IdentifierTable.cpp revision 1eb4433ac451dc16f4133a88af2d002ac26c58ef
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 <cstdio> 20 21using namespace clang; 22 23//===----------------------------------------------------------------------===// 24// IdentifierInfo Implementation 25//===----------------------------------------------------------------------===// 26 27IdentifierInfo::IdentifierInfo() { 28 TokenID = tok::identifier; 29 ObjCOrBuiltinID = 0; 30 HasMacro = false; 31 IsExtension = false; 32 IsPoisoned = false; 33 IsCPPOperatorKeyword = false; 34 NeedsHandleIdentifier = false; 35 FETokenInfo = 0; 36 Entry = 0; 37} 38 39//===----------------------------------------------------------------------===// 40// IdentifierTable Implementation 41//===----------------------------------------------------------------------===// 42 43IdentifierInfoLookup::~IdentifierInfoLookup() {} 44 45ExternalIdentifierLookup::~ExternalIdentifierLookup() {} 46 47IdentifierTable::IdentifierTable(const LangOptions &LangOpts, 48 IdentifierInfoLookup* externalLookup) 49 : HashTable(8192), // Start with space for 8K identifiers. 50 ExternalLookup(externalLookup) { 51 52 // Populate the identifier table with info about keywords for the current 53 // language. 54 AddKeywords(LangOpts); 55} 56 57//===----------------------------------------------------------------------===// 58// Language Keyword Implementation 59//===----------------------------------------------------------------------===// 60 61// Constants for TokenKinds.def 62namespace { 63 enum { 64 KEYALL = 1, 65 KEYC99 = 2, 66 KEYCXX = 4, 67 KEYCXX0X = 8, 68 KEYGNU = 16, 69 KEYMS = 32, 70 BOOLSUPPORT = 64 71 }; 72} 73 74/// AddKeyword - This method is used to associate a token ID with specific 75/// identifiers because they are language keywords. This causes the lexer to 76/// automatically map matching identifiers to specialized token codes. 77/// 78/// The C90/C99/CPP/CPP0x flags are set to 0 if the token should be 79/// enabled in the specified langauge, set to 1 if it is an extension 80/// in the specified language, and set to 2 if disabled in the 81/// specified language. 82static void AddKeyword(const char *Keyword, unsigned KWLen, 83 tok::TokenKind TokenCode, unsigned Flags, 84 const LangOptions &LangOpts, IdentifierTable &Table) { 85 unsigned AddResult = 0; 86 if (Flags & KEYALL) AddResult = 2; 87 else if (LangOpts.CPlusPlus && (Flags & KEYCXX)) AddResult = 2; 88 else if (LangOpts.CPlusPlus0x && (Flags & KEYCXX0X)) AddResult = 2; 89 else if (LangOpts.C99 && (Flags & KEYC99)) AddResult = 2; 90 else if (LangOpts.GNUMode && (Flags & KEYGNU)) AddResult = 1; 91 else if (LangOpts.Microsoft && (Flags & KEYMS)) AddResult = 1; 92 else if (LangOpts.Bool && (Flags & BOOLSUPPORT)) AddResult = 2; 93 94 // Don't add this keyword if disabled in this language. 95 if (AddResult == 0) return; 96 97 IdentifierInfo &Info = Table.get(Keyword, Keyword+KWLen); 98 Info.setTokenID(TokenCode); 99 Info.setIsExtensionToken(AddResult == 1); 100} 101 102/// AddCXXOperatorKeyword - Register a C++ operator keyword alternative 103/// representations. 104static void AddCXXOperatorKeyword(const char *Keyword, unsigned KWLen, 105 tok::TokenKind TokenCode, 106 IdentifierTable &Table) { 107 IdentifierInfo &Info = Table.get(Keyword, Keyword + KWLen); 108 Info.setTokenID(TokenCode); 109 Info.setIsCPlusPlusOperatorKeyword(); 110} 111 112/// AddObjCKeyword - Register an Objective-C @keyword like "class" "selector" or 113/// "property". 