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/SmallString.h" 20 #include "llvm/ADT/StringSwitch.h" 21 #include "llvm/Support/raw_ostream.h" 22 #include "llvm/Support/ErrorHandling.h" 23 #include <cstdio> 24 25 using namespace clang; 26 27 //===----------------------------------------------------------------------===// 28 // IdentifierInfo Implementation 29 //===----------------------------------------------------------------------===// 30 31 IdentifierInfo::IdentifierInfo() { 32 TokenID = tok::identifier; 33 ObjCOrBuiltinID = 0; 34 HasMacro = false; 35 IsExtension = false; 36 IsCXX11CompatKeyword = false; 37 IsPoisoned = false; 38 IsCPPOperatorKeyword = false; 39 NeedsHandleIdentifier = false; 40 IsFromAST = false; 41 ChangedAfterLoad = false; 42 RevertedTokenID = false; 43 OutOfDate = false; 44 IsModulesImport = false; 45 FETokenInfo = 0; 46 Entry = 0; 47 } 48 49 //===----------------------------------------------------------------------===// 50 // IdentifierTable Implementation 51 //===----------------------------------------------------------------------===// 52 53 IdentifierIterator::~IdentifierIterator() { } 54 55 IdentifierInfoLookup::~IdentifierInfoLookup() {} 56 57 namespace { 58 /// \brief A simple identifier lookup iterator that represents an 59 /// empty sequence of identifiers. 60 class EmptyLookupIterator : public IdentifierIterator 61 { 62 public: 63 virtual StringRef Next() { return StringRef(); } 64 }; 65 } 66 67 IdentifierIterator *IdentifierInfoLookup::getIdentifiers() const { 68 return new EmptyLookupIterator(); 69 } 70 71 ExternalIdentifierLookup::~ExternalIdentifierLookup() {} 72 73 IdentifierTable::IdentifierTable(const LangOptions &LangOpts, 74 IdentifierInfoLookup* externalLookup) 75 : HashTable(8192), // Start with space for 8K identifiers. 76 ExternalLookup(externalLookup) { 77 78 // Populate the identifier table with info about keywords for the current 79 // language. 80 AddKeywords(LangOpts); 81 82 83 // Add the '_experimental_modules_import' contextual keyword. 84 get("__experimental_modules_import").setModulesImport(true); 85 } 86 87 //===----------------------------------------------------------------------===// 88 // Language Keyword Implementation 89 //===----------------------------------------------------------------------===// 90 91 // Constants for TokenKinds.def 92 namespace { 93 enum { 94 KEYC99 = 0x1, 95 KEYCXX = 0x2, 96 KEYCXX0X = 0x4, 97 KEYGNU = 0x8, 98 KEYMS = 0x10, 99 BOOLSUPPORT = 0x20, 100 KEYALTIVEC = 0x40, 101 KEYNOCXX = 0x80, 102 KEYBORLAND = 0x100, 103 KEYOPENCL = 0x200, 104 KEYC11 = 0x400, 105 KEYARC = 0x800, 106 KEYALL = 0x0fff 107 }; 108 } 109 110 /// AddKeyword - This method is used to associate a token ID with specific 111 /// identifiers because they are language keywords. This causes the lexer to 112 /// automatically map matching identifiers to specialized token codes. 113 /// 114 /// The C90/C99/CPP/CPP0x flags are set to 3 if the token is a keyword in a 115 /// future language standard, set to 2 if the token should be enabled in the 116 /// specified language, set to 1 if it is an extension in the specified 117 /// language, and set to 0 if disabled in the specified language. 