from AnalyzeComparison. 3887static void AnalyzeImpConvsInComparison(Sema &S, BinaryOperator *E) { 3888 AnalyzeImplicitConversions(S, E->getLHS(), E->getOperatorLoc()); 3889 AnalyzeImplicitConversions(S, E->getRHS(), E->getOperatorLoc()); 3890} 3891 3892/// \brief Implements -Wsign-compare. 3893/// 3894/// \param E the binary operator to check for warnings 3895static void AnalyzeComparison(Sema &S, BinaryOperator *E) { 3896 // The type the comparison is being performed in. 3897 QualType T = E->getLHS()->getType(); 3898 assert(S.Context.hasSameUnqualifiedType(T, E->getRHS()->getType()) 3899 && "comparison with mismatched types"); 3900 3901 // We don't do anything special if this isn't an unsigned integral 3902 // comparison: we're only interested in integral comparisons, and 3903 // signed comparisons only happen in cases we don't care to warn about. 3904 // 3905 // We also don't care about value-dependent expressions or expressions 3906 // whose result is a constant. 3907 if (!T->hasUnsignedIntegerRepresentation() 3908 || E->isValueDependent() || E->isIntegerConstantExpr(S.Context)) 3909 return AnalyzeImpConvsInComparison(S, E); 3910 3911 Expr *LHS = E->getLHS()->IgnoreParenImpCasts(); 3912 Expr *RHS = E->getRHS()->IgnoreParenImpCasts(); 3913 3914 // Check to see if one of the (unmodified) operands is of different 3915 // signedness. 3916 Expr *signedOperand, *unsignedOperand; 3917 if (LHS->getType()->hasSignedIntegerRepresentation()) { 3918 assert(!RHS->getType()->hasSignedIntegerRepresentation() && 3919 "unsigned comparison between two signed integer expressions?"); 3920 signedOperand = LHS; 3921 unsignedOperand = RHS; 3922 } else if (RHS->getType()->hasSignedIntegerRepresentation()) { 3923 signedOperand = RHS; 3924 unsignedOperand = LHS; 3925 } else { 3926 CheckTrivialUnsignedComparison(S, E); 3927 return AnalyzeImpConvsInComparison(S, E); 3928 } 3929 3930 // Otherwise, calculate the effective range of the signed operand. 3931 IntRange signedRange = GetExprRange(S.Context, signedOperand); 3932 3933 // Go ahead and analyze implicit conversions in the operands. Note 3934 // that we skip the implicit conversions on both sides. 3935 AnalyzeImplicitConversions(S, LHS, E->getOperatorLoc()); 3936 AnalyzeImplicitConversions(S, RHS, E->getOperatorLoc()); 3937 3938 // If the signed range is non-negative, -Wsign-compare won't fire, 3939 // but we should still check for comparisons which are always true 3940 // or false. 3941 if (signedRange.NonNegative) 3942 return CheckTrivialUnsignedComparison(S, E); 3943 3944 // For (in)equality comparisons, if the unsigned operand is a 3945 // constant which cannot collide with a overflowed signed operand, 3946 // then reinterpreting the signed operand as unsigned will not 3947 // change the result of the comparison. 3948 if (E->isEqualityOp()) { 3949 unsigned comparisonWidth = S.Context.getIntWidth(T); 3950 IntRange unsignedRange = GetExprRange(S.Context, unsignedOperand); 3951 3952 // We should never be unable to prove that the unsigned operand is 3953 // non-negative. 3954 assert(unsignedRange.NonNegative && "unsigned range includes negative?"); 3955 3956 if (unsignedRange.Width < comparisonWidth) 3957 return; 3958 } 3959 3960 S.DiagRuntimeBehavior(E->getOperatorLoc(), E, 3961 S.PDiag(diag::warn_mixed_sign_comparison) 3962 << LHS->getType() << RHS->getType() 3963 << LHS->getSourceRange() << RHS->getSourceRange()); 3964} 3965 3966/// Analyzes an attempt to assign the given value to a bitfield. 3967/// 3968/// Returns true if there was something fishy about the attempt. 3969static bool AnalyzeBitFieldAssignment(Sema &S, FieldDecl *Bitfield, Expr *Init, 3970 SourceLocation InitLoc) { 3971 assert(Bitfield->isBitField()); 3972 if (Bitfield->isInvalidDecl()) 3973 return false; 3974 3975 // White-list bool bitfields. 3976 if (Bitfield->getType()->isBooleanType()) 3977 return false; 3978 3979 // Ignore value- or type-dependent expressions. 3980 if (Bitfield->getBitWidth()->isValueDependent() || 3981 Bitfield->getBitWidth()->isTypeDependent() || 3982 Init->isValueDependent() || 3983 Init->isTypeDependent()) 3984 return false; 3985 3986 Expr *OriginalInit = Init->IgnoreParenImpCasts(); 3987 3988 llvm::APSInt Value; 3989 if (!OriginalInit->EvaluateAsInt(Value, S.Context, Expr::SE_AllowSideEffects)) 3990 return false; 3991 3992 unsigned OriginalWidth = Value.getBitWidth(); 3993 unsigned FieldWidth = Bitfield->getBitWidthValue(S.Context); 3994 3995 if (OriginalWidth <= FieldWidth) 3996 return false; 3997 3998 // Compute the value which the bitfield will contain. 3999 llvm::APSInt TruncatedValue = Value.trunc(FieldWidth); 4000 TruncatedValue.setIsSigned(Bitfield->getType()->isSignedIntegerType()); 4001 4002 // Check whether the stored value is equal to the original value. 4003 TruncatedValue = TruncatedValue.extend(OriginalWidth); 4004 if (Value == TruncatedValue) 4005 return false; 4006 4007 // Special-case bitfields of width 1: booleans are naturally 0/1, and 4008 // therefore don't strictly fit into a signed bitfield of width 1. 4009 if (FieldWidth == 1 && Value == 1) 4010 return false; 4011 4012 std::string PrettyValue = Value.toString(10); 4013 std::string PrettyTrunc = TruncatedValue.toString(10); 4014 4015 S.Diag(InitLoc, diag::warn_impcast_bitfield_precision_constant) 4016 << PrettyValue << PrettyTrunc << OriginalInit->getType() 4017 << Init->getSourceRange(); 4018 4019 return true; 4020} 4021 4022/// Analyze the given simple or compound assignment for warning-worthy 4023/// operations. 4024static void AnalyzeAssignment(Sema &S, BinaryOperator *E) { 4025 // Just recurse on the LHS. 4026 AnalyzeImplicitConversions(S, E->getLHS(), E->getOperatorLoc()); 4027 4028 // We want to recurse on the RHS as normal unless we're assigning to 4029 // a bitfield. 