1 //===--- IdentifierTable.cpp - Hash table for identifier lookup -----------===// 2 // 3 // The LLVM Compiler Infrastructure 4 // 5 // This file is distributed under the University of Illinois Open Source 6 // License. See LICENSE.TXT for details. 7 // 8 //===----------------------------------------------------------------------===// 9 // 10 // This file implements the IdentifierInfo, IdentifierVisitor, and 11 // IdentifierTable interfaces. 12 // 13 //===----------------------------------------------------------------------===// 14 15 #include "clang/Basic/IdentifierTable.h" 16 #include "clang/Basic/LangOptions.h" 17 #include "llvm/ADT/FoldingSet.h" 18 #include "llvm/ADT/DenseMap.h" 19 #include "llvm/ADT/StringRef.h" 20 #include "llvm/Support/raw_ostream.h" 21 #include <cstdio> 22 23 using namespace clang; 24 25 //===----------------------------------------------------------------------===// 26 // IdentifierInfo Implementation 27 //===----------------------------------------------------------------------===// 28 29 IdentifierInfo::IdentifierInfo() { 30 TokenID = tok::identifier; 31 ObjCOrBuiltinID = 0; 32 HasMacro = false; 33 IsExtension = false; 34 IsPoisoned = false; 35 IsCPPOperatorKeyword = false; 36 NeedsHandleIdentifier = false; 37 FETokenInfo = 0; 38 Entry = 0; 39 } 40 41 //===----------------------------------------------------------------------===// 42 // IdentifierTable Implementation 43 //===----------------------------------------------------------------------===// 44 45 IdentifierInfoLookup::~IdentifierInfoLookup() {} 46 47 ExternalIdentifierLookup::~ExternalIdentifierLookup() {} 48 49 IdentifierTable::IdentifierTable(const LangOptions &LangOpts, 50 IdentifierInfoLookup* externalLookup) 51 : HashTable(8192), // Start with space for 8K identifiers. 52 ExternalLookup(externalLookup) { 53 54 // Populate the identifier table with info about keywords for the current 55 // language. 56 AddKeywords(LangOpts); 57 } 58 59 //===----------------------------------------------------------------------===// 60 // Language Keyword Implementation 61 //===----------------------------------------------------------------------===// 62 63 // Constants for TokenKinds.def 64 namespace { 65 enum { 66 KEYALL = 1, 67 KEYC99 = 2, 68 KEYCXX = 4, 69 KEYCXX0X = 8, 70 KEYGNU = 16, 71 KEYMS = 32, 72 BOOLSUPPORT = 64, 73 KEYALTIVEC = 128, 74 KEYNOMS = 256 75 }; 76 } 77 78 /// AddKeyword - This method is used to associate a token ID with specific 79 /// identifiers because they are language keywords. This causes the lexer to 80 /// automatically map matching identifiers to specialized token codes. 81 /// 82 /// The C90/C99/CPP/CPP0x flags are set to 2 if the token should be 83 /// enabled in the specified langauge, set to 1 if it is an extension 84 /// in the specified language, and set to 0 if disabled in the 85 /// specified language. 86 static void AddKeyword(llvm::StringRef Keyword, 87 tok::TokenKind TokenCode, unsigned Flags, 88 const LangOptions &LangOpts, IdentifierTable &Table) { 89 unsigned AddResult = 0; 90 if (Flags & KEYALL) AddResult = 2; 91 else if (LangOpts.CPlusPlus && (Flags & KEYCXX)) AddResult = 2; 92 else if (LangOpts.CPlusPlus0x && (Flags & KEYCXX0X)) AddResult = 2; 93 else if (LangOpts.C99 && (Flags & KEYC99)) AddResult = 2; 94 else if (LangOpts.GNUKeywords && (Flags & KEYGNU)) AddResult = 1; 95 else if (LangOpts.Microsoft && (Flags & KEYMS)) AddResult = 1; 96 else if (LangOpts.Bool && (Flags & BOOLSUPPORT)) AddResult = 2; 97 else if (LangOpts.AltiVec && (Flags & KEYALTIVEC)) AddResult = 2; 98 else if (!LangOpts.Microsoft && (Flags & KEYNOMS)) AddResult = 2; 99 100 // Don't add this keyword if disabled in this language. 