118 static void AddKeyword(StringRef Keyword, 119 tok::TokenKind TokenCode, unsigned Flags, 120 const LangOptions &LangOpts, IdentifierTable &Table) { 121 unsigned AddResult = 0; 122 if (Flags == KEYALL) AddResult = 2; 123 else if (LangOpts.CPlusPlus && (Flags & KEYCXX)) AddResult = 2; 124 else if (LangOpts.CPlusPlus0x && (Flags & KEYCXX0X)) AddResult = 2; 125 else if (LangOpts.C99 && (Flags & KEYC99)) AddResult = 2; 126 else if (LangOpts.GNUKeywords && (Flags & KEYGNU)) AddResult = 1; 127 else if (LangOpts.MicrosoftExt && (Flags & KEYMS)) AddResult = 1; 128 else if (LangOpts.Borland && (Flags & KEYBORLAND)) AddResult = 1; 129 else if (LangOpts.Bool && (Flags & BOOLSUPPORT)) AddResult = 2; 130 else if (LangOpts.AltiVec && (Flags & KEYALTIVEC)) AddResult = 2; 131 else if (LangOpts.OpenCL && (Flags & KEYOPENCL)) AddResult = 2; 132 else if (!LangOpts.CPlusPlus && (Flags & KEYNOCXX)) AddResult = 2; 133 else if (LangOpts.C11 && (Flags & KEYC11)) AddResult = 2; 134 // We treat bridge casts as objective-C keywords so we can warn on them 135 // in non-arc mode. 136 else if (LangOpts.ObjC2 && (Flags & KEYARC)) AddResult = 2; 137 else if (LangOpts.CPlusPlus && (Flags & KEYCXX0X)) AddResult = 3; 138 139 // Don't add this keyword if disabled in this language. 140 if (AddResult == 0) return; 141 142 IdentifierInfo &Info = 143 Table.get(Keyword, AddResult == 3 ? tok::identifier : TokenCode); 144 Info.setIsExtensionToken(AddResult == 1); 145 Info.setIsCXX11CompatKeyword(AddResult == 3); 146 } 147 148 /// AddCXXOperatorKeyword - Register a C++ operator keyword alternative 149 /// representations. 150 static void AddCXXOperatorKeyword(StringRef Keyword, 151 tok::TokenKind TokenCode, 152 IdentifierTable &Table) { 153 IdentifierInfo &Info = Table.get(Keyword, TokenCode); 154 Info.setIsCPlusPlusOperatorKeyword(); 155 } 156 157 /// AddObjCKeyword - Register an Objective-C @keyword like "class" "selector" or 158 /// "property". 159 static void AddObjCKeyword(StringRef Name, 160 tok::ObjCKeywordKind ObjCID, 161 IdentifierTable &Table) { 162 Table.get(Name).setObjCKeywordID(ObjCID); 163 } 164 165 /// AddKeywords - Add all keywords to the symbol table. 166 /// 167 void IdentifierTable::AddKeywords(const LangOptions &LangOpts) { 168 // Add keywords and tokens for the current language. 169 #define KEYWORD(NAME, FLAGS) \ 170 AddKeyword(StringRef(#NAME), tok::kw_ ## NAME, \ 171 FLAGS, LangOpts, *this); 172 #define ALIAS(NAME, TOK, FLAGS) \ 173 AddKeyword(StringRef(NAME), tok::kw_ ## TOK, \ 174 FLAGS, LangOpts, *this); 175 #define CXX_KEYWORD_OPERATOR(NAME, ALIAS) \ 176 if (LangOpts.CXXOperatorNames) \ 177 AddCXXOperatorKeyword(StringRef(#NAME), tok::ALIAS, *this); 178 #define OBJC1_AT_KEYWORD(NAME) \ 179 if (LangOpts.ObjC1) \ 180 AddObjCKeyword(StringRef(#NAME), tok::objc_##NAME, *this); 181 #define OBJC2_AT_KEYWORD(NAME) \ 182 if (LangOpts.ObjC2) \ 183 AddObjCKeyword(StringRef(#NAME), tok::objc_##NAME, *this); 184 #define TESTING_KEYWORD(NAME, FLAGS) 185 #include "clang/Basic/TokenKinds.def" 186 187 if (LangOpts.ParseUnknownAnytype) 188 AddKeyword("__unknown_anytype", tok::kw___unknown_anytype, KEYALL, 189 LangOpts, *this); 190 } 191 192 tok::PPKeywordKind IdentifierInfo::getPPKeywordID() const { 193 // We use a perfect hash function here involving the length of the keyword, 194 // the first and third character. For preprocessor ID's there are no 195 // collisions (if there were, the switch below would complain about duplicate 196 // case values). Note that this depends on 'if' being null terminated. 