4030 if (FieldDecl *Bitfield = E->getLHS()->getBitField()) { 4031 if (AnalyzeBitFieldAssignment(S, Bitfield, E->getRHS(), 4032 E->getOperatorLoc())) { 4033 // Recurse, ignoring any implicit conversions on the RHS. 4034 return AnalyzeImplicitConversions(S, E->getRHS()->IgnoreParenImpCasts(), 4035 E->getOperatorLoc()); 4036 } 4037 } 4038 4039 AnalyzeImplicitConversions(S, E->getRHS(), E->getOperatorLoc()); 4040} 4041 4042/// Diagnose an implicit cast; purely a helper for CheckImplicitConversion. 4043static void DiagnoseImpCast(Sema &S, Expr *E, QualType SourceType, QualType T, 4044 SourceLocation CContext, unsigned diag, 4045 bool pruneControlFlow = false) { 4046 if (pruneControlFlow) { 4047 S.DiagRuntimeBehavior(E->getExprLoc(), E, 4048 S.PDiag(diag) 4049 << SourceType << T << E->getSourceRange() 4050 << SourceRange(CContext)); 4051 return; 4052 } 4053 S.Diag(E->getExprLoc(), diag) 4054 << SourceType << T << E->getSourceRange() << SourceRange(CContext); 4055} 4056 4057/// Diagnose an implicit cast; purely a helper for CheckImplicitConversion. 4058static void DiagnoseImpCast(Sema &S, Expr *E, QualType T, 4059 SourceLocation CContext, unsigned diag, 4060 bool pruneControlFlow = false) { 4061 DiagnoseImpCast(S, E, E->getType(), T, CContext, diag, pruneControlFlow); 4062} 4063 4064/// Diagnose an implicit cast from a literal expression. Does not warn when the 4065/// cast wouldn't lose information. 4066void DiagnoseFloatingLiteralImpCast(Sema &S, FloatingLiteral *FL, QualType T, 4067 SourceLocation CContext) { 4068 // Try to convert the literal exactly to an integer. If we can, don't warn. 4069 bool isExact = false; 4070 const llvm::APFloat &Value = FL->getValue(); 4071 llvm::APSInt IntegerValue(S.Context.getIntWidth(T), 4072 T->hasUnsignedIntegerRepresentation()); 4073 if (Value.convertToInteger(IntegerValue, 4074 llvm::APFloat::rmTowardZero, &isExact) 4075 == llvm::APFloat::opOK && isExact) 4076 return; 4077 4078 S.Diag(FL->getExprLoc(), diag::warn_impcast_literal_float_to_integer) 4079 << FL->getType() << T << FL->getSourceRange() << SourceRange(CContext); 4080} 4081 4082std::string PrettyPrintInRange(const llvm::APSInt &Value, IntRange Range) { 4083 if (!Range.Width) return "0"; 4084 4085 llvm::APSInt ValueInRange = Value; 4086 ValueInRange.setIsSigned(!Range.NonNegative); 4087 ValueInRange = ValueInRange.trunc(Range.Width); 4088 return ValueInRange.toString(10); 4089} 4090 4091void CheckImplicitConversion(Sema &S, Expr *E, QualType T, 4092 SourceLocation CC, bool *ICContext = 0) { 4093 if (E->isTypeDependent() || E->isValueDependent()) return; 4094 4095 const Type *Source = S.Context.getCanonicalType(E->getType()).getTypePtr(); 4096 const Type *Target = S.Context.getCanonicalType(T).getTypePtr(); 4097 if (Source == Target) return; 4098 if (Target->isDependentType()) return; 4099 4100 // If the conversion context location is invalid don't complain. We also 4101 // don't want to emit a warning if the issue occurs from the expansion of 4102 // a system macro. The problem is that 'getSpellingLoc()' is slow, so we 4103 // delay this check as long as possible. Once we detect we are in that 4104 // scenario, we just return. 4105 if (CC.isInvalid()) 4106 return; 4107 4108 // Diagnose implicit casts to bool. 4109 if (Target->isSpecificBuiltinType(BuiltinType::Bool)) { 4110 if (isa<StringLiteral>(E)) 4111 // Warn on string literal to bool. Checks for string literals in logical 4112 // expressions, for instances, assert(0 && "error here"), is prevented 4113 // by a check in AnalyzeImplicitConversions(). 4114 return DiagnoseImpCast(S, E, T, CC, 4115 diag::warn_impcast_string_literal_to_bool); 4116 if (Source->isFunctionType()) { 4117 // Warn on function to bool. Checks free functions and static member 4118 // functions. Weakly imported functions are excluded from the check, 4119 // since it's common to test their value to check whether the linker 4120 // found a definition for them. 4121 ValueDecl *D = 0; 4122 if (DeclRefExpr* R = dyn_cast<DeclRefExpr>(E)) { 4123 D = R->getDecl(); 4124 } else if (MemberExpr *M = dyn_cast<MemberExpr>(E)) { 4125 D = M->getMemberDecl(); 4126 } 4127 4128 if (D && !D->isWeak()) { 4129 if (FunctionDecl* F = dyn_cast<FunctionDecl>(D)) { 4130 S.Diag(E->getExprLoc(), diag::warn_impcast_function_to_bool) 4131 << F << E->getSourceRange() << SourceRange(CC); 4132 S.Diag(E->getExprLoc(), diag::note_function_to_bool_silence) 4133 << FixItHint::CreateInsertion(E->getExprLoc(), "&"); 4134 QualType ReturnType; 4135 UnresolvedSet<4> NonTemplateOverloads; 4136 S.isExprCallable(*E, ReturnType, NonTemplateOverloads); 4137 if (!ReturnType.isNull() 4138 && ReturnType->isSpecificBuiltinType(BuiltinType::Bool)) 4139 S.Diag(E->getExprLoc(), diag::note_function_to_bool_call) 4140 << FixItHint::CreateInsertion( 4141 S.getPreprocessor().getLocForEndOfToken(E->getLocEnd()), "()"); 4142 return; 4143 } 4144 } 4145 } 4146 return; // Other casts to bool are not checked. 4147 } 4148 4149 // Strip vector types. 4150 if (isa<VectorType>(Source)) { 4151 if (!isa<VectorType>(Target)) { 4152 if (S.SourceMgr.isInSystemMacro(CC)) 4153 return; 4154 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_vector_scalar); 4155 } 4156 4157 // If the vector cast is cast between two vectors of the same size, it is 4158 // a bitcast, not a conversion. 4159 if (S.Context.getTypeSize(Source) == S.Context.getTypeSize(Target)) 4160 return; 4161 4162 Source = cast<VectorType>(Source)->getElementType().getTypePtr(); 4163 Target = cast<VectorType>(Target)->getElementType().getTypePtr(); 4164 } 4165 4166 // Strip complex types. 4167 if (isa<ComplexType>(Source)) { 4168 if (!isa<ComplexType>(Target)) { 4169 if (S.SourceMgr.isInSystemMacro(CC)) 4170 return; 4171 4172 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_complex_scalar); 4173 } 4174 4175 Source = cast<ComplexType>(Source)->getElementType().getTypePtr(); 4176 Target = cast<ComplexType>(Target)->getElementType().getTypePtr(); 4177 } 4178 4179 const BuiltinType *SourceBT = dyn_cast<BuiltinType>(Source); 4180 const BuiltinType *TargetBT = dyn_cast<BuiltinType>(Target); 4181 4182 // If the source is floating point... 