IdentifierTable.cpp revision f7ccbad5d9949e7ddd1cbef43d482553b811e026
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 25using namespace clang; 26 27//===----------------------------------------------------------------------===// 28// IdentifierInfo Implementation 29//===----------------------------------------------------------------------===// 30 31IdentifierInfo::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 IsImport = false; 45 FETokenInfo = 0; 46 Entry = 0; 47} 48 49//===----------------------------------------------------------------------===// 50// IdentifierTable Implementation 51//===----------------------------------------------------------------------===// 52 53IdentifierIterator::~IdentifierIterator() { } 54 55IdentifierInfoLookup::~IdentifierInfoLookup() {} 56 57namespace { 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 67IdentifierIterator *IdentifierInfoLookup::getIdentifiers() const { 68 return new EmptyLookupIterator(); 69} 70 71ExternalIdentifierLookup::~ExternalIdentifierLookup() {} 72 73IdentifierTable::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//===----------------------------------------------------------------------===// 84// Language Keyword Implementation 85//===----------------------------------------------------------------------===// 86 87// Constants for TokenKinds.def 88namespace { 89 enum { 90 KEYC99 = 0x1, 91 KEYCXX = 0x2, 92 KEYCXX0X = 0x4, 93 KEYGNU = 0x8, 94 KEYMS = 0x10, 95 BOOLSUPPORT = 0x20, 96 KEYALTIVEC = 0x40, 97 KEYNOCXX = 0x80, 98 KEYBORLAND = 0x100, 99 KEYOPENCL = 0x200, 100 KEYC11 = 0x400, 101 KEYARC = 0x800, 102 KEYALL = 0x0fff 103 }; 104} 105 106/// AddKeyword - This method is used to associate a token ID with specific 107/// identifiers because they are language keywords. This causes the lexer to 108/// automatically map matching identifiers to specialized token codes. 109/// 110/// The C90/C99/CPP/CPP0x flags are set to 3 if the token is a keyword in a 111/// future language standard, set to 2 if the token should be enabled in the 112/// specified langauge, set to 1 if it is an extension in the specified 113/// language, and set to 0 if disabled in the specified language. 114static void AddKeyword(StringRef Keyword, 115 tok::TokenKind TokenCode, unsigned Flags, 116 const LangOptions &LangOpts, IdentifierTable &Table) { 117 unsigned AddResult = 0; 118 if (Flags == KEYALL) AddResult = 2; 119 else if (LangOpts.CPlusPlus && (Flags & KEYCXX)) AddResult = 2; 120 else if (LangOpts.CPlusPlus0x && (Flags & KEYCXX0X)) AddResult = 2; 121 else if (LangOpts.C99 && (Flags & KEYC99)) AddResult = 2; 122 else if (LangOpts.GNUKeywords && (Flags & KEYGNU)) AddResult = 1; 123 else if (LangOpts.MicrosoftExt && (Flags & KEYMS)) AddResult = 1; 124 else if (LangOpts.Borland && (Flags & KEYBORLAND)) AddResult = 1; 125 else if (LangOpts.Bool && (Flags & BOOLSUPPORT)) AddResult = 2; 126 else if (LangOpts.AltiVec && (Flags & KEYALTIVEC)) AddResult = 2; 127 else if (LangOpts.OpenCL && (Flags & KEYOPENCL)) AddResult = 2; 128 else if (!LangOpts.CPlusPlus && (Flags & KEYNOCXX)) AddResult = 2; 129 else if (LangOpts.C11 && (Flags & KEYC11)) AddResult = 2; 130 // We treat bridge casts as objective-C keywords so we can warn on them 131 // in non-arc mode. 132 else if (LangOpts.ObjC2 && (Flags & KEYARC)) AddResult = 2; 133 else if (LangOpts.CPlusPlus && (Flags & KEYCXX0X)) AddResult = 3; 134 135 // Don't add this keyword if disabled in this language. 