101 if (AddResult == 0) return; 102 103 IdentifierInfo &Info = Table.get(Keyword); 104 Info.setTokenID(TokenCode); 105 Info.setIsExtensionToken(AddResult == 1); 106 } 107 108 /// AddCXXOperatorKeyword - Register a C++ operator keyword alternative 109 /// representations. 110 static void AddCXXOperatorKeyword(llvm::StringRef Keyword, 111 tok::TokenKind TokenCode, 112 IdentifierTable &Table) { 113 IdentifierInfo &Info = Table.get(Keyword); 114 Info.setTokenID(TokenCode); 115 Info.setIsCPlusPlusOperatorKeyword(); 116 } 117 118 /// AddObjCKeyword - Register an Objective-C @keyword like "class" "selector" or 119 /// "property". 120 static void AddObjCKeyword(llvm::StringRef Name, 121 tok::ObjCKeywordKind ObjCID, 122 IdentifierTable &Table) { 123 Table.get(Name).setObjCKeywordID(ObjCID); 124 } 125 126 /// AddKeywords - Add all keywords to the symbol table. 127 /// 128 void IdentifierTable::AddKeywords(const LangOptions &LangOpts) { 129 // Add keywords and tokens for the current language. 130 #define KEYWORD(NAME, FLAGS) \ 131 AddKeyword(llvm::StringRef(#NAME), tok::kw_ ## NAME, \ 132 FLAGS, LangOpts, *this); 133 #define ALIAS(NAME, TOK, FLAGS) \ 134 AddKeyword(llvm::StringRef(NAME), tok::kw_ ## TOK, \ 135 FLAGS, LangOpts, *this); 136 #define CXX_KEYWORD_OPERATOR(NAME, ALIAS) \ 137 if (LangOpts.CXXOperatorNames) \ 138 AddCXXOperatorKeyword(llvm::StringRef(#NAME), tok::ALIAS, *this); 139 #define OBJC1_AT_KEYWORD(NAME) \ 140 if (LangOpts.ObjC1) \ 141 AddObjCKeyword(llvm::StringRef(#NAME), tok::objc_##NAME, *this); 142 #define OBJC2_AT_KEYWORD(NAME) \ 143 if (LangOpts.ObjC2) \ 144 AddObjCKeyword(llvm::StringRef(#NAME), tok::objc_##NAME, *this); 145 #include "clang/Basic/TokenKinds.def" 146 } 147 148 tok::PPKeywordKind IdentifierInfo::getPPKeywordID() const { 149 // We use a perfect hash function here involving the length of the keyword, 150 // the first and third character. For preprocessor ID's there are no 151 // collisions (if there were, the switch below would complain about duplicate 152 // case values). Note that this depends on 'if' being null terminated. 153 154 #define HASH(LEN, FIRST, THIRD) \ 155 (LEN << 5) + (((FIRST-'a') + (THIRD-'a')) & 31) 156 #define CASE(LEN, FIRST, THIRD, NAME) \ 157 case HASH(LEN, FIRST, THIRD): \ 158 return memcmp(Name, #NAME, LEN) ? tok::pp_not_keyword : tok::pp_ ## NAME 159 160 unsigned Len = getLength(); 161 if (Len < 2) return tok::pp_not_keyword; 162 const char *Name = getNameStart(); 163 switch (HASH(Len, Name[0], Name[2])) { 164 default: return tok::pp_not_keyword; 165 CASE( 2, 'i', '\0', if); 166 CASE( 4, 'e', 'i', elif); 167 CASE( 4, 'e', 's', else); 168 CASE( 4, 'l', 'n', line); 169 CASE( 4, 's', 'c', sccs); 170 CASE( 5, 'e', 'd', endif); 171 CASE( 5, 'e', 'r', error); 172 CASE( 5, 'i', 'e', ident); 173 CASE( 5, 'i', 'd', ifdef); 174 CASE( 5, 'u', 'd', undef); 175 176 CASE( 6, 'a', 's', assert); 177 CASE( 6, 'd', 'f', define); 178 CASE( 6, 'i', 'n', ifndef); 179 CASE( 6, 'i', 'p', import); 180 CASE( 6, 'p', 'a', pragma); 181 182 CASE( 7, 'd', 'f', defined); 183 CASE( 7, 'i', 'c', include); 184 CASE( 7, 'w', 'r', warning); 185 186 CASE( 8, 'u', 'a', unassert); 187 CASE(12, 'i', 'c', include_next); 188 189 CASE(16, '_', 'i', __include_macros); 190 #undef CASE 191 #undef HASH 192 } 193 } 194 195 //===----------------------------------------------------------------------===// 196 // Stats Implementation 197 //===----------------------------------------------------------------------===// 198 199 /// PrintStats - Print statistics about how well the identifier table is doing 200 /// at hashing identifiers. 