4183 if (SourceBT && SourceBT->isFloatingPoint()) { 4184 // ...and the target is floating point... 4185 if (TargetBT && TargetBT->isFloatingPoint()) { 4186 // ...then warn if we're dropping FP rank. 4187 4188 // Builtin FP kinds are ordered by increasing FP rank. 4189 if (SourceBT->getKind() > TargetBT->getKind()) { 4190 // Don't warn about float constants that are precisely 4191 // representable in the target type. 4192 Expr::EvalResult result; 4193 if (E->EvaluateAsRValue(result, S.Context)) { 4194 // Value might be a float, a float vector, or a float complex. 4195 if (IsSameFloatAfterCast(result.Val, 4196 S.Context.getFloatTypeSemantics(QualType(TargetBT, 0)), 4197 S.Context.getFloatTypeSemantics(QualType(SourceBT, 0)))) 4198 return; 4199 } 4200 4201 if (S.SourceMgr.isInSystemMacro(CC)) 4202 return; 4203 4204 DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_float_precision); 4205 } 4206 return; 4207 } 4208 4209 // If the target is integral, always warn. 4210 if ((TargetBT && TargetBT->isInteger())) { 4211 if (S.SourceMgr.isInSystemMacro(CC)) 4212 return; 4213 4214 Expr *InnerE = E->IgnoreParenImpCasts(); 4215 // We also want to warn on, e.g., "int i = -1.234" 4216 if (UnaryOperator *UOp = dyn_cast<UnaryOperator>(InnerE)) 4217 if (UOp->getOpcode() == UO_Minus || UOp->getOpcode() == UO_Plus) 4218 InnerE = UOp->getSubExpr()->IgnoreParenImpCasts(); 4219 4220 if (FloatingLiteral *FL = dyn_cast<FloatingLiteral>(InnerE)) { 4221 DiagnoseFloatingLiteralImpCast(S, FL, T, CC); 4222 } else { 4223 DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_float_integer); 4224 } 4225 } 4226 4227 return; 4228 } 4229 4230 if (!Source->isIntegerType() || !Target->isIntegerType()) 4231 return; 4232 4233 if ((E->isNullPointerConstant(S.Context, Expr::NPC_ValueDependentIsNotNull) 4234 == Expr::NPCK_GNUNull) && Target->isIntegerType()) { 4235 SourceLocation Loc = E->getSourceRange().getBegin(); 4236 if (Loc.isMacroID()) 4237 Loc = S.SourceMgr.getImmediateExpansionRange(Loc).first; 4238 S.Diag(Loc, diag::warn_impcast_null_pointer_to_integer) 4239 << T << clang::SourceRange(CC) 4240 << FixItHint::CreateReplacement(Loc, S.getFixItZeroLiteralForType(T)); 4241 return; 4242 } 4243 4244 IntRange SourceRange = GetExprRange(S.Context, E); 4245 IntRange TargetRange = IntRange::forTargetOfCanonicalType(S.Context, Target); 4246 4247 if (SourceRange.Width > TargetRange.Width) { 4248 // If the source is a constant, use a default-on diagnostic. 4249 // TODO: this should happen for bitfield stores, too. 4250 llvm::APSInt Value(32); 4251 if (E->isIntegerConstantExpr(Value, S.Context)) { 4252 if (S.SourceMgr.isInSystemMacro(CC)) 4253 return; 4254 4255 std::string PrettySourceValue = Value.toString(10); 4256 std::string PrettyTargetValue = PrettyPrintInRange(Value, TargetRange); 4257 4258 S.DiagRuntimeBehavior(E->getExprLoc(), E, 4259 S.PDiag(diag::warn_impcast_integer_precision_constant) 4260 << PrettySourceValue << PrettyTargetValue 4261 << E->getType() << T << E->getSourceRange() 4262 << clang::SourceRange(CC)); 4263 return; 4264 } 4265 4266 // People want to build with -Wshorten-64-to-32 and not -Wconversion. 4267 if (S.SourceMgr.isInSystemMacro(CC)) 4268 return; 4269 4270 if (TargetRange.Width == 32 && S.Context.getIntWidth(E->getType()) == 64) 4271 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_integer_64_32, 4272 /* pruneControlFlow */ true); 4273 return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_integer_precision); 4274 } 4275 4276 if ((TargetRange.NonNegative && !SourceRange.NonNegative) || 4277 (!TargetRange.NonNegative && SourceRange.NonNegative && 4278 SourceRange.Width == TargetRange.Width)) { 4279 4280 if (S.SourceMgr.isInSystemMacro(CC)) 4281 return; 4282 4283 unsigned DiagID = diag::warn_impcast_integer_sign; 4284 4285 // Traditionally, gcc has warned about this under -Wsign-compare. 4286 // We also want to warn about it in -Wconversion. 4287 // So if -Wconversion is off, use a completely identical diagnostic 4288 // in the sign-compare group. 4289 // The conditional-checking code will 4290 if (ICContext) { 4291 DiagID = diag::warn_impcast_integer_sign_conditional; 4292 *ICContext = true; 4293 } 4294 4295 return DiagnoseImpCast(S, E, T, CC, DiagID); 4296 } 4297 4298 // Diagnose conversions between different enumeration types. 4299 // In C, we pretend that the type of an EnumConstantDecl is its enumeration 4300 // type, to give us better diagnostics. 4301 QualType SourceType = E->getType(); 4302 if (!S.getLangOpts().CPlusPlus) { 4303 if (DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(E)) 4304 if (EnumConstantDecl *ECD = dyn_cast<EnumConstantDecl>(DRE->getDecl())) { 4305 EnumDecl *Enum = cast<EnumDecl>(ECD->getDeclContext()); 4306 SourceType = S.Context.getTypeDeclType(Enum); 4307 Source = S.Context.getCanonicalType(SourceType).getTypePtr(); 4308 } 4309 } 4310 4311 if (const EnumType *SourceEnum = Source->getAs<EnumType>()) 4312 if (const EnumType *TargetEnum = Target->getAs<EnumType>()) 4313 if ((SourceEnum->getDecl()->getIdentifier() || 4314 SourceEnum->getDecl()->getTypedefNameForAnonDecl()) && 4315 (TargetEnum->getDecl()->getIdentifier() || 4316 TargetEnum->getDecl()->getTypedefNameForAnonDecl()) && 4317 SourceEnum != TargetEnum) { 4318 if (S.SourceMgr.isInSystemMacro(CC)) 4319 return; 4320 4321 return DiagnoseImpCast(S, E, SourceType, T, CC, 4322 diag::warn_impcast_different_enum_types); 4323 } 4324 4325 return; 4326} 4327 4328void CheckConditionalOperator(Sema &S, ConditionalOperator *E, QualType T); 4329 4330void CheckConditionalOperand(Sema &S, Expr *E, QualType T, 4331 SourceLocation CC, bool &ICContext) { 4332 E = E->IgnoreParenImpCasts(); 4333 4334 if (isa<ConditionalOperator>(E)) 4335 return CheckConditionalOperator(S, cast<ConditionalOperator>(E), T); 4336 4337 AnalyzeImplicitConversions(S, E, CC); 4338 if (E->getType() != T) 4339 return CheckImplicitConversion(S, E, T, CC, &ICContext); 4340 return; 4341} 4342 4343void CheckConditionalOperator(Sema &S, ConditionalOperator *E, QualType T) { 4344 SourceLocation CC = E->getQuestionLoc(); 4345 4346 AnalyzeImplicitConversions(S, E->getCond(), CC); 4347 4348 bool Suspicious = false; 4349 CheckConditionalOperand(S, E->getTrueExpr(), T, CC, Suspicious); 4350 CheckConditionalOperand(S, E->getFalseExpr(), T, CC, Suspicious); 4351 4352 // If -Wconversion would have warned about either of the candidates 4353 // for a signedness conversion to the context type... 4354 if (!Suspicious) return; 4355 4356 // ...but it's currently ignored... 