136 if (AddResult == 0) return; 137 138 IdentifierInfo &Info = 139 Table.get(Keyword, AddResult == 3 ? tok::identifier : TokenCode); 140 Info.setIsExtensionToken(AddResult == 1); 141 Info.setIsCXX11CompatKeyword(AddResult == 3); 142} 143 144/// AddCXXOperatorKeyword - Register a C++ operator keyword alternative 145/// representations. 146static void AddCXXOperatorKeyword(StringRef Keyword, 147 tok::TokenKind TokenCode, 148 IdentifierTable &Table) { 149 IdentifierInfo &Info = Table.get(Keyword, TokenCode); 150 Info.setIsCPlusPlusOperatorKeyword(); 151} 152 153/// AddObjCKeyword - Register an Objective-C @keyword like "class" "selector" or 154/// "property". 155static void AddObjCKeyword(StringRef Name, 156 tok::ObjCKeywordKind ObjCID, 157 IdentifierTable &Table) { 158 Table.get(Name).setObjCKeywordID(ObjCID); 159} 160 161/// AddKeywords - Add all keywords to the symbol table. 162/// 163void IdentifierTable::AddKeywords(const LangOptions &LangOpts) { 164 // Add keywords and tokens for the current language. 165#define KEYWORD(NAME, FLAGS) \ 166 AddKeyword(StringRef(#NAME), tok::kw_ ## NAME, \ 167 FLAGS, LangOpts, *this); 168#define ALIAS(NAME, TOK, FLAGS) \ 169 AddKeyword(StringRef(NAME), tok::kw_ ## TOK, \ 170 FLAGS, LangOpts, *this); 171#define CXX_KEYWORD_OPERATOR(NAME, ALIAS) \ 172 if (LangOpts.CXXOperatorNames) \ 173 AddCXXOperatorKeyword(StringRef(#NAME), tok::ALIAS, *this); 174#define OBJC1_AT_KEYWORD(NAME) \ 175 if (LangOpts.ObjC1) \ 176 AddObjCKeyword(StringRef(#NAME), tok::objc_##NAME, *this); 177#define OBJC2_AT_KEYWORD(NAME) \ 178 if (LangOpts.ObjC2) \ 179 AddObjCKeyword(StringRef(#NAME), tok::objc_##NAME, *this); 180#define TESTING_KEYWORD(NAME, FLAGS) 181#include "clang/Basic/TokenKinds.def" 182 183 if (LangOpts.ParseUnknownAnytype) 184 AddKeyword("__unknown_anytype", tok::kw___unknown_anytype, KEYALL, 185 LangOpts, *this); 186} 187 188tok::PPKeywordKind IdentifierInfo::getPPKeywordID() const { 189 // We use a perfect hash function here involving the length of the keyword, 190 // the first and third character. For preprocessor ID's there are no 191 // collisions (if there were, the switch below would complain about duplicate 192 // case values). Note that this depends on 'if' being null terminated. 193 194#define HASH(LEN, FIRST, THIRD) \ 195 (LEN << 5) + (((FIRST-'a') + (THIRD-'a')) & 31) 196#define CASE(LEN, FIRST, THIRD, NAME) \ 197 case HASH(LEN, FIRST, THIRD): \ 198 return memcmp(Name, #NAME, LEN) ? tok::pp_not_keyword : tok::pp_ ## NAME 199 200 unsigned Len = getLength(); 201 if (Len < 2) return tok::pp_not_keyword; 202 const char *Name = getNameStart(); 203 switch (HASH(Len, Name[0], Name[2])) { 204 default: return tok::pp_not_keyword; 205 CASE( 2, 'i', '\0', if); 206 CASE( 4, 'e', 'i', elif); 207 CASE( 4, 'e', 's', else); 208 CASE( 4, 'l', 'n', line); 209 CASE( 4, 's', 'c', sccs); 210 CASE( 5, 'e', 'd', endif); 211 CASE( 5, 'e', 'r', error); 212 CASE( 5, 'i', 'e', ident); 213 CASE( 5, 'i', 'd', ifdef); 214 CASE( 5, 'u', 'd', undef); 215 216 CASE( 6, 'a', 's', assert); 217 CASE( 6, 'd', 'f', define); 218 CASE( 6, 'i', 'n', ifndef); 219 CASE( 6, 'i', 'p', import); 220 CASE( 6, 'p', 'a', pragma); 221 222 CASE( 7, 'd', 'f', defined); 223 CASE( 7, 'i', 'c', include); 224 CASE( 7, 'w', 'r', warning); 225 226 CASE( 8, 'u', 