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/ADT/StringSwitch.h" 21 #include "llvm/Support/raw_ostream.h" 22 #include <cstdio> 23 24 using namespace clang; 25 26 //===----------------------------------------------------------------------===// 27 // IdentifierInfo Implementation 28 //===----------------------------------------------------------------------===// 29 30 IdentifierInfo::IdentifierInfo() { 31 TokenID = tok::identifier; 32 ObjCOrBuiltinID = 0; 33 HasMacro = false; 34 IsExtension = false; 35 IsCXX11CompatKeyword = false; 36 IsPoisoned = false; 37 IsCPPOperatorKeyword = false; 38 NeedsHandleIdentifier = false; 39 IsFromAST = false; 40 RevertedTokenID = false; 41 FETokenInfo = 0; 42 Entry = 0; 43 } 44 45 //===----------------------------------------------------------------------===// 46 // IdentifierTable Implementation 47 //===----------------------------------------------------------------------===// 48 49 IdentifierIterator::~IdentifierIterator() { } 50 51 IdentifierInfoLookup::~IdentifierInfoLookup() {} 52 53 namespace { 54 /// \brief A simple identifier lookup iterator that represents an 55 /// empty sequence of identifiers. 56 class EmptyLookupIterator : public IdentifierIterator 57 { 58 public: 59 virtual StringRef Next() { return StringRef(); } 60 }; 61 } 62 63 IdentifierIterator *IdentifierInfoLookup::getIdentifiers() const { 64 return new EmptyLookupIterator(); 65 } 66 67 ExternalIdentifierLookup::~ExternalIdentifierLookup() {} 68 69 IdentifierTable::IdentifierTable(const LangOptions &LangOpts, 70 IdentifierInfoLookup* externalLookup) 71 : HashTable(8192), // Start with space for 8K identifiers. 72 ExternalLookup(externalLookup) { 73 74 // Populate the identifier table with info about keywords for the current 75 // language. 76 AddKeywords(LangOpts); 77 } 78 79 //===----------------------------------------------------------------------===// 80 // Language Keyword Implementation 81 //===----------------------------------------------------------------------===// 82 83 // Constants for TokenKinds.def 84 namespace { 85 enum { 86 KEYC99 = 0x1, 87 KEYCXX = 0x2, 88 KEYCXX0X = 0x4, 89 KEYGNU = 0x8, 90 KEYMS = 0x10, 91 BOOLSUPPORT = 0x20, 92 KEYALTIVEC = 0x40, 93 KEYNOCXX = 0x80, 94 KEYBORLAND = 0x100, 95 KEYOPENCL = 0x200, 96 KEYC1X = 0x400, 97 KEYARC = 0x800, 98 KEYALL = 0x0fff 99 }; 100 } 101 102 /// AddKeyword - This method is used to associate a token ID with specific 103 /// identifiers because they are language keywords. This causes the lexer to 104 /// automatically map matching identifiers to specialized token codes. 105 /// 106 /// The C90/C99/CPP/CPP0x flags are set to 3 if the token is a keyword in a 107 /// future language standard, set to 2 if the token should be enabled in the 108 /// specified langauge, set to 1 if it is an extension in the specified 109 /// language, and set to 0 if disabled in the specified language. 110 static void AddKeyword(StringRef Keyword, 111 tok::TokenKind TokenCode, unsigned Flags, 112 const LangOptions &LangOpts, IdentifierTable &Table) { 113 unsigned AddResult = 0; 114 if (Flags == KEYALL) AddResult = 2; 115 else if (LangOpts.CPlusPlus && (Flags & KEYCXX)) AddResult = 2; 116 else if (LangOpts.CPlusPlus0x && (Flags & KEYCXX0X)) AddResult = 2; 117 else if (LangOpts.C99 && (Flags & KEYC99)) AddResult = 2; 118 else if (LangOpts.GNUKeywords && (Flags & KEYGNU)) AddResult = 1; 119 else if (LangOpts.MicrosoftExt && (Flags & KEYMS)) AddResult = 1; 120 else if (LangOpts.Borland && (Flags & KEYBORLAND)) AddResult = 1; 121 else if (LangOpts.Bool && (Flags & BOOLSUPPORT)) AddResult = 2; 122 else if (LangOpts.AltiVec && (Flags & KEYALTIVEC)) AddResult = 2; 123 else if (LangOpts.OpenCL && (Flags & KEYOPENCL)) AddResult = 2; 124 else if (!LangOpts.CPlusPlus && (Flags & KEYNOCXX)) AddResult = 2; 125 else if (LangOpts.C1X && (Flags & KEYC1X)) AddResult = 2; 126 else if (LangOpts.ObjCAutoRefCount && (Flags & KEYARC)) AddResult = 2; 127 else if (LangOpts.CPlusPlus && (Flags & KEYCXX0X)) AddResult = 3; 128 129 // Don't add this keyword if disabled in this language. 130 if (AddResult == 0) return; 131 132 IdentifierInfo &Info = 133 Table.get(Keyword, AddResult == 3 ? tok::identifier : TokenCode); 134 Info.setIsExtensionToken(AddResult == 1); 135 Info.setIsCXX11CompatKeyword(AddResult == 3); 136 } 137 138 /// AddCXXOperatorKeyword - Register a C++ operator keyword alternative 139 /// representations. 140 static void AddCXXOperatorKeyword(StringRef Keyword, 141 tok::TokenKind TokenCode, 142 IdentifierTable &Table) { 143 IdentifierInfo &Info = Table.get(Keyword, TokenCode); 144 Info.setIsCPlusPlusOperatorKeyword(); 145 } 146 147 /// AddObjCKeyword - Register an Objective-C @keyword like "class" "selector" or 148 /// "property". 149 static void AddObjCKeyword(StringRef Name, 150 tok::ObjCKeywordKind ObjCID, 151 IdentifierTable &Table) { 152 Table.get(Name).setObjCKeywordID(ObjCID); 153 } 154 155 /// AddKeywords - Add all keywords to the symbol table. 156 /// 157 void IdentifierTable::AddKeywords(const LangOptions &LangOpts) { 158 // Add keywords and tokens for the current language. 159 #define KEYWORD(NAME, FLAGS) \ 160 AddKeyword(StringRef(#NAME), tok::kw_ ## NAME, \ 161 FLAGS, LangOpts, *this); 162 #define ALIAS(NAME, TOK, FLAGS) \ 163 AddKeyword(StringRef(NAME), tok::kw_ ## TOK, \ 164 FLAGS, LangOpts, *this); 165 #define CXX_KEYWORD_OPERATOR(NAME, ALIAS) \ 166 if (LangOpts.CXXOperatorNames) \ 167 AddCXXOperatorKeyword(StringRef(#NAME), tok::ALIAS, *this); 168 #define OBJC1_AT_KEYWORD(NAME) \ 169 if (LangOpts.ObjC1) \ 170 AddObjCKeyword(StringRef(#NAME), tok::objc_##NAME, *this); 171 #define OBJC2_AT_KEYWORD(NAME) \ 172 if (LangOpts.ObjC2) \ 173 AddObjCKeyword(StringRef(#NAME), tok::objc_##NAME, *this); 174 #define TESTING_KEYWORD(NAME, FLAGS) 175 #include "clang/Basic/TokenKinds.def" 176 177 if (LangOpts.ParseUnknownAnytype) 178 AddKeyword("__unknown_anytype", tok::kw___unknown_anytype, KEYALL, 179 LangOpts, *this); 180 } 181 182 tok::PPKeywordKind IdentifierInfo::getPPKeywordID() const { 183 // We use a perfect hash function here involving the length of the keyword, 184 // the first and third character. For preprocessor ID's there are no 185 // collisions (if there were, the switch below would complain about duplicate 186 // case values). Note that this depends on 'if' being null terminated. 