4357 if (S.Diags.getDiagnosticLevel(diag::warn_impcast_integer_sign_conditional, 4358 CC)) 4359 return; 4360 4361 // ...then check whether it would have warned about either of the 4362 // candidates for a signedness conversion to the condition type. 4363 if (E->getType() == T) return; 4364 4365 Suspicious = false; 4366 CheckImplicitConversion(S, E->getTrueExpr()->IgnoreParenImpCasts(), 4367 E->getType(), CC, &Suspicious); 4368 if (!Suspicious) 4369 CheckImplicitConversion(S, E->getFalseExpr()->IgnoreParenImpCasts(), 4370 E->getType(), CC, &Suspicious); 4371} 4372 4373/// AnalyzeImplicitConversions - Find and report any interesting 4374/// implicit conversions in the given expression. There are a couple 4375/// of competing diagnostics here, -Wconversion and -Wsign-compare. 4376void AnalyzeImplicitConversions(Sema &S, Expr *OrigE, SourceLocation CC) { 4377 QualType T = OrigE->getType(); 4378 Expr *E = OrigE->IgnoreParenImpCasts(); 4379 4380 if (E->isTypeDependent() || E->isValueDependent()) 4381 return; 4382 4383 // For conditional operators, we analyze the arguments as if they 4384 // were being fed directly into the output. 4385 if (isa<ConditionalOperator>(E)) { 4386 ConditionalOperator *CO = cast<ConditionalOperator>(E); 4387 CheckConditionalOperator(S, CO, T); 4388 return; 4389 } 4390 4391 // Go ahead and check any implicit conversions we might have skipped. 4392 // The non-canonical typecheck is just an optimization; 4393 // CheckImplicitConversion will filter out dead implicit conversions. 4394 if (E->getType() != T) 4395 CheckImplicitConversion(S, E, T, CC); 4396 4397 // Now continue drilling into this expression. 4398 4399 // Skip past explicit casts. 4400 if (isa<ExplicitCastExpr>(E)) { 4401 E = cast<ExplicitCastExpr>(E)->getSubExpr()->IgnoreParenImpCasts(); 4402 return AnalyzeImplicitConversions(S, E, CC); 4403 } 4404 4405 if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) { 4406 // Do a somewhat different check with comparison operators. 4407 if (BO->isComparisonOp()) 4408 return AnalyzeComparison(S, BO); 4409 4410 // And with simple assignments. 4411 if (BO->getOpcode() == BO_Assign) 4412 return AnalyzeAssignment(S, BO); 4413 } 4414 4415 // These break the otherwise-useful invariant below. Fortunately, 4416 // we don't really need to recurse into them, because any internal 4417 // expressions should have been analyzed already when they were 4418 // built into statements. 4419 if (isa<StmtExpr>(E)) return; 4420 4421 // Don't descend into unevaluated contexts. 4422 if (isa<UnaryExprOrTypeTraitExpr>(E)) return; 4423 4424 // Now just recurse over the expression's children. 4425 CC = E->getExprLoc(); 4426 BinaryOperator *BO = dyn_cast<BinaryOperator>(E); 4427 bool IsLogicalOperator = BO && BO->isLogicalOp(); 4428 for (Stmt::child_range I = E->children(); I; ++I) { 4429 Expr *ChildExpr = dyn_cast_or_null<Expr>(*I); 4430 if (!ChildExpr) 4431 continue; 4432 4433 if (IsLogicalOperator && 4434 isa<StringLiteral>(ChildExpr->IgnoreParenImpCasts())) 4435 // Ignore checking string literals that are in logical operators. 4436 continue; 4437 AnalyzeImplicitConversions(S, ChildExpr, CC); 4438 } 4439} 4440 4441} // end anonymous namespace 4442 4443/// Diagnoses "dangerous" implicit conversions within the given 4444/// expression (which is a full expression). Implements -Wconversion 4445/// and -Wsign-compare. 4446/// 4447/// \param CC the "context" location of the implicit conversion, i.e. 4448/// the most location of the syntactic entity requiring the implicit 4449/// conversion 4450void Sema::CheckImplicitConversions(Expr *E, SourceLocation CC) { 4451 // Don't diagnose in unevaluated contexts. 4452 if (ExprEvalContexts.back().Context == Sema::Unevaluated) 4453 return; 4454 4455 // Don't diagnose for value- or type-dependent expressions. 4456 if (E->isTypeDependent() || E->isValueDependent()) 4457 return; 4458 4459 // Check for array bounds violations in cases where the check isn't triggered 4460 // elsewhere for other Expr types (like BinaryOperators), e.g. when an 4461 // ArraySubscriptExpr is on the RHS of a variable initialization. 4462 CheckArrayAccess(E); 4463 4464 // This is not the right CC for (e.g.) a variable initialization. 4465 AnalyzeImplicitConversions(*this, E, CC); 4466} 4467 4468void Sema::CheckBitFieldInitialization(SourceLocation InitLoc, 4469 FieldDecl *BitField, 4470 Expr *Init) { 4471 (void) AnalyzeBitFieldAssignment(*this, BitField, Init, InitLoc); 4472} 4473 4474/// CheckParmsForFunctionDef - Check that the parameters of the given 4475/// function are appropriate for the definition of a function. This 4476/// takes care of any checks that cannot be performed on the 4477/// declaration itself, e.g., that the types of each of the function 4478/// parameters are complete. 4479bool Sema::CheckParmsForFunctionDef(ParmVarDecl **P, ParmVarDecl **PEnd, 4480 bool CheckParameterNames) { 4481 bool HasInvalidParm = false; 4482 for (; P != PEnd; ++P) { 4483 ParmVarDecl *Param = *P; 4484 4485 // C99 6.7.5.3p4: the parameters in a parameter type list in a 4486 // function declarator that is part of a function definition of 4487 // that function shall not have incomplete type. 4488 // 4489 // This is also C++ [dcl.fct]p6. 4490 if (!Param->isInvalidDecl() && 4491 RequireCompleteType(Param->getLocation(), Param->getType(), 4492 diag::err_typecheck_decl_incomplete_type)) { 4493 Param->setInvalidDecl(); 4494 HasInvalidParm = true; 4495 } 4496 4497 // C99 6.9.1p5: If the declarator includes a parameter type list, the 4498 // declaration of each parameter shall include an identifier. 