114static void AddObjCKeyword(tok::ObjCKeywordKind ObjCID, 115 const char *Name, unsigned NameLen, 116 IdentifierTable &Table) { 117 Table.get(Name, Name+NameLen).setObjCKeywordID(ObjCID); 118} 119 120/// AddKeywords - Add all keywords to the symbol table. 121/// 122void IdentifierTable::AddKeywords(const LangOptions &LangOpts) { 123 // Add keywords and tokens for the current language. 124#define KEYWORD(NAME, FLAGS) \ 125 AddKeyword(#NAME, strlen(#NAME), tok::kw_ ## NAME, \ 126 FLAGS, LangOpts, *this); 127#define ALIAS(NAME, TOK, FLAGS) \ 128 AddKeyword(NAME, strlen(NAME), tok::kw_ ## TOK, \ 129 FLAGS, LangOpts, *this); 130#define CXX_KEYWORD_OPERATOR(NAME, ALIAS) \ 131 if (LangOpts.CXXOperatorNames) \ 132 AddCXXOperatorKeyword(#NAME, strlen(#NAME), tok::ALIAS, *this); 133#define OBJC1_AT_KEYWORD(NAME) \ 134 if (LangOpts.ObjC1) \ 135 AddObjCKeyword(tok::objc_##NAME, #NAME, strlen(#NAME), *this); 136#define OBJC2_AT_KEYWORD(NAME) \ 137 if (LangOpts.ObjC2) \ 138 AddObjCKeyword(tok::objc_##NAME, #NAME, strlen(#NAME), *this); 139#include "clang/Basic/TokenKinds.def" 140} 141 142tok::PPKeywordKind IdentifierInfo::getPPKeywordID() const { 143 // We use a perfect hash function here involving the length of the keyword, 144 // the first and third character. For preprocessor ID's there are no 145 // collisions (if there were, the switch below would complain about duplicate 146 // case values). Note that this depends on 'if' being null terminated. 147 148#define HASH(LEN, FIRST, THIRD) \ 149 (LEN << 5) + (((FIRST-'a') + (THIRD-'a')) & 31) 150#define CASE(LEN, FIRST, THIRD, NAME) \ 151 case HASH(LEN, FIRST, THIRD): \ 152 return memcmp(Name, #NAME, LEN) ? tok::pp_not_keyword : tok::pp_ ## NAME 153 154 unsigned Len = getLength(); 155 if (Len < 2) return tok::pp_not_keyword; 156 const char *Name = getName(); 157 switch (HASH(Len, Name[0], Name[2])) { 158 default: return tok::pp_not_keyword; 159 CASE( 2, 'i', '\0', if); 160 CASE( 4, 'e', 'i', elif); 161 CASE( 4, 'e', 's', else); 162 CASE( 4, 'l', 'n', line); 163 CASE( 4, 's', 'c', sccs); 164 CASE( 5, 'e', 'd', endif); 165 CASE( 5, 'e', 'r', error); 166 CASE( 5, 'i', 'e', ident); 167 CASE( 5, 'i', 'd', ifdef); 168 CASE( 5, 'u', 'd', undef); 169 170 CASE( 6, 'a', 's', assert); 171 CASE( 6, 'd', 'f', define); 172 CASE( 6, 'i', 'n', ifndef); 173 CASE( 6, 'i', 'p', import); 174 CASE( 6, 'p', 'a', pragma); 175 176 CASE( 7, 'd', 'f', defined); 177 CASE( 7, 'i', 'c', include); 178 CASE( 7, 'w', 'r', warning); 179 180 CASE( 8, 'u', 'a', unassert); 181 CASE(12, 'i', 'c', include_next); 182 183 CASE(16, '_', 'i', __include_macros); 184#undef CASE 185#undef HASH 186 } 187} 188 189//===----------------------------------------------------------------------===// 190// Stats Implementation 191//===----------------------------------------------------------------------===// 192 193/// PrintStats - Print statistics about how well the identifier table is doing 194/// at hashing identifiers. 195void IdentifierTable::PrintStats() const { 196 unsigned NumBuckets = HashTable.getNumBuckets(); 197 unsigned NumIdentifiers = HashTable.getNumItems(); 198 unsigned NumEmptyBuckets = NumBuckets-NumIdentifiers; 199 unsigned AverageIdentifierSize = 0; 200 unsigned MaxIdentifierLength = 0; 201 202 // TODO: Figure out maximum times an identifier had to probe for -stats. 