187 188 #define HASH(LEN, FIRST, THIRD) \ 189 (LEN << 5) + (((FIRST-'a') + (THIRD-'a')) & 31) 190 #define CASE(LEN, FIRST, THIRD, NAME) \ 191 case HASH(LEN, FIRST, THIRD): \ 192 return memcmp(Name, #NAME, LEN) ? tok::pp_not_keyword : tok::pp_ ## NAME 193 194 unsigned Len = getLength(); 195 if (Len < 2) return tok::pp_not_keyword; 196 const char *Name = getNameStart(); 197 switch (HASH(Len, Name[0], Name[2])) { 198 default: return tok::pp_not_keyword; 199 CASE( 2, 'i', '\0', if); 200 CASE( 4, 'e', 'i', elif); 201 CASE( 4, 'e', 's', else); 202 CASE( 4, 'l', 'n', line); 203 CASE( 4, 's', 'c', sccs); 204 CASE( 5, 'e', 'd', endif); 205 CASE( 5, 'e', 'r', error); 206 CASE( 5, 'i', 'e', ident); 207 CASE( 5, 'i', 'd', ifdef); 208 CASE( 5, 'u', 'd', undef); 209 210 CASE( 6, 'a', 's', assert); 211 CASE( 6, 'd', 'f', define); 212 CASE( 6, 'i', 'n', ifndef); 213 CASE( 6, 'i', 'p', import); 214 CASE( 6, 'p', 'a', pragma); 215 216 CASE( 7, 'd', 'f', defined); 217 CASE( 7, 'i', 'c', include); 218 CASE( 7, 'w', 'r', warning); 219 220 CASE( 8, 'u', 'a', unassert); 221 CASE(12, 'i', 'c', include_next); 222 223 CASE(16, '_', 'i', __include_macros); 224 CASE(16, '_', 'e', __export_macro__); 225 226 CASE(17, '_', 'p', __private_macro__); 227 #undef CASE 228 #undef HASH 229 } 230 } 231 232 //===----------------------------------------------------------------------===// 233 // Stats Implementation 234 //===----------------------------------------------------------------------===// 235 236 /// PrintStats - Print statistics about how well the identifier table is doing 237 /// at hashing identifiers. 238 void IdentifierTable::PrintStats() const { 239 unsigned NumBuckets = HashTable.getNumBuckets(); 240 unsigned NumIdentifiers = HashTable.getNumItems(); 241 unsigned NumEmptyBuckets = NumBuckets-NumIdentifiers; 242 unsigned AverageIdentifierSize = 0; 243 unsigned MaxIdentifierLength = 0; 244 245 // TODO: Figure out maximum times an identifier had to probe for -stats. 246 for (llvm::StringMap<IdentifierInfo*, llvm::BumpPtrAllocator>::const_iterator 247 I = HashTable.begin(), E = HashTable.end(); I != E; ++I) { 248 unsigned IdLen = I->getKeyLength(); 249 AverageIdentifierSize += IdLen; 250 if (MaxIdentifierLength < IdLen) 251 MaxIdentifierLength = IdLen; 252 } 253 254 fprintf(stderr, "\n*** Identifier Table Stats:\n"); 255 fprintf(stderr, "# Identifiers: %d\n", NumIdentifiers); 256 fprintf(stderr, "# Empty Buckets: %d\n", NumEmptyBuckets); 257 fprintf(stderr, "Hash density (#identifiers per bucket): %f\n", 258 NumIdentifiers/(double)NumBuckets); 259 fprintf(stderr, "Ave identifier length: %f\n", 260 (AverageIdentifierSize/(double)NumIdentifiers)); 261 fprintf(stderr, "Max identifier length: %d\n", MaxIdentifierLength); 262 263 // Compute statistics about the memory allocated for identifiers. 264 HashTable.getAllocator().PrintStats(); 265 } 266 267 //===----------------------------------------------------------------------===// 268 // SelectorTable Implementation 269 //===----------------------------------------------------------------------===// 270 271 unsigned llvm::DenseMapInfo<clang::Selector>::getHashValue(clang::Selector S) { 272 return DenseMapInfo<void*>::getHashValue(S.getAsOpaquePtr()); 273 } 274 275 namespace clang { 276 /// MultiKeywordSelector - One of these variable length records is kept for each 277 /// selector containing more than one keyword. We use a folding set 278 /// to unique aggregate names (keyword selectors in ObjC parlance). Access to 279 /// this class is provided strictly through Selector. 280 class MultiKeywordSelector 281 : public DeclarationNameExtra, public llvm::FoldingSetNode { 282 MultiKeywordSelector(unsigned nKeys) { 283 ExtraKindOrNumArgs = NUM_EXTRA_KINDS + nKeys; 284 } 285 public: 286 // Constructor for keyword selectors. 287 MultiKeywordSelector(unsigned nKeys, IdentifierInfo **IIV) { 288 assert((nKeys > 1) && "not a multi-keyword selector"); 289 ExtraKindOrNumArgs = NUM_EXTRA_KINDS + nKeys; 290 291 // Fill in the trailing keyword array. 