201 void IdentifierTable::PrintStats() const { 202 unsigned NumBuckets = HashTable.getNumBuckets(); 203 unsigned NumIdentifiers = HashTable.getNumItems(); 204 unsigned NumEmptyBuckets = NumBuckets-NumIdentifiers; 205 unsigned AverageIdentifierSize = 0; 206 unsigned MaxIdentifierLength = 0; 207 208 // TODO: Figure out maximum times an identifier had to probe for -stats. 209 for (llvm::StringMap<IdentifierInfo*, llvm::BumpPtrAllocator>::const_iterator 210 I = HashTable.begin(), E = HashTable.end(); I != E; ++I) { 211 unsigned IdLen = I->getKeyLength(); 212 AverageIdentifierSize += IdLen; 213 if (MaxIdentifierLength < IdLen) 214 MaxIdentifierLength = IdLen; 215 } 216 217 fprintf(stderr, "\n*** Identifier Table Stats:\n"); 218 fprintf(stderr, "# Identifiers: %d\n", NumIdentifiers); 219 fprintf(stderr, "# Empty Buckets: %d\n", NumEmptyBuckets); 220 fprintf(stderr, "Hash density (#identifiers per bucket): %f\n", 221 NumIdentifiers/(double)NumBuckets); 222 fprintf(stderr, "Ave identifier length: %f\n", 223 (AverageIdentifierSize/(double)NumIdentifiers)); 224 fprintf(stderr, "Max identifier length: %d\n", MaxIdentifierLength); 225 226 // Compute statistics about the memory allocated for identifiers. 227 HashTable.getAllocator().PrintStats(); 228 } 229 230 //===----------------------------------------------------------------------===// 231 // SelectorTable Implementation 232 //===----------------------------------------------------------------------===// 233 234 unsigned llvm::DenseMapInfo<clang::Selector>::getHashValue(clang::Selector S) { 235 return DenseMapInfo<void*>::getHashValue(S.getAsOpaquePtr()); 236 } 237 238 namespace clang { 239 /// MultiKeywordSelector - One of these variable length records is kept for each 240 /// selector containing more than one keyword. We use a folding set 241 /// to unique aggregate names (keyword selectors in ObjC parlance). Access to 242 /// this class is provided strictly through Selector. 243 class MultiKeywordSelector 244 : public DeclarationNameExtra, public llvm::FoldingSetNode { 245 MultiKeywordSelector(unsigned nKeys) { 246 ExtraKindOrNumArgs = NUM_EXTRA_KINDS + nKeys; 247 } 248 public: 249 // Constructor for keyword selectors. 250 MultiKeywordSelector(unsigned nKeys, IdentifierInfo **IIV) { 251 assert((nKeys > 1) && "not a multi-keyword selector"); 252 ExtraKindOrNumArgs = NUM_EXTRA_KINDS + nKeys; 253 254 // Fill in the trailing keyword array. 255 IdentifierInfo **KeyInfo = reinterpret_cast<IdentifierInfo **>(this+1); 256 for (unsigned i = 0; i != nKeys; ++i) 257 KeyInfo[i] = IIV[i]; 258 } 259 260 // getName - Derive the full selector name and return it. 261 std::string getName() const; 262 263 unsigned getNumArgs() const { return ExtraKindOrNumArgs - NUM_EXTRA_KINDS; } 264 265 typedef IdentifierInfo *const *keyword_iterator; 266 keyword_iterator keyword_begin() const { 267 return reinterpret_cast<keyword_iterator>(this+1); 268 } 269 keyword_iterator keyword_end() const { 270 return keyword_begin()+getNumArgs(); 271 } 272 IdentifierInfo *getIdentifierInfoForSlot(unsigned i) const { 273 assert(i < getNumArgs() && "getIdentifierInfoForSlot(): illegal index"); 274 return keyword_begin()[i]; 275 } 276 static void Profile(llvm::FoldingSetNodeID &ID, 277 keyword_iterator ArgTys, unsigned NumArgs) { 278 ID.AddInteger(NumArgs); 279 for (unsigned i = 0; i != NumArgs; ++i) 280 ID.AddPointer(ArgTys[i]); 281 } 282 void Profile(llvm::FoldingSetNodeID &ID) { 283 Profile(ID, keyword_begin(), getNumArgs()); 284 } 285 }; 286 } // end namespace clang. 