203 for (llvm::StringMap<IdentifierInfo*, llvm::BumpPtrAllocator>::const_iterator 204 I = HashTable.begin(), E = HashTable.end(); I != E; ++I) { 205 unsigned IdLen = I->getKeyLength(); 206 AverageIdentifierSize += IdLen; 207 if (MaxIdentifierLength < IdLen) 208 MaxIdentifierLength = IdLen; 209 } 210 211 fprintf(stderr, "\n*** Identifier Table Stats:\n"); 212 fprintf(stderr, "# Identifiers: %d\n", NumIdentifiers); 213 fprintf(stderr, "# Empty Buckets: %d\n", NumEmptyBuckets); 214 fprintf(stderr, "Hash density (#identifiers per bucket): %f\n", 215 NumIdentifiers/(double)NumBuckets); 216 fprintf(stderr, "Ave identifier length: %f\n", 217 (AverageIdentifierSize/(double)NumIdentifiers)); 218 fprintf(stderr, "Max identifier length: %d\n", MaxIdentifierLength); 219 220 // Compute statistics about the memory allocated for identifiers. 221 HashTable.getAllocator().PrintStats(); 222} 223 224//===----------------------------------------------------------------------===// 225// SelectorTable Implementation 226//===----------------------------------------------------------------------===// 227 228unsigned llvm::DenseMapInfo<clang::Selector>::getHashValue(clang::Selector S) { 229 return DenseMapInfo<void*>::getHashValue(S.getAsOpaquePtr()); 230} 231 232namespace clang { 233/// MultiKeywordSelector - One of these variable length records is kept for each 234/// selector containing more than one keyword. We use a folding set 235/// to unique aggregate names (keyword selectors in ObjC parlance). Access to 236/// this class is provided strictly through Selector. 237class MultiKeywordSelector 238 : public DeclarationNameExtra, public llvm::FoldingSetNode { 239 MultiKeywordSelector(unsigned nKeys) { 240 ExtraKindOrNumArgs = NUM_EXTRA_KINDS + nKeys; 241 } 242public: 243 // Constructor for keyword selectors. 244 MultiKeywordSelector(unsigned nKeys, IdentifierInfo **IIV) { 245 assert((nKeys > 1) && "not a multi-keyword selector"); 246 ExtraKindOrNumArgs = NUM_EXTRA_KINDS + nKeys; 247 248 // Fill in the trailing keyword array. 249 IdentifierInfo **KeyInfo = reinterpret_cast<IdentifierInfo **>(this+1); 250 for (unsigned i = 0; i != nKeys; ++i) 251 KeyInfo[i] = IIV[i]; 252 } 253 254 // getName - Derive the full selector name and return it. 255 std::string getName() const; 256 257 unsigned getNumArgs() const { return ExtraKindOrNumArgs - NUM_EXTRA_KINDS; } 258 259 typedef IdentifierInfo *const *keyword_iterator; 260 keyword_iterator keyword_begin() const { 261 return reinterpret_cast<keyword_iterator>(this+1); 262 } 263 keyword_iterator keyword_end() const { 264 return keyword_begin()+getNumArgs(); 265 } 266 IdentifierInfo *getIdentifierInfoForSlot(unsigned i) const { 267 assert(i < getNumArgs() && "getIdentifierInfoForSlot(): illegal index"); 268 return keyword_begin()[i]; 269 } 270 static void Profile(llvm::FoldingSetNodeID &ID, 271 keyword_iterator ArgTys, unsigned NumArgs) { 272 ID.AddInteger(NumArgs); 273 for (unsigned i = 0; i != NumArgs; ++i) 274 ID.AddPointer(ArgTys[i]); 275 } 276 void Profile(llvm::FoldingSetNodeID &ID) { 277 Profile(ID, keyword_begin(), getNumArgs()); 278 } 279}; 280} // end namespace clang. 