292 IdentifierInfo **KeyInfo = reinterpret_cast<IdentifierInfo **>(this+1); 293 for (unsigned i = 0; i != nKeys; ++i) 294 KeyInfo[i] = IIV[i]; 295 } 296 297 // getName - Derive the full selector name and return it. 298 std::string getName() const; 299 300 unsigned getNumArgs() const { return ExtraKindOrNumArgs - NUM_EXTRA_KINDS; } 301 302 typedef IdentifierInfo *const *keyword_iterator; 303 keyword_iterator keyword_begin() const { 304 return reinterpret_cast<keyword_iterator>(this+1); 305 } 306 keyword_iterator keyword_end() const { 307 return keyword_begin()+getNumArgs(); 308 } 309 IdentifierInfo *getIdentifierInfoForSlot(unsigned i) const { 310 assert(i < getNumArgs() && "getIdentifierInfoForSlot(): illegal index"); 311 return keyword_begin()[i]; 312 } 313 static void Profile(llvm::FoldingSetNodeID &ID, 314 keyword_iterator ArgTys, unsigned NumArgs) { 315 ID.AddInteger(NumArgs); 316 for (unsigned i = 0; i != NumArgs; ++i) 317 ID.AddPointer(ArgTys[i]); 318 } 319 void Profile(llvm::FoldingSetNodeID &ID) { 320 Profile(ID, keyword_begin(), getNumArgs()); 321 } 322 }; 323 } // end namespace clang. 324 325 unsigned Selector::getNumArgs() const { 326 unsigned IIF = getIdentifierInfoFlag(); 327 if (IIF == ZeroArg) 328 return 0; 329 if (IIF == OneArg) 330 return 1; 331 // We point to a MultiKeywordSelector (pointer doesn't contain any flags). 332 MultiKeywordSelector *SI = reinterpret_cast<MultiKeywordSelector *>(InfoPtr); 333 return SI->getNumArgs(); 334 } 335 336 IdentifierInfo *Selector::getIdentifierInfoForSlot(unsigned argIndex) const { 337 if (getIdentifierInfoFlag()) { 338 assert(argIndex == 0 && "illegal keyword index"); 339 return getAsIdentifierInfo(); 340 } 341 // We point to a MultiKeywordSelector (pointer doesn't contain any flags). 342 MultiKeywordSelector *SI = reinterpret_cast<MultiKeywordSelector *>(InfoPtr); 343 return SI->getIdentifierInfoForSlot(argIndex); 344 } 345 346 StringRef Selector::getNameForSlot(unsigned int argIndex) const { 347 IdentifierInfo *II = getIdentifierInfoForSlot(argIndex); 348 return II? II->getName() : StringRef(); 349 } 350 351 std::string MultiKeywordSelector::getName() const { 352 llvm::SmallString<256> Str; 353 llvm::raw_svector_ostream OS(Str); 354 for (keyword_iterator I = keyword_begin(), E = keyword_end(); I != E; ++I) { 355 if (*I) 356 OS << (*I)->getName(); 357 OS << ':'; 358 } 359 360 return OS.str(); 361 } 362 363 std::string Selector::getAsString() const { 364 if (InfoPtr == 0) 365 return "<null selector>"; 366 367 if (InfoPtr & ArgFlags) { 368 IdentifierInfo *II = getAsIdentifierInfo(); 369 370 // If the number of arguments is 0 then II is guaranteed to not be null. 371 if (getNumArgs() == 0) 372 return II->getName(); 373 374 if (!II) 375 return ":"; 376 377 return II->getName().str() + ":"; 378 } 379 380 // We have a multiple keyword selector (no embedded flags). 381 return reinterpret_cast<MultiKeywordSelector *>(InfoPtr)->getName(); 382 } 383 384 /// Interpreting the given string using the normal CamelCase 385 /// conventions, determine whether the given string starts with the 386 /// given "word", which is assumed to end in a lowercase letter. 