4499 if (CheckParameterNames && 4500 Param->getIdentifier() == 0 && 4501 !Param->isImplicit() && 4502 !getLangOpts().CPlusPlus) 4503 Diag(Param->getLocation(), diag::err_parameter_name_omitted); 4504 4505 // C99 6.7.5.3p12: 4506 // If the function declarator is not part of a definition of that 4507 // function, parameters may have incomplete type and may use the [*] 4508 // notation in their sequences of declarator specifiers to specify 4509 // variable length array types. 4510 QualType PType = Param->getOriginalType(); 4511 if (const ArrayType *AT = Context.getAsArrayType(PType)) { 4512 if (AT->getSizeModifier() == ArrayType::Star) { 4513 // FIXME: This diagnosic should point the the '[*]' if source-location 4514 // information is added for it. 4515 Diag(Param->getLocation(), diag::err_array_star_in_function_definition); 4516 } 4517 } 4518 } 4519 4520 return HasInvalidParm; 4521} 4522 4523/// CheckCastAlign - Implements -Wcast-align, which warns when a 4524/// pointer cast increases the alignment requirements. 4525void Sema::CheckCastAlign(Expr *Op, QualType T, SourceRange TRange) { 4526 // This is actually a lot of work to potentially be doing on every 4527 // cast; don't do it if we're ignoring -Wcast_align (as is the default). 4528 if (getDiagnostics().getDiagnosticLevel(diag::warn_cast_align, 4529 TRange.getBegin()) 4530 == DiagnosticsEngine::Ignored) 4531 return; 4532 4533 // Ignore dependent types. 4534 if (T->isDependentType() || Op->getType()->isDependentType()) 4535 return; 4536 4537 // Require that the destination be a pointer type. 4538 const PointerType *DestPtr = T->getAs<PointerType>(); 4539 if (!DestPtr) return; 4540 4541 // If the destination has alignment 1, we're done. 4542 QualType DestPointee = DestPtr->getPointeeType(); 4543 if (DestPointee->isIncompleteType()) return; 4544 CharUnits DestAlign = Context.getTypeAlignInChars(DestPointee); 4545 if (DestAlign.isOne()) return; 4546 4547 // Require that the source be a pointer type. 4548 const PointerType *SrcPtr = Op->getType()->getAs<PointerType>(); 4549 if (!SrcPtr) return; 4550 QualType SrcPointee = SrcPtr->getPointeeType(); 4551 4552 // Whitelist casts from cv void*. We already implicitly 4553 // whitelisted casts to cv void*, since they have alignment 1. 4554 // Also whitelist casts involving incomplete types, which implicitly 4555 // includes 'void'. 4556 if (SrcPointee->isIncompleteType()) return; 4557 4558 CharUnits SrcAlign = Context.getTypeAlignInChars(SrcPointee); 4559 if (SrcAlign >= DestAlign) return; 4560 4561 Diag(TRange.getBegin(), diag::warn_cast_align) 4562 << Op->getType() << T 4563 << static_cast<unsigned>(SrcAlign.getQuantity()) 4564 << static_cast<unsigned>(DestAlign.getQuantity()) 4565 << TRange << Op->getSourceRange(); 4566} 4567 4568static const Type* getElementType(const Expr *BaseExpr) { 4569 const Type* EltType = BaseExpr->getType().getTypePtr(); 4570 if (EltType->isAnyPointerType()) 4571 return EltType->getPointeeType().getTypePtr(); 4572 else if (EltType->isArrayType()) 4573 return EltType->getBaseElementTypeUnsafe(); 4574 return EltType; 4575} 4576 4577/// \brief Check whether this array fits the idiom of a size-one tail padded 4578/// array member of a struct. 4579/// 4580/// We avoid emitting out-of-bounds access warnings for such arrays as they are 4581/// commonly used to emulate flexible arrays in C89 code. 4582static bool IsTailPaddedMemberArray(Sema &S, llvm::APInt Size, 4583 const NamedDecl *ND) { 4584 if (Size != 1 || !ND) return false; 4585 4586 const FieldDecl *FD = dyn_cast<FieldDecl>(ND); 4587 if (!FD) return false; 4588 4589 // Don't consider sizes resulting from macro expansions or template argument 4590 // substitution to form C89 tail-padded arrays. 4591 4592 TypeSourceInfo *TInfo = FD->getTypeSourceInfo(); 4593 while (TInfo) { 4594 TypeLoc TL = TInfo->getTypeLoc(); 4595 // Look through typedefs. 4596 const TypedefTypeLoc *TTL = dyn_cast<TypedefTypeLoc>(&TL); 4597 if (TTL) { 4598 const TypedefNameDecl *TDL = TTL->getTypedefNameDecl(); 4599 TInfo = TDL->getTypeSourceInfo(); 4600 continue; 4601 } 4602 ConstantArrayTypeLoc CTL = cast<ConstantArrayTypeLoc>(TL); 4603 const Expr *SizeExpr = dyn_cast<IntegerLiteral>(CTL.getSizeExpr()); 4604 if (!SizeExpr || SizeExpr->getExprLoc().isMacroID()) 4605 return false; 4606 break; 4607 } 4608 4609 const RecordDecl *RD = dyn_cast<RecordDecl>(FD->getDeclContext()); 4610 if (!RD) return false; 4611 if (RD->isUnion()) return false; 4612 if (const CXXRecordDecl *CRD = dyn_cast<CXXRecordDecl>(RD)) { 4613 if (!CRD->isStandardLayout()) return false; 4614 } 4615 4616 // See if this is the last field decl in the record. 4617 const Decl *D = FD; 4618 while ((D = D->getNextDeclInContext())) 4619 if (isa<FieldDecl>(D)) 4620 return false; 4621 return true; 4622} 4623 4624void Sema::CheckArrayAccess(const Expr *BaseExpr, const Expr *IndexExpr, 4625 const ArraySubscriptExpr *ASE, 4626 bool AllowOnePastEnd, bool IndexNegated) { 4627 IndexExpr = IndexExpr->IgnoreParenImpCasts(); 4628 if (IndexExpr->isValueDependent()) 4629 return; 4630 4631 const Type *EffectiveType = getElementType(BaseExpr); 4632 BaseExpr = BaseExpr->IgnoreParenCasts(); 4633 const ConstantArrayType *ArrayTy = 4634 Context.getAsConstantArrayType(BaseExpr->getType()); 4635 if (!ArrayTy) 4636 return; 4637 4638 llvm::APSInt index; 4639 if (!IndexExpr->EvaluateAsInt(index, Context)) 4640 return; 4641 if (IndexNegated) 4642 index = -index; 4643 4644 const NamedDecl *ND = NULL; 4645 if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(BaseExpr)) 4646 ND = dyn_cast<NamedDecl>(DRE->getDecl()); 4647 if (const MemberExpr *ME = dyn_cast<MemberExpr>(BaseExpr)) 4648 ND = dyn_cast<NamedDecl>(ME->getMemberDecl()); 4649 4650 if (index.isUnsigned() || !index.isNegative()) { 4651 llvm::APInt size = ArrayTy->getSize(); 4652 if (!size.isStrictlyPositive()) 4653 return; 