'a', unassert); 227 CASE(12, 'i', 'c', include_next); 228 229 CASE(14, '_', 'p', __public_macro); 230 231 CASE(15, '_', 'p', __private_macro); 232 233 CASE(16, '_', 'i', __include_macros); 234#undef CASE 235#undef HASH 236 } 237} 238 239//===----------------------------------------------------------------------===// 240// Stats Implementation 241//===----------------------------------------------------------------------===// 242 243/// PrintStats - Print statistics about how well the identifier table is doing 244/// at hashing identifiers. 245void IdentifierTable::PrintStats() const { 246 unsigned NumBuckets = HashTable.getNumBuckets(); 247 unsigned NumIdentifiers = HashTable.getNumItems(); 248 unsigned NumEmptyBuckets = NumBuckets-NumIdentifiers; 249 unsigned AverageIdentifierSize = 0; 250 unsigned MaxIdentifierLength = 0; 251 252 // TODO: Figure out maximum times an identifier had to probe for -stats. 253 for (llvm::StringMap<IdentifierInfo*, llvm::BumpPtrAllocator>::const_iterator 254 I = HashTable.begin(), E = HashTable.end(); I != E; ++I) { 255 unsigned IdLen = I->getKeyLength(); 256 AverageIdentifierSize += IdLen; 257 if (MaxIdentifierLength < IdLen) 258 MaxIdentifierLength = IdLen; 259 } 260 261 fprintf(stderr, "\n*** Identifier Table Stats:\n"); 262 fprintf(stderr, "# Identifiers: %d\n", NumIdentifiers); 263 fprintf(stderr, "# Empty Buckets: %d\n", NumEmptyBuckets); 264 fprintf(stderr, "Hash density (#identifiers per bucket): %f\n", 265 NumIdentifiers/(double)NumBuckets); 266 fprintf(stderr, "Ave identifier length: %f\n", 267 (AverageIdentifierSize/(double)NumIdentifiers)); 268 fprintf(stderr, "Max identifier length: %d\n", MaxIdentifierLength); 269 270 // Compute statistics about the memory allocated for identifiers. 271 HashTable.getAllocator().PrintStats(); 272} 273 274//===----------------------------------------------------------------------===// 275// SelectorTable Implementation 276//===----------------------------------------------------------------------===// 277 278unsigned llvm::DenseMapInfo<clang::Selector>::getHashValue(clang::Selector S) { 279 return DenseMapInfo<void*>::getHashValue(S.getAsOpaquePtr()); 280} 281 282namespace clang { 283/// MultiKeywordSelector - One of these variable length records is kept for each 284/// selector containing more than one keyword. We use a folding set 285/// to unique aggregate names (keyword selectors in ObjC parlance). Access to 286/// this class is provided strictly through Selector. 287class MultiKeywordSelector 288 : public DeclarationNameExtra, public llvm::FoldingSetNode { 289 MultiKeywordSelector(unsigned nKeys) { 290 ExtraKindOrNumArgs = NUM_EXTRA_KINDS + nKeys; 291 } 292public: 293 // Constructor for keyword selectors. 294 MultiKeywordSelector(unsigned nKeys, IdentifierInfo **IIV) { 295 assert((nKeys > 1) && "not a multi-keyword selector"); 296 ExtraKindOrNumArgs = NUM_EXTRA_KINDS + nKeys; 297 298 // Fill in the trailing keyword array. 