197 198 #define HASH(LEN, FIRST, THIRD) \ 199 (LEN << 5) + (((FIRST-'a') + (THIRD-'a')) & 31) 200 #define CASE(LEN, FIRST, THIRD, NAME) \ 201 case HASH(LEN, FIRST, THIRD): \ 202 return memcmp(Name, #NAME, LEN) ? tok::pp_not_keyword : tok::pp_ ## NAME 203 204 unsigned Len = getLength(); 205 if (Len < 2) return tok::pp_not_keyword; 206 const char *Name = getNameStart(); 207 switch (HASH(Len, Name[0], Name[2])) { 208 default: return tok::pp_not_keyword; 209 CASE( 2, 'i', '\0', if); 210 CASE( 4, 'e', 'i', elif); 211 CASE( 4, 'e', 's', else); 212 CASE( 4, 'l', 'n', line); 213 CASE( 4, 's', 'c', sccs); 214 CASE( 5, 'e', 'd', endif); 215 CASE( 5, 'e', 'r', error); 216 CASE( 5, 'i', 'e', ident); 217 CASE( 5, 'i', 'd', ifdef); 218 CASE( 5, 'u', 'd', undef); 219 220 CASE( 6, 'a', 's', assert); 221 CASE( 6, 'd', 'f', define); 222 CASE( 6, 'i', 'n', ifndef); 223 CASE( 6, 'i', 'p', import); 224 CASE( 6, 'p', 'a', pragma); 225 226 CASE( 7, 'd', 'f', defined); 227 CASE( 7, 'i', 'c', include); 228 CASE( 7, 'w', 'r', warning); 229 230 CASE( 8, 'u', 'a', unassert); 231 CASE(12, 'i', 'c', include_next); 232 233 CASE(14, '_', 'p', __public_macro); 234 235 CASE(15, '_', 'p', __private_macro); 236 237 CASE(16, '_', 'i', __include_macros); 238 #undef CASE 239 #undef HASH 240 } 241 } 242 243 //===----------------------------------------------------------------------===// 244 // Stats Implementation 245 //===----------------------------------------------------------------------===// 246 247 /// PrintStats - Print statistics about how well the identifier table is doing 248 /// at hashing identifiers. 249 void IdentifierTable::PrintStats() const { 250 unsigned NumBuckets = HashTable.getNumBuckets(); 251 unsigned NumIdentifiers = HashTable.getNumItems(); 252 unsigned NumEmptyBuckets = NumBuckets-NumIdentifiers; 253 unsigned AverageIdentifierSize = 0; 254 unsigned MaxIdentifierLength = 0; 255 256 // TODO: Figure out maximum times an identifier had to probe for -stats. 257 for (llvm::StringMap<IdentifierInfo*, llvm::BumpPtrAllocator>::const_iterator 258 I = HashTable.begin(), E = HashTable.end(); I != E; ++I) { 259 unsigned IdLen = I->getKeyLength(); 260 AverageIdentifierSize += IdLen; 261 if (MaxIdentifierLength < IdLen) 262 MaxIdentifierLength = IdLen; 263 } 264 265 fprintf(stderr, "\n*** Identifier Table Stats:\n"); 266 fprintf(stderr, "# Identifiers: %d\n", NumIdentifiers); 267 fprintf(stderr, "# Empty Buckets: %d\n", NumEmptyBuckets); 268 fprintf(stderr, "Hash density (#identifiers per bucket): %f\n", 269 NumIdentifiers/(double)NumBuckets); 270 fprintf(stderr, "Ave identifier length: %f\n", 271 (AverageIdentifierSize/(double)NumIdentifiers)); 272 fprintf(stderr, "Max identifier length: %d\n", MaxIdentifierLength); 273 274 // Compute statistics about the memory allocated for identifiers. 275 HashTable.getAllocator().PrintStats(); 276 } 277 278 //===----------------------------------------------------------------------===// 279 // SelectorTable Implementation 280 //===----------------------------------------------------------------------===// 281 282 unsigned llvm::DenseMapInfo<clang::Selector>::getHashValue(clang::Selector S) { 283 return DenseMapInfo<void*>::getHashValue(S.getAsOpaquePtr()); 284 } 285 286 namespace clang { 287 /// MultiKeywordSelector - One of these variable length records is kept for each 288 /// selector containing more than one keyword. We use a folding set 289 /// to unique aggregate names (keyword selectors in ObjC parlance). Access to 290 /// this class is provided strictly through Selector. 291 class MultiKeywordSelector 292 : public DeclarationNameExtra, public llvm::FoldingSetNode { 293 MultiKeywordSelector(unsigned nKeys) { 294 ExtraKindOrNumArgs = NUM_EXTRA_KINDS + nKeys; 295 } 296 public: 297 // Constructor for keyword selectors. 298 MultiKeywordSelector(unsigned nKeys, IdentifierInfo **IIV) { 299 assert((nKeys > 1) && "not a multi-keyword selector"); 300 ExtraKindOrNumArgs = NUM_EXTRA_KINDS + nKeys; 301 302 // Fill in the trailing keyword array. 