4654 4655 const Type* BaseType = getElementType(BaseExpr); 4656 if (BaseType != EffectiveType) { 4657 // Make sure we're comparing apples to apples when comparing index to size 4658 uint64_t ptrarith_typesize = Context.getTypeSize(EffectiveType); 4659 uint64_t array_typesize = Context.getTypeSize(BaseType); 4660 // Handle ptrarith_typesize being zero, such as when casting to void* 4661 if (!ptrarith_typesize) ptrarith_typesize = 1; 4662 if (ptrarith_typesize != array_typesize) { 4663 // There's a cast to a different size type involved 4664 uint64_t ratio = array_typesize / ptrarith_typesize; 4665 // TODO: Be smarter about handling cases where array_typesize is not a 4666 // multiple of ptrarith_typesize 4667 if (ptrarith_typesize * ratio == array_typesize) 4668 size *= llvm::APInt(size.getBitWidth(), ratio); 4669 } 4670 } 4671 4672 if (size.getBitWidth() > index.getBitWidth()) 4673 index = index.zext(size.getBitWidth()); 4674 else if (size.getBitWidth() < index.getBitWidth()) 4675 size = size.zext(index.getBitWidth()); 4676 4677 // For array subscripting the index must be less than size, but for pointer 4678 // arithmetic also allow the index (offset) to be equal to size since 4679 // computing the next address after the end of the array is legal and 4680 // commonly done e.g. in C++ iterators and range-based for loops. 4681 if (AllowOnePastEnd ? index.ule(size) : index.ult(size)) 4682 return; 4683 4684 // Also don't warn for arrays of size 1 which are members of some 4685 // structure. These are often used to approximate flexible arrays in C89 4686 // code. 4687 if (IsTailPaddedMemberArray(*this, size, ND)) 4688 return; 4689 4690 // Suppress the warning if the subscript expression (as identified by the 4691 // ']' location) and the index expression are both from macro expansions 4692 // within a system header. 4693 if (ASE) { 4694 SourceLocation RBracketLoc = SourceMgr.getSpellingLoc( 4695 ASE->getRBracketLoc()); 4696 if (SourceMgr.isInSystemHeader(RBracketLoc)) { 4697 SourceLocation IndexLoc = SourceMgr.getSpellingLoc( 4698 IndexExpr->getLocStart()); 4699 if (SourceMgr.isFromSameFile(RBracketLoc, IndexLoc)) 4700 return; 4701 } 4702 } 4703 4704 unsigned DiagID = diag::warn_ptr_arith_exceeds_bounds; 4705 if (ASE) 4706 DiagID = diag::warn_array_index_exceeds_bounds; 4707 4708 DiagRuntimeBehavior(BaseExpr->getLocStart(), BaseExpr, 4709 PDiag(DiagID) << index.toString(10, true) 4710 << size.toString(10, true) 4711 << (unsigned)size.getLimitedValue(~0U) 4712 << IndexExpr->getSourceRange()); 4713 } else { 4714 unsigned DiagID = diag::warn_array_index_precedes_bounds; 4715 if (!ASE) { 4716 DiagID = diag::warn_ptr_arith_precedes_bounds; 4717 if (index.isNegative()) index = -index; 4718 } 4719 4720 DiagRuntimeBehavior(BaseExpr->getLocStart(), BaseExpr, 4721 PDiag(DiagID) << index.toString(10, true) 4722 << IndexExpr->getSourceRange()); 4723 } 4724 4725 if (!ND) { 4726 // Try harder to find a NamedDecl to point at in the note. 4727 while (const ArraySubscriptExpr *ASE = 4728 dyn_cast<ArraySubscriptExpr>(BaseExpr)) 4729 BaseExpr = ASE->getBase()->IgnoreParenCasts(); 4730 if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(BaseExpr)) 4731 ND = dyn_cast<NamedDecl>(DRE->getDecl()); 4732 if (const MemberExpr *ME = dyn_cast<MemberExpr>(BaseExpr)) 4733 ND = dyn_cast<NamedDecl>(ME->getMemberDecl()); 4734 } 4735 4736 if (ND) 4737 DiagRuntimeBehavior(ND->getLocStart(), BaseExpr, 4738 PDiag(diag::note_array_index_out_of_bounds) 4739 << ND->getDeclName()); 4740} 4741 4742void Sema::CheckArrayAccess(const Expr *expr) { 4743 int AllowOnePastEnd = 0; 4744 while (expr) { 4745 expr = expr->IgnoreParenImpCasts(); 4746 switch (expr->getStmtClass()) { 4747 case Stmt::ArraySubscriptExprClass: { 4748 const ArraySubscriptExpr *ASE = cast<ArraySubscriptExpr>(expr); 4749 CheckArrayAccess(ASE->getBase(), ASE->getIdx(), ASE, 4750 AllowOnePastEnd > 0); 4751 return; 4752 } 4753 case Stmt::UnaryOperatorClass: { 4754 // Only unwrap the * and & unary operators 4755 const UnaryOperator *UO = cast<UnaryOperator>(expr); 4756 expr = UO->getSubExpr(); 4757 switch (UO->getOpcode()) { 4758 case UO_AddrOf: 4759 AllowOnePastEnd++; 4760 break; 4761 case UO_Deref: 4762 AllowOnePastEnd--; 4763 break; 4764 default: 4765 return; 4766 } 4767 break; 4768 } 4769 case Stmt::ConditionalOperatorClass: { 4770 const ConditionalOperator *cond = cast<ConditionalOperator>(expr); 4771 if (const Expr *lhs = cond->getLHS()) 4772 CheckArrayAccess(lhs); 4773 if (const Expr *rhs = cond->getRHS()) 4774 CheckArrayAccess(rhs); 4775 return; 4776 } 4777 default: 4778 return; 4779 } 4780 } 4781} 4782 4783//===--- CHECK: Objective-C retain cycles ----------------------------------// 4784 4785namespace { 4786 struct RetainCycleOwner { 4787 RetainCycleOwner() : Variable(0), Indirect(false) {} 4788 VarDecl *Variable; 4789 SourceRange Range; 4790 SourceLocation Loc; 4791 bool Indirect; 4792 4793 void setLocsFrom(Expr *e) { 4794 Loc = e->getExprLoc(); 4795 Range = e->getSourceRange(); 4796 } 4797 }; 4798} 4799 4800/// Consider whether capturing the given variable can possibly lead to 4801/// a retain cycle. 4802static bool considerVariable(VarDecl *var, Expr *ref, RetainCycleOwner &owner) { 4803 // In ARC, it's captured strongly iff the variable has __strong 4804 // lifetime. In MRR, it's captured strongly if the variable is 4805 // __block and has an appropriate type. 4806 if (var->getType().getObjCLifetime() != Qualifiers::OCL_Strong) 4807 return false; 4808 4809 owner.Variable = var; 4810 owner.setLocsFrom(ref); 4811 return true; 4812} 4813 4814static bool findRetainCycleOwner(Sema &S, Expr *e, RetainCycleOwner &owner) { 4815 while (true) { 4816 e = e->IgnoreParens(); 4817 if (CastExpr *cast = dyn_cast<CastExpr>(e)) { 4818 switch (cast->getCastKind()) { 4819 case CK_BitCast: 4820 case CK_LValueBitCast: 4821 case CK_LValueToRValue: 4822 case CK_ARCReclaimReturnedObject: 4823 e = cast->getSubExpr(); 4824 continue; 4825 4826 default: 4827 return false; 4828 } 4829 } 4830 4831 if (ObjCIvarRefExpr *ref = dyn_cast<ObjCIvarRefExpr>(e)) { 4832 ObjCIvarDecl *ivar = ref->getDecl(); 4833 if (ivar->getType().getObjCLifetime() != Qualifiers::OCL_Strong) 4834 return false; 4835 4836 // Try to find a retain cycle in the base. 4837 if (!findRetainCycleOwner(S, ref->getBase(), owner)) 4838 return false; 4839 4840 if (ref->isFreeIvar()) owner.setLocsFrom(ref); 4841 owner.Indirect = true; 4842 return true; 4843 } 4844 4845 if (DeclRefExpr *ref = dyn_cast<DeclRefExpr>(e)) { 4846 VarDecl *var = dyn_cast<VarDecl>(ref->getDecl()); 4847 if (!var) return false; 4848 return considerVariable(var, ref, owner); 4849 } 4850 4851 if (MemberExpr *member = dyn_cast<MemberExpr>(e)) { 4852 if (member->isArrow()) return false; 4853 4854 // Don't count this as an indirect ownership. 