299 IdentifierInfo **KeyInfo = reinterpret_cast<IdentifierInfo **>(this+1); 300 for (unsigned i = 0; i != nKeys; ++i) 301 KeyInfo[i] = IIV[i]; 302 } 303 304 // getName - Derive the full selector name and return it. 305 std::string getName() const; 306 307 unsigned getNumArgs() const { return ExtraKindOrNumArgs - NUM_EXTRA_KINDS; } 308 309 typedef IdentifierInfo *const *keyword_iterator; 310 keyword_iterator keyword_begin() const { 311 return reinterpret_cast<keyword_iterator>(this+1); 312 } 313 keyword_iterator keyword_end() const { 314 return keyword_begin()+getNumArgs(); 315 } 316 IdentifierInfo *getIdentifierInfoForSlot(unsigned i) const { 317 assert(i < getNumArgs() && "getIdentifierInfoForSlot(): illegal index"); 318 return keyword_begin()[i]; 319 } 320 static void Profile(llvm::FoldingSetNodeID &ID, 321 keyword_iterator ArgTys, unsigned NumArgs) { 322 ID.AddInteger(NumArgs); 323 for (unsigned i = 0; i != NumArgs; ++i) 324 ID.AddPointer(ArgTys[i]); 325 } 326 void Profile(llvm::FoldingSetNodeID &ID) { 327 Profile(ID, keyword_begin(), getNumArgs()); 328 } 329}; 330} // end namespace clang. 331 332unsigned Selector::getNumArgs() const { 333 unsigned IIF = getIdentifierInfoFlag(); 334 if (IIF == ZeroArg) 335 return 0; 336 if (IIF == OneArg) 337 return 1; 338 // We point to a MultiKeywordSelector (pointer doesn't contain any flags). 339 MultiKeywordSelector *SI = reinterpret_cast<MultiKeywordSelector *>(InfoPtr); 340 return SI->getNumArgs(); 341} 342 343IdentifierInfo *Selector::getIdentifierInfoForSlot(unsigned argIndex) const { 344 if (getIdentifierInfoFlag()) { 345 assert(argIndex == 0 && "illegal keyword index"); 346 return getAsIdentifierInfo(); 347 } 348 // We point to a MultiKeywordSelector (pointer doesn't contain any flags). 349 MultiKeywordSelector *SI = reinterpret_cast<MultiKeywordSelector *>(InfoPtr); 350 return SI->getIdentifierInfoForSlot(argIndex); 351} 352 353StringRef Selector::getNameForSlot(unsigned int argIndex) const { 354 IdentifierInfo *II = getIdentifierInfoForSlot(argIndex); 355 return II? II->getName() : StringRef(); 356} 357 358std::string MultiKeywordSelector::getName() const { 359 SmallString<256> Str; 360 llvm::raw_svector_ostream OS(Str); 361 for (keyword_iterator I = keyword_begin(), E = keyword_end(); I != E; ++I) { 362 if (*I) 363 OS << (*I)->getName(); 364 OS << ':'; 365 } 366 367 return OS.str(); 368} 369 370std::string Selector::getAsString() const { 371 if (InfoPtr == 0) 372 return "<null selector>"; 373 374 if (InfoPtr & ArgFlags) { 375 IdentifierInfo *II = getAsIdentifierInfo(); 376 377 // If the number of arguments is 0 then II is guaranteed to not be null. 378 if (getNumArgs() == 0) 379 return II->getName(); 380 381 if (!II) 382 return ":"; 383 384 return II->getName().str() + ":"; 385 } 386 387 // We have a multiple keyword selector (no embedded flags). 388 return reinterpret_cast<MultiKeywordSelector *>(InfoPtr)->getName(); 389} 390 391/// Interpreting the given string using the normal CamelCase 392/// conventions, determine whether the given string starts with the 393/// given "word", which is assumed to end in a lowercase letter. 394static bool startsWithWord(StringRef name, StringRef word) { 395 if (name.size() < word.size()) return false; 396 return ((name.size() == word.size() || 397 !islower(name[word.size()])) 398 && name.startswith(word)); 399} 400 401ObjCMethodFamily Selector::getMethodFamilyImpl(Selector sel) { 402 IdentifierInfo *first = sel.getIdentifierInfoForSlot(0); 403 if (!first) return OMF_None; 404 405 StringRef name = first->getName(); 406 if (sel.isUnarySelector()) { 407 if (name == "autorelease") return OMF_autorelease; 408 if (name == "dealloc") return OMF_dealloc; 409 if (name == "finalize") return OMF_finalize; 410 if (name == "release") return OMF_release; 411 if (name == "retain") return OMF_retain; 412 if (name == "retainCount") return OMF_retainCount; 413 if (name == "self") return OMF_self; 414 } 415 416 if (name == "performSelector") return OMF_performSelector; 417 418 // The other method families may begin with a prefix of underscores. 