303 IdentifierInfo **KeyInfo = reinterpret_cast<IdentifierInfo **>(this+1); 304 for (unsigned i = 0; i != nKeys; ++i) 305 KeyInfo[i] = IIV[i]; 306 } 307 308 // getName - Derive the full selector name and return it. 309 std::string getName() const; 310 311 unsigned getNumArgs() const { return ExtraKindOrNumArgs - NUM_EXTRA_KINDS; } 312 313 typedef IdentifierInfo *const *keyword_iterator; 314 keyword_iterator keyword_begin() const { 315 return reinterpret_cast<keyword_iterator>(this+1); 316 } 317 keyword_iterator keyword_end() const { 318 return keyword_begin()+getNumArgs(); 319 } 320 IdentifierInfo *getIdentifierInfoForSlot(unsigned i) const { 321 assert(i < getNumArgs() && "getIdentifierInfoForSlot(): illegal index"); 322 return keyword_begin()[i]; 323 } 324 static void Profile(llvm::FoldingSetNodeID &ID, 325 keyword_iterator ArgTys, unsigned NumArgs) { 326 ID.AddInteger(NumArgs); 327 for (unsigned i = 0; i != NumArgs; ++i) 328 ID.AddPointer(ArgTys[i]); 329 } 330 void Profile(llvm::FoldingSetNodeID &ID) { 331 Profile(ID, keyword_begin(), getNumArgs()); 332 } 333 }; 334 } // end namespace clang. 335 336 unsigned Selector::getNumArgs() const { 337 unsigned IIF = getIdentifierInfoFlag(); 338 if (IIF == ZeroArg) 339 return 0; 340 if (IIF == OneArg) 341 return 1; 342 // We point to a MultiKeywordSelector (pointer doesn't contain any flags). 343 MultiKeywordSelector *SI = reinterpret_cast<MultiKeywordSelector *>(InfoPtr); 344 return SI->getNumArgs(); 345 } 346 347 IdentifierInfo *Selector::getIdentifierInfoForSlot(unsigned argIndex) const { 348 if (getIdentifierInfoFlag()) { 349 assert(argIndex == 0 && "illegal keyword index"); 350 return getAsIdentifierInfo(); 351 } 352 // We point to a MultiKeywordSelector (pointer doesn't contain any flags). 353 MultiKeywordSelector *SI = reinterpret_cast<MultiKeywordSelector *>(InfoPtr); 354 return SI->getIdentifierInfoForSlot(argIndex); 355 } 356 357 StringRef Selector::getNameForSlot(unsigned int argIndex) const { 358 IdentifierInfo *II = getIdentifierInfoForSlot(argIndex); 359 return II? II->getName() : StringRef(); 360 } 361 362 std::string MultiKeywordSelector::getName() const { 363 SmallString<256> Str; 364 llvm::raw_svector_ostream OS(Str); 365 for (keyword_iterator I = keyword_begin(), E = keyword_end(); I != E; ++I) { 366 if (*I) 367 OS << (*I)->getName(); 368 OS << ':'; 369 } 370 371 return OS.str(); 372 } 373 374 std::string Selector::getAsString() const { 375 if (InfoPtr == 0) 376 return "<null selector>"; 377 378 if (InfoPtr & ArgFlags) { 379 IdentifierInfo *II = getAsIdentifierInfo(); 380 381 // If the number of arguments is 0 then II is guaranteed to not be null. 382 if (getNumArgs() == 0) 383 return II->getName(); 384 385 if (!II) 386 return ":"; 387 388 return II->getName().str() + ":"; 389 } 390 391 // We have a multiple keyword selector (no embedded flags). 392 return reinterpret_cast<MultiKeywordSelector *>(InfoPtr)->getName(); 393 } 394 395 /// Interpreting the given string using the normal CamelCase 396 /// conventions, determine whether the given string starts with the 397 /// given "word", which is assumed to end in a lowercase letter. 