4855 e = member->getBase(); 4856 continue; 4857 } 4858 4859 if (PseudoObjectExpr *pseudo = dyn_cast<PseudoObjectExpr>(e)) { 4860 // Only pay attention to pseudo-objects on property references. 4861 ObjCPropertyRefExpr *pre 4862 = dyn_cast<ObjCPropertyRefExpr>(pseudo->getSyntacticForm() 4863 ->IgnoreParens()); 4864 if (!pre) return false; 4865 if (pre->isImplicitProperty()) return false; 4866 ObjCPropertyDecl *property = pre->getExplicitProperty(); 4867 if (!property->isRetaining() && 4868 !(property->getPropertyIvarDecl() && 4869 property->getPropertyIvarDecl()->getType() 4870 .getObjCLifetime() == Qualifiers::OCL_Strong)) 4871 return false; 4872 4873 owner.Indirect = true; 4874 if (pre->isSuperReceiver()) { 4875 owner.Variable = S.getCurMethodDecl()->getSelfDecl(); 4876 if (!owner.Variable) 4877 return false; 4878 owner.Loc = pre->getLocation(); 4879 owner.Range = pre->getSourceRange(); 4880 return true; 4881 } 4882 e = const_cast<Expr*>(cast<OpaqueValueExpr>(pre->getBase()) 4883 ->getSourceExpr()); 4884 continue; 4885 } 4886 4887 // Array ivars? 4888 4889 return false; 4890 } 4891} 4892 4893namespace { 4894 struct FindCaptureVisitor : EvaluatedExprVisitor<FindCaptureVisitor> { 4895 FindCaptureVisitor(ASTContext &Context, VarDecl *variable) 4896 : EvaluatedExprVisitor<FindCaptureVisitor>(Context), 4897 Variable(variable), Capturer(0) {} 4898 4899 VarDecl *Variable; 4900 Expr *Capturer; 4901 4902 void VisitDeclRefExpr(DeclRefExpr *ref) { 4903 if (ref->getDecl() == Variable && !Capturer) 4904 Capturer = ref; 4905 } 4906 4907 void VisitObjCIvarRefExpr(ObjCIvarRefExpr *ref) { 4908 if (Capturer) return; 4909 Visit(ref->getBase()); 4910 if (Capturer && ref->isFreeIvar()) 4911 Capturer = ref; 4912 } 4913 4914 void VisitBlockExpr(BlockExpr *block) { 4915 // Look inside nested blocks 4916 if (block->getBlockDecl()->capturesVariable(Variable)) 4917 Visit(block->getBlockDecl()->getBody()); 4918 } 4919 }; 4920} 4921 4922/// Check whether the given argument is a block which captures a 4923/// variable. 4924static Expr *findCapturingExpr(Sema &S, Expr *e, RetainCycleOwner &owner) { 4925 assert(owner.Variable && owner.Loc.isValid()); 4926 4927 e = e->IgnoreParenCasts(); 4928 BlockExpr *block = dyn_cast<BlockExpr>(e); 4929 if (!block || !block->getBlockDecl()->capturesVariable(owner.Variable)) 4930 return 0; 4931 4932 FindCaptureVisitor visitor(S.Context, owner.Variable); 4933 visitor.Visit(block->getBlockDecl()->getBody()); 4934 return visitor.Capturer; 4935} 4936 4937static void diagnoseRetainCycle(Sema &S, Expr *capturer, 4938 RetainCycleOwner &owner) { 4939 assert(capturer); 4940 assert(owner.Variable && owner.Loc.isValid()); 4941 4942 S.Diag(capturer->getExprLoc(), diag::warn_arc_retain_cycle) 4943 << owner.Variable << capturer->getSourceRange(); 4944 S.Diag(owner.Loc, diag::note_arc_retain_cycle_owner) 4945 << owner.Indirect << owner.Range; 4946} 4947 4948/// Check for a keyword selector that starts with the word 'add' or 4949/// 'set'. 4950static bool isSetterLikeSelector(Selector sel) { 4951 if (sel.isUnarySelector()) return false; 4952 4953 StringRef str = sel.getNameForSlot(0); 4954 while (!str.empty() && str.front() == '_') str = str.substr(1); 4955 if (str.startswith("set")) 4956 str = str.substr(3); 4957 else if (str.startswith("add")) { 4958 // Specially whitelist 'addOperationWithBlock:'. 4959 if (sel.getNumArgs() == 1 && str.startswith("addOperationWithBlock")) 4960 return false; 4961 str = str.substr(3); 4962 } 4963 else 4964 return false; 4965 4966 if (str.empty()) return true; 4967 return !islower(str.front()); 4968} 4969 4970/// Check a message send to see if it's likely to cause a retain cycle. 4971void Sema::checkRetainCycles(ObjCMessageExpr *msg) { 4972 // Only check instance methods whose selector looks like a setter. 4973 if (!msg->isInstanceMessage() || !isSetterLikeSelector(msg->getSelector())) 4974 return; 4975 4976 // Try to find a variable that the receiver is strongly owned by. 4977 RetainCycleOwner owner; 4978 if (msg->getReceiverKind() == ObjCMessageExpr::Instance) { 4979 if (!findRetainCycleOwner(*this, msg->getInstanceReceiver(), owner)) 4980 return; 4981 } else { 4982 assert(msg->getReceiverKind() == ObjCMessageExpr::SuperInstance); 4983 owner.Variable = getCurMethodDecl()->getSelfDecl(); 4984 owner.Loc = msg->getSuperLoc(); 4985 owner.Range = msg->getSuperLoc(); 4986 } 4987 4988 // Check whether the receiver is captured by any of the arguments. 4989 for (unsigned i = 0, e = msg->getNumArgs(); i != e; ++i) 4990 if (Expr *capturer = findCapturingExpr(*this, msg->getArg(i), owner)) 4991 return diagnoseRetainCycle(*this, capturer, owner); 4992} 4993 4994/// Check a property assign to see if it's likely to cause a retain cycle. 