387 static bool startsWithWord(StringRef name, StringRef word) { 388 if (name.size() < word.size()) return false; 389 return ((name.size() == word.size() || 390 !islower(name[word.size()])) 391 && name.startswith(word)); 392 } 393 394 ObjCMethodFamily Selector::getMethodFamilyImpl(Selector sel) { 395 IdentifierInfo *first = sel.getIdentifierInfoForSlot(0); 396 if (!first) return OMF_None; 397 398 StringRef name = first->getName(); 399 if (sel.isUnarySelector()) { 400 if (name == "autorelease") return OMF_autorelease; 401 if (name == "dealloc") return OMF_dealloc; 402 if (name == "finalize") return OMF_finalize; 403 if (name == "release") return OMF_release; 404 if (name == "retain") return OMF_retain; 405 if (name == "retainCount") return OMF_retainCount; 406 if (name == "self") return OMF_self; 407 } 408 409 if (name == "performSelector") return OMF_performSelector; 410 411 // The other method families may begin with a prefix of underscores. 412 while (!name.empty() && name.front() == '_') 413 name = name.substr(1); 414 415 if (name.empty()) return OMF_None; 416 switch (name.front()) { 417 case 'a': 418 if (startsWithWord(name, "alloc")) return OMF_alloc; 419 break; 420 case 'c': 421 if (startsWithWord(name, "copy")) return OMF_copy; 422 break; 423 case 'i': 424 if (startsWithWord(name, "init")) return OMF_init; 425 break; 426 case 'm': 427 if (startsWithWord(name, "mutableCopy")) return OMF_mutableCopy; 428 break; 429 case 'n': 430 if (startsWithWord(name, "new")) return OMF_new; 431 break; 432 default: 433 break; 434 } 435 436 return OMF_None; 437 } 438 439 namespace { 440 struct SelectorTableImpl { 441 llvm::FoldingSet<MultiKeywordSelector> Table; 442 llvm::BumpPtrAllocator Allocator; 443 }; 444 } // end anonymous namespace. 445 446 static SelectorTableImpl &getSelectorTableImpl(void *P) { 447 return *static_cast<SelectorTableImpl*>(P); 448 } 449 450 size_t SelectorTable::getTotalMemory() const { 451 SelectorTableImpl &SelTabImpl = getSelectorTableImpl(Impl); 452 return SelTabImpl.Allocator.getTotalMemory(); 453 } 454 455 Selector SelectorTable::getSelector(unsigned nKeys, IdentifierInfo **IIV) { 456 if (nKeys < 2) 457 return Selector(IIV[0], nKeys); 458 459 SelectorTableImpl &SelTabImpl = getSelectorTableImpl(Impl); 460 461 // Unique selector, to guarantee there is one per name. 462 llvm::FoldingSetNodeID ID; 463 MultiKeywordSelector::Profile(ID, IIV, nKeys); 464 465 void *InsertPos = 0; 466 if (MultiKeywordSelector *SI = 467 SelTabImpl.Table.FindNodeOrInsertPos(ID, InsertPos)) 468 return Selector(SI); 469 470 // MultiKeywordSelector objects are not allocated with new because they have a 471 // variable size array (for parameter types) at the end of them. 472 unsigned Size = sizeof(MultiKeywordSelector) + nKeys*sizeof(IdentifierInfo *); 473 MultiKeywordSelector *SI = 474 (MultiKeywordSelector*)SelTabImpl.Allocator.Allocate(Size, 475 llvm::alignOf<MultiKeywordSelector>()); 476 new (SI) MultiKeywordSelector(nKeys, IIV); 477 SelTabImpl.Table.InsertNode(SI, InsertPos); 478 return Selector(SI); 479 } 480 481 SelectorTable::SelectorTable() { 482 Impl = new SelectorTableImpl(); 483 } 484 485 SelectorTable::~SelectorTable() { 486 delete &getSelectorTableImpl(Impl); 487 } 488 489 const char *clang::getOperatorSpelling(OverloadedOperatorKind Operator) { 490 switch (Operator) { 491 case OO_None: 492 case NUM_OVERLOADED_OPERATORS: 493 return 0; 494 495 #define OVERLOADED_OPERATOR(Name,Spelling,Token,Unary,Binary,MemberOnly) \ 496 case OO_##Name: return Spelling; 497 #include "clang/Basic/OperatorKinds.def" 498 } 499 500 return 0; 501 } 502