281 282unsigned Selector::getNumArgs() const { 283 unsigned IIF = getIdentifierInfoFlag(); 284 if (IIF == ZeroArg) 285 return 0; 286 if (IIF == OneArg) 287 return 1; 288 // We point to a MultiKeywordSelector (pointer doesn't contain any flags). 289 MultiKeywordSelector *SI = reinterpret_cast<MultiKeywordSelector *>(InfoPtr); 290 return SI->getNumArgs(); 291} 292 293IdentifierInfo *Selector::getIdentifierInfoForSlot(unsigned argIndex) const { 294 if (getIdentifierInfoFlag()) { 295 assert(argIndex == 0 && "illegal keyword index"); 296 return getAsIdentifierInfo(); 297 } 298 // We point to a MultiKeywordSelector (pointer doesn't contain any flags). 299 MultiKeywordSelector *SI = reinterpret_cast<MultiKeywordSelector *>(InfoPtr); 300 return SI->getIdentifierInfoForSlot(argIndex); 301} 302 303std::string MultiKeywordSelector::getName() const { 304 std::string Result; 305 unsigned Length = 0; 306 for (keyword_iterator I = keyword_begin(), E = keyword_end(); I != E; ++I) { 307 if (*I) 308 Length += (*I)->getLength(); 309 ++Length; // : 310 } 311 312 Result.reserve(Length); 313 314 for (keyword_iterator I = keyword_begin(), E = keyword_end(); I != E; ++I) { 315 if (*I) 316 Result.insert(Result.end(), (*I)->getName(), 317 (*I)->getName()+(*I)->getLength()); 318 Result.push_back(':'); 319 } 320 321 return Result; 322} 323 324std::string Selector::getAsString() const { 325 if (InfoPtr == 0) 326 return "<null selector>"; 327 328 if (InfoPtr & ArgFlags) { 329 IdentifierInfo *II = getAsIdentifierInfo(); 330 331 // If the number of arguments is 0 then II is guaranteed to not be null. 332 if (getNumArgs() == 0) 333 return II->getName(); 334 335 std::string Res = II ? II->getName() : ""; 336 Res += ":"; 337 return Res; 338 } 339 340 // We have a multiple keyword selector (no embedded flags). 341 return reinterpret_cast<MultiKeywordSelector *>(InfoPtr)->getName(); 342} 343 344 345namespace { 346 struct SelectorTableImpl { 347 llvm::FoldingSet<MultiKeywordSelector> Table; 348 llvm::BumpPtrAllocator Allocator; 349 }; 350} // end anonymous namespace. 351 352static SelectorTableImpl &getSelectorTableImpl(void *P) { 353 return *static_cast<SelectorTableImpl*>(P); 354} 355 356 357Selector SelectorTable::getSelector(unsigned nKeys, IdentifierInfo **IIV) { 358 if (nKeys < 2) 359 return Selector(IIV[0], nKeys); 360 361 SelectorTableImpl &SelTabImpl = getSelectorTableImpl(Impl); 362 363 // Unique selector, to guarantee there is one per name. 364 llvm::FoldingSetNodeID ID; 365 MultiKeywordSelector::Profile(ID, IIV, nKeys); 366 367 void *InsertPos = 0; 368 if (MultiKeywordSelector *SI = 369 SelTabImpl.Table.FindNodeOrInsertPos(ID, InsertPos)) 370 return Selector(SI); 371 372 // MultiKeywordSelector objects are not allocated with new because they have a 373 // variable size array (for parameter types) at the end of them. 374 unsigned Size = sizeof(MultiKeywordSelector) + nKeys*sizeof(IdentifierInfo *); 375 MultiKeywordSelector *SI = 376 (MultiKeywordSelector*)SelTabImpl.Allocator.Allocate(Size, 377 llvm::alignof<MultiKeywordSelector>()); 378 new (SI) MultiKeywordSelector(nKeys, IIV); 379 SelTabImpl.Table.InsertNode(SI, InsertPos); 380 return Selector(SI); 381} 382 383SelectorTable::SelectorTable() { 384 Impl = new SelectorTableImpl(); 385} 386 387SelectorTable::~SelectorTable() { 388 delete &getSelectorTableImpl(Impl); 389} 390 391