398 static bool startsWithWord(StringRef name, StringRef word) { 399 if (name.size() < word.size()) return false; 400 return ((name.size() == word.size() || 401 !islower(name[word.size()])) 402 && name.startswith(word)); 403 } 404 405 ObjCMethodFamily Selector::getMethodFamilyImpl(Selector sel) { 406 IdentifierInfo *first = sel.getIdentifierInfoForSlot(0); 407 if (!first) return OMF_None; 408 409 StringRef name = first->getName(); 410 if (sel.isUnarySelector()) { 411 if (name == "autorelease") return OMF_autorelease; 412 if (name == "dealloc") return OMF_dealloc; 413 if (name == "finalize") return OMF_finalize; 414 if (name == "release") return OMF_release; 415 if (name == "retain") return OMF_retain; 416 if (name == "retainCount") return OMF_retainCount; 417 if (name == "self") return OMF_self; 418 } 419 420 if (name == "performSelector") return OMF_performSelector; 421 422 // The other method families may begin with a prefix of underscores. 423 while (!name.empty() && name.front() == '_') 424 name = name.substr(1); 425 426 if (name.empty()) return OMF_None; 427 switch (name.front()) { 428 case 'a': 429 if (startsWithWord(name, "alloc")) return OMF_alloc; 430 break; 431 case 'c': 432 if (startsWithWord(name, "copy")) return OMF_copy; 433 break; 434 case 'i': 435 if (startsWithWord(name, "init")) return OMF_init; 436 break; 437 case 'm': 438 if (startsWithWord(name, "mutableCopy")) return OMF_mutableCopy; 439 break; 440 case 'n': 441 if (startsWithWord(name, "new")) return OMF_new; 442 break; 443 default: 444 break; 445 } 446 447 return OMF_None; 448 } 449 450 namespace { 451 struct SelectorTableImpl { 452 llvm::FoldingSet<MultiKeywordSelector> Table; 453 llvm::BumpPtrAllocator Allocator; 454 }; 455 } // end anonymous namespace. 456 457 static SelectorTableImpl &getSelectorTableImpl(void *P) { 458 return *static_cast<SelectorTableImpl*>(P); 459 } 460 461 /*static*/ Selector 462 SelectorTable::constructSetterName(IdentifierTable &Idents, 463 SelectorTable &SelTable, 464 const IdentifierInfo *Name) { 465 SmallString<100> SelectorName; 466 SelectorName = "set"; 467 SelectorName += Name->getName(); 468 SelectorName[3] = toupper(SelectorName[3]); 469 IdentifierInfo *SetterName = &Idents.get(SelectorName); 470 return SelTable.getUnarySelector(SetterName); 471 } 472 473 size_t SelectorTable::getTotalMemory() const { 474 SelectorTableImpl &SelTabImpl = getSelectorTableImpl(Impl); 475 return SelTabImpl.Allocator.getTotalMemory(); 476 } 477 478 Selector SelectorTable::getSelector(unsigned nKeys, IdentifierInfo **IIV) { 479 if (nKeys < 2) 480 return Selector(IIV[0], nKeys); 481 482 SelectorTableImpl &SelTabImpl = getSelectorTableImpl(Impl); 483 484 // Unique selector, to guarantee there is one per name. 485 llvm::FoldingSetNodeID ID; 486 MultiKeywordSelector::Profile(ID, IIV, nKeys); 487 488 void *InsertPos = 0; 489 if (MultiKeywordSelector *SI = 490 SelTabImpl.Table.FindNodeOrInsertPos(ID, InsertPos)) 491 return Selector(SI); 492 493 // MultiKeywordSelector objects are not allocated with new because they have a 494 // variable size array (for parameter types) at the end of them. 495 unsigned Size = sizeof(MultiKeywordSelector) + nKeys*sizeof(IdentifierInfo *); 496 MultiKeywordSelector *SI = 497 (MultiKeywordSelector*)SelTabImpl.Allocator.Allocate(Size, 498 llvm::alignOf<MultiKeywordSelector>()); 499 new (SI) MultiKeywordSelector(nKeys, IIV); 500 SelTabImpl.Table.InsertNode(SI, InsertPos); 501 return Selector(SI); 502 } 503 504 SelectorTable::SelectorTable() { 505 Impl = new SelectorTableImpl(); 506 } 507 508 SelectorTable::~SelectorTable() { 509 delete &getSelectorTableImpl(Impl); 510 } 511 512 const char *clang::getOperatorSpelling(OverloadedOperatorKind Operator) { 513 switch (Operator) { 514 case OO_None: 515 case NUM_OVERLOADED_OPERATORS: 516 return 0; 517 518 #define OVERLOADED_OPERATOR(Name,Spelling,Token,Unary,Binary,MemberOnly) \ 519 case OO_##Name: return Spelling; 520 #include "clang/Basic/OperatorKinds.def" 521 } 522 523 llvm_unreachable("Invalid OverloadedOperatorKind!"); 524 } 525