419 while (!name.empty() && name.front() == '_') 420 name = name.substr(1); 421 422 if (name.empty()) return OMF_None; 423 switch (name.front()) { 424 case 'a': 425 if (startsWithWord(name, "alloc")) return OMF_alloc; 426 break; 427 case 'c': 428 if (startsWithWord(name, "copy")) return OMF_copy; 429 break; 430 case 'i': 431 if (startsWithWord(name, "init")) return OMF_init; 432 break; 433 case 'm': 434 if (startsWithWord(name, "mutableCopy")) return OMF_mutableCopy; 435 break; 436 case 'n': 437 if (startsWithWord(name, "new")) return OMF_new; 438 break; 439 default: 440 break; 441 } 442 443 return OMF_None; 444} 445 446namespace { 447 struct SelectorTableImpl { 448 llvm::FoldingSet<MultiKeywordSelector> Table; 449 llvm::BumpPtrAllocator Allocator; 450 }; 451} // end anonymous namespace. 452 453static SelectorTableImpl &getSelectorTableImpl(void *P) { 454 return *static_cast<SelectorTableImpl*>(P); 455} 456 457/*static*/ Selector 458SelectorTable::constructSetterName(IdentifierTable &Idents, 459 SelectorTable &SelTable, 460 const IdentifierInfo *Name) { 461 SmallString<100> SelectorName; 462 SelectorName = "set"; 463 SelectorName += Name->getName(); 464 SelectorName[3] = toupper(SelectorName[3]); 465 IdentifierInfo *SetterName = &Idents.get(SelectorName); 466 return SelTable.getUnarySelector(SetterName); 467} 468 469size_t SelectorTable::getTotalMemory() const { 470 SelectorTableImpl &SelTabImpl = getSelectorTableImpl(Impl); 471 return SelTabImpl.Allocator.getTotalMemory(); 472} 473 474Selector SelectorTable::getSelector(unsigned nKeys, IdentifierInfo **IIV) { 475 if (nKeys < 2) 476 return Selector(IIV[0], nKeys); 477 478 SelectorTableImpl &SelTabImpl = getSelectorTableImpl(Impl); 479 480 // Unique selector, to guarantee there is one per name. 481 llvm::FoldingSetNodeID ID; 482 MultiKeywordSelector::Profile(ID, IIV, nKeys); 483 484 void *InsertPos = 0; 485 if (MultiKeywordSelector *SI = 486 SelTabImpl.Table.FindNodeOrInsertPos(ID, InsertPos)) 487 return Selector(SI); 488 489 // MultiKeywordSelector objects are not allocated with new because they have a 490 // variable size array (for parameter types) at the end of them. 491 unsigned Size = sizeof(MultiKeywordSelector) + nKeys*sizeof(IdentifierInfo *); 492 MultiKeywordSelector *SI = 493 (MultiKeywordSelector*)SelTabImpl.Allocator.Allocate(Size, 494 llvm::alignOf<MultiKeywordSelector>()); 495 new (SI) MultiKeywordSelector(nKeys, IIV); 496 SelTabImpl.Table.InsertNode(SI, InsertPos); 497 return Selector(SI); 498} 499 500SelectorTable::SelectorTable() { 501 Impl = new SelectorTableImpl(); 502} 503 504SelectorTable::~SelectorTable() { 505 delete &getSelectorTableImpl(Impl); 506} 507 508const char *clang::getOperatorSpelling(OverloadedOperatorKind Operator) { 509 switch (Operator) { 510 case OO_None: 511 case NUM_OVERLOADED_OPERATORS: 512 return 0; 513 514#define OVERLOADED_OPERATOR(Name,Spelling,Token,Unary,Binary,MemberOnly) \ 515 case OO_##Name: return Spelling; 516#include "clang/Basic/OperatorKinds.def" 517 } 518 519 llvm_unreachable("Invalid OverloadedOperatorKind!"); 520} 521