4995void Sema::checkRetainCycles(Expr *receiver, Expr *argument) { 4996 RetainCycleOwner owner; 4997 if (!findRetainCycleOwner(*this, receiver, owner)) 4998 return; 4999 5000 if (Expr *capturer = findCapturingExpr(*this, argument, owner)) 5001 diagnoseRetainCycle(*this, capturer, owner); 5002} 5003 5004bool Sema::checkUnsafeAssigns(SourceLocation Loc, 5005 QualType LHS, Expr *RHS) { 5006 Qualifiers::ObjCLifetime LT = LHS.getObjCLifetime(); 5007 if (LT != Qualifiers::OCL_Weak && LT != Qualifiers::OCL_ExplicitNone) 5008 return false; 5009 // strip off any implicit cast added to get to the one arc-specific 5010 while (ImplicitCastExpr *cast = dyn_cast<ImplicitCastExpr>(RHS)) { 5011 if (cast->getCastKind() == CK_ARCConsumeObject) { 5012 Diag(Loc, diag::warn_arc_retained_assign) 5013 << (LT == Qualifiers::OCL_ExplicitNone) 5014 << RHS->getSourceRange(); 5015 return true; 5016 } 5017 RHS = cast->getSubExpr(); 5018 } 5019 return false; 5020} 5021 5022void Sema::checkUnsafeExprAssigns(SourceLocation Loc, 5023 Expr *LHS, Expr *RHS) { 5024 QualType LHSType; 5025 // PropertyRef on LHS type need be directly obtained from 5026 // its declaration as it has a PsuedoType. 5027 ObjCPropertyRefExpr *PRE 5028 = dyn_cast<ObjCPropertyRefExpr>(LHS->IgnoreParens()); 5029 if (PRE && !PRE->isImplicitProperty()) { 5030 const ObjCPropertyDecl *PD = PRE->getExplicitProperty(); 5031 if (PD) 5032 LHSType = PD->getType(); 5033 } 5034 5035 if (LHSType.isNull()) 5036 LHSType = LHS->getType(); 5037 if (checkUnsafeAssigns(Loc, LHSType, RHS)) 5038 return; 5039 Qualifiers::ObjCLifetime LT = LHSType.getObjCLifetime(); 5040 // FIXME. Check for other life times. 5041 if (LT != Qualifiers::OCL_None) 5042 return; 5043 5044 if (PRE) { 5045 if (PRE->isImplicitProperty()) 5046 return; 5047 const ObjCPropertyDecl *PD = PRE->getExplicitProperty(); 5048 if (!PD) 5049 return; 5050 5051 unsigned Attributes = PD->getPropertyAttributes(); 5052 if (Attributes & ObjCPropertyDecl::OBJC_PR_assign) { 5053 // when 'assign' attribute was not explicitly specified 5054 // by user, ignore it and rely on property type itself 5055 // for lifetime info. 5056 unsigned AsWrittenAttr = PD->getPropertyAttributesAsWritten(); 5057 if (!(AsWrittenAttr & ObjCPropertyDecl::OBJC_PR_assign) && 5058 LHSType->isObjCRetainableType()) 5059 return; 5060 5061 while (ImplicitCastExpr *cast = dyn_cast<ImplicitCastExpr>(RHS)) { 5062 if (cast->getCastKind() == CK_ARCConsumeObject) { 5063 Diag(Loc, diag::warn_arc_retained_property_assign) 5064 << RHS->getSourceRange(); 5065 return; 5066 } 5067 RHS = cast->getSubExpr(); 5068 } 5069 } 5070 } 5071} 5072 5073//===--- CHECK: Empty statement body (-Wempty-body) ---------------------===// 5074 5075namespace { 5076bool ShouldDiagnoseEmptyStmtBody(const SourceManager &SourceMgr, 5077 SourceLocation StmtLoc, 5078 const NullStmt *Body) { 5079 // Do not warn if the body is a macro that expands to nothing, e.g: 5080 // 5081 // #define CALL(x) 5082 // if (condition) 5083 // CALL(0); 5084 // 5085 if (Body->hasLeadingEmptyMacro()) 5086 return false; 5087 5088 // Get line numbers of statement and body. 5089 bool StmtLineInvalid; 5090 unsigned StmtLine = SourceMgr.getSpellingLineNumber(StmtLoc, 5091 &StmtLineInvalid); 5092 if (StmtLineInvalid) 5093 return false; 5094 5095 bool BodyLineInvalid; 5096 unsigned BodyLine = SourceMgr.getSpellingLineNumber(Body->getSemiLoc(), 5097 &BodyLineInvalid); 5098 if (BodyLineInvalid) 5099 return false; 5100 5101 // Warn if null statement and body are on the same line. 5102 if (StmtLine != BodyLine) 5103 return false; 5104 5105 return true; 5106} 5107} // Unnamed namespace 5108 5109void Sema::DiagnoseEmptyStmtBody(SourceLocation StmtLoc, 5110 const Stmt *Body, 5111 unsigned DiagID) { 5112 // Since this is a syntactic check, don't emit diagnostic for template 5113 // instantiations, this just adds noise. 5114 if (CurrentInstantiationScope) 5115 return; 5116 5117 // The body should be a null statement. 5118 const NullStmt *NBody = dyn_cast<NullStmt>(Body); 5119 if (!NBody) 5120 return; 5121 5122 // Do the usual checks. 5123 if (!ShouldDiagnoseEmptyStmtBody(SourceMgr, StmtLoc, NBody)) 5124 return; 5125 5126 Diag(NBody->getSemiLoc(), DiagID); 5127 Diag(NBody->getSemiLoc(), diag::note_empty_body_on_separate_line); 5128} 5129 5130void Sema::DiagnoseEmptyLoopBody(const Stmt *S, 5131 const Stmt *PossibleBody) { 5132 assert(!CurrentInstantiationScope); // Ensured by caller 5133 5134 SourceLocation StmtLoc; 5135 const Stmt *Body; 5136 unsigned DiagID; 5137 if (const ForStmt *FS = dyn_cast<ForStmt>(S)) { 5138 StmtLoc = FS->getRParenLoc(); 5139 Body = FS->getBody(); 5140 DiagID = diag::warn_empty_for_body; 5141 } else if (const WhileStmt *WS = dyn_cast<WhileStmt>(S)) { 5142 StmtLoc = WS->getCond()->getSourceRange().getEnd(); 5143 Body = WS->getBody(); 5144 DiagID = diag::warn_empty_while_body; 5145 } else 5146 return; // Neither `for' nor `while'. 5147 5148 // The body should be a null statement. 5149 const NullStmt *NBody = dyn_cast<NullStmt>(Body); 5150 if (!NBody) 5151 return; 5152 5153 // Skip expensive checks if diagnostic is disabled. 5154 if (Diags.getDiagnosticLevel(DiagID, NBody->getSemiLoc()) == 5155 DiagnosticsEngine::Ignored) 5156 return; 5157 5158 // Do the usual checks. 5159 if (!ShouldDiagnoseEmptyStmtBody(SourceMgr, StmtLoc, NBody)) 5160 return; 5161 5162 // `for(...);' and `while(...);' are popular idioms, so in order to keep 5163 // noise level low, emit diagnostics only if for/while is followed by a 5164 // CompoundStmt, e.g.: 5165 // for (int i = 0; i < n; i++); 5166 // { 5167 // a(i); 5168 // } 5169 // or if for/while is followed by a statement with more indentation 5170 // than for/while itself: 5171 // for (int i = 0; i < n; i++); 5172 // a(i); 5173 bool ProbableTypo = isa<CompoundStmt>(PossibleBody); 5174 if (!ProbableTypo) { 5175 bool BodyColInvalid; 5176 unsigned BodyCol = SourceMgr.getPresumedColumnNumber( 5177 PossibleBody->getLocStart(), 5178 &BodyColInvalid); 5179 if (BodyColInvalid) 5180 return; 5181 5182 bool StmtColInvalid; 5183 unsigned StmtCol = SourceMgr.getPresumedColumnNumber( 5184 S->getLocStart(), 5185 &StmtColInvalid); 5186 if (StmtColInvalid) 5187 return; 5188 5189 if (BodyCol > StmtCol) 5190 ProbableTypo = true; 5191 } 5192 5193 if (ProbableTypo) { 5194 Diag(NBody->getSemiLoc(), DiagID); 5195 Diag(NBody->getSemiLoc(), diag::note_empty_body_on_separate_line); 5196 } 5197} 5198