287 288 unsigned Selector::getNumArgs() const { 289 unsigned IIF = getIdentifierInfoFlag(); 290 if (IIF == ZeroArg) 291 return 0; 292 if (IIF == OneArg) 293 return 1; 294 // We point to a MultiKeywordSelector (pointer doesn't contain any flags). 295 MultiKeywordSelector *SI = reinterpret_cast<MultiKeywordSelector *>(InfoPtr); 296 return SI->getNumArgs(); 297 } 298 299 IdentifierInfo *Selector::getIdentifierInfoForSlot(unsigned argIndex) const { 300 if (getIdentifierInfoFlag()) { 301 assert(argIndex == 0 && "illegal keyword index"); 302 return getAsIdentifierInfo(); 303 } 304 // We point to a MultiKeywordSelector (pointer doesn't contain any flags). 305 MultiKeywordSelector *SI = reinterpret_cast<MultiKeywordSelector *>(InfoPtr); 306 return SI->getIdentifierInfoForSlot(argIndex); 307 } 308 309 std::string MultiKeywordSelector::getName() const { 310 llvm::SmallString<256> Str; 311 llvm::raw_svector_ostream OS(Str); 312 for (keyword_iterator I = keyword_begin(), E = keyword_end(); I != E; ++I) { 313 if (*I) 314 OS << (*I)->getName(); 315 OS << ':'; 316 } 317 318 return OS.str(); 319 } 320 321 std::string Selector::getAsString() const { 322 if (InfoPtr == 0) 323 return "<null selector>"; 324 325 if (InfoPtr & ArgFlags) { 326 IdentifierInfo *II = getAsIdentifierInfo(); 327 328 // If the number of arguments is 0 then II is guaranteed to not be null. 329 if (getNumArgs() == 0) 330 return II->getName(); 331 332 if (!II) 333 return ":"; 334 335 return II->getName().str() + ":"; 336 } 337 338 // We have a multiple keyword selector (no embedded flags). 339 return reinterpret_cast<MultiKeywordSelector *>(InfoPtr)->getName(); 340 } 341 342 343 namespace { 344 struct SelectorTableImpl { 345 llvm::FoldingSet<MultiKeywordSelector> Table; 346 llvm::BumpPtrAllocator Allocator; 347 }; 348 } // end anonymous namespace. 349 350 static SelectorTableImpl &getSelectorTableImpl(void *P) { 351 return *static_cast<SelectorTableImpl*>(P); 352 } 353 354 355 Selector SelectorTable::getSelector(unsigned nKeys, IdentifierInfo **IIV) { 356 if (nKeys < 2) 357 return Selector(IIV[0], nKeys); 358 359 SelectorTableImpl &SelTabImpl = getSelectorTableImpl(Impl); 360 361 // Unique selector, to guarantee there is one per name. 362 llvm::FoldingSetNodeID ID; 363 MultiKeywordSelector::Profile(ID, IIV, nKeys); 364 365 void *InsertPos = 0; 366 if (MultiKeywordSelector *SI = 367 SelTabImpl.Table.FindNodeOrInsertPos(ID, InsertPos)) 368 return Selector(SI); 369 370 // MultiKeywordSelector objects are not allocated with new because they have a 371 // variable size array (for parameter types) at the end of them. 372 unsigned Size = sizeof(MultiKeywordSelector) + nKeys*sizeof(IdentifierInfo *); 373 MultiKeywordSelector *SI = 374 (MultiKeywordSelector*)SelTabImpl.Allocator.Allocate(Size, 375 llvm::alignof<MultiKeywordSelector>()); 376 new (SI) MultiKeywordSelector(nKeys, IIV); 377 SelTabImpl.Table.InsertNode(SI, InsertPos); 378 return Selector(SI); 379 } 380 381 SelectorTable::SelectorTable() { 382 Impl = new SelectorTableImpl(); 383 } 384 385 SelectorTable::~SelectorTable() { 386 delete &getSelectorTableImpl(Impl); 387 } 388 389 const char *clang::getOperatorSpelling(OverloadedOperatorKind Operator) { 390 switch (Operator) { 391 case OO_None: 392 case NUM_OVERLOADED_OPERATORS: 393 return 0; 394 395 #define OVERLOADED_OPERATOR(Name,Spelling,Token,Unary,Binary,MemberOnly) \ 396 case OO_##Name: return Spelling; 397 #include "clang/Basic/OperatorKinds.def" 398 } 399 400 return 0; 401 } 402 403