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 "clang/Basic/CharInfo.h"
18#include "llvm/ADT/DenseMap.h"
19#include "llvm/ADT/FoldingSet.h"
20#include "llvm/Support/ErrorHandling.h"
21#include "llvm/Support/raw_ostream.h"
22#include <cstdio>
23
24using namespace clang;
25
26//===----------------------------------------------------------------------===//
27// IdentifierInfo Implementation
28//===----------------------------------------------------------------------===//
29
30IdentifierInfo::IdentifierInfo() {
31  TokenID = tok::identifier;
32  ObjCOrBuiltinID = 0;
33  HasMacro = false;
34  HadMacro = 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
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() {
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  // Add the '_experimental_modules_import' contextual keyword.
84  get("import").setModulesImport(true);
85}
86
87//===----------------------------------------------------------------------===//
88// Language Keyword Implementation
89//===----------------------------------------------------------------------===//
90
91// Constants for TokenKinds.def
92namespace {
93  enum {
94    KEYC99 = 0x1,
95    KEYCXX = 0x2,
96    KEYCXX11 = 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    KEYNOMS = 0x01000,
107    WCHARSUPPORT = 0x02000,
108    KEYALL = (0xffff & ~KEYNOMS) // Because KEYNOMS is used to exclude.
109  };
110}
111
112/// AddKeyword - This method is used to associate a token ID with specific
113/// identifiers because they are language keywords.  This causes the lexer to
114/// automatically map matching identifiers to specialized token codes.
115///
116/// The C90/C99/CPP/CPP0x flags are set to 3 if the token is a keyword in a
117/// future language standard, set to 2 if the token should be enabled in the
118/// specified language, set to 1 if it is an extension in the specified
119/// language, and set to 0 if disabled in the specified language.
120static void AddKeyword(StringRef Keyword,
121                       tok::TokenKind TokenCode, unsigned Flags,
122                       const LangOptions &LangOpts, IdentifierTable &Table) {
123  unsigned AddResult = 0;
124  if (Flags == KEYALL) AddResult = 2;
125  else if (LangOpts.CPlusPlus && (Flags & KEYCXX)) AddResult = 2;
126  else if (LangOpts.CPlusPlus11 && (Flags & KEYCXX11)) AddResult = 2;
127  else if (LangOpts.C99 && (Flags & KEYC99)) AddResult = 2;
128  else if (LangOpts.GNUKeywords && (Flags & KEYGNU)) AddResult = 1;
129  else if (LangOpts.MicrosoftExt && (Flags & KEYMS)) AddResult = 1;
130  else if (LangOpts.Borland && (Flags & KEYBORLAND)) AddResult = 1;
131  else if (LangOpts.Bool && (Flags & BOOLSUPPORT)) AddResult = 2;
132  else if (LangOpts.WChar && (Flags & WCHARSUPPORT)) AddResult = 2;
133  else if (LangOpts.AltiVec && (Flags & KEYALTIVEC)) AddResult = 2;
134  else if (LangOpts.OpenCL && (Flags & KEYOPENCL)) AddResult = 2;
135  else if (!LangOpts.CPlusPlus && (Flags & KEYNOCXX)) AddResult = 2;
136  else if (LangOpts.C11 && (Flags & KEYC11)) AddResult = 2;
137  // We treat bridge casts as objective-C keywords so we can warn on them
138  // in non-arc mode.
139  else if (LangOpts.ObjC2 && (Flags & KEYARC)) AddResult = 2;
140  else if (LangOpts.CPlusPlus && (Flags & KEYCXX11)) AddResult = 3;
141
142  // Don't add this keyword under MicrosoftMode.
143  if (LangOpts.MicrosoftMode && (Flags & KEYNOMS))
144     return;
145  // Don't add this keyword if disabled in this language.
146  if (AddResult == 0) return;
147
148  IdentifierInfo &Info =
149      Table.get(Keyword, AddResult == 3 ? tok::identifier : TokenCode);
150  Info.setIsExtensionToken(AddResult == 1);
151  Info.setIsCXX11CompatKeyword(AddResult == 3);
152}
153
154/// AddCXXOperatorKeyword - Register a C++ operator keyword alternative
155/// representations.
156static void AddCXXOperatorKeyword(StringRef Keyword,
157                                  tok::TokenKind TokenCode,
158                                  IdentifierTable &Table) {
159  IdentifierInfo &Info = Table.get(Keyword, TokenCode);
160  Info.setIsCPlusPlusOperatorKeyword();
161}
162
163/// AddObjCKeyword - Register an Objective-C \@keyword like "class" "selector"
164/// or "property".
165static void AddObjCKeyword(StringRef Name,
166                           tok::ObjCKeywordKind ObjCID,
167                           IdentifierTable &Table) {
168  Table.get(Name).setObjCKeywordID(ObjCID);
169}
170
171/// AddKeywords - Add all keywords to the symbol table.
172///
173void IdentifierTable::AddKeywords(const LangOptions &LangOpts) {
174  // Add keywords and tokens for the current language.
175#define KEYWORD(NAME, FLAGS) \
176  AddKeyword(StringRef(#NAME), tok::kw_ ## NAME,  \
177             FLAGS, LangOpts, *this);
178#define ALIAS(NAME, TOK, FLAGS) \
179  AddKeyword(StringRef(NAME), tok::kw_ ## TOK,  \
180             FLAGS, LangOpts, *this);
181#define CXX_KEYWORD_OPERATOR(NAME, ALIAS) \
182  if (LangOpts.CXXOperatorNames)          \
183    AddCXXOperatorKeyword(StringRef(#NAME), tok::ALIAS, *this);
184#define OBJC1_AT_KEYWORD(NAME) \
185  if (LangOpts.ObjC1)          \
186    AddObjCKeyword(StringRef(#NAME), tok::objc_##NAME, *this);
187#define OBJC2_AT_KEYWORD(NAME) \
188  if (LangOpts.ObjC2)          \
189    AddObjCKeyword(StringRef(#NAME), tok::objc_##NAME, *this);
190#define TESTING_KEYWORD(NAME, FLAGS)
191#include "clang/Basic/TokenKinds.def"
192
193  if (LangOpts.ParseUnknownAnytype)
194    AddKeyword("__unknown_anytype", tok::kw___unknown_anytype, KEYALL,
195               LangOpts, *this);
196}
197
198tok::PPKeywordKind IdentifierInfo::getPPKeywordID() const {
199  // We use a perfect hash function here involving the length of the keyword,
200  // the first and third character.  For preprocessor ID's there are no
201  // collisions (if there were, the switch below would complain about duplicate
202  // case values).  Note that this depends on 'if' being null terminated.
203
204#define HASH(LEN, FIRST, THIRD) \
205  (LEN << 5) + (((FIRST-'a') + (THIRD-'a')) & 31)
206#define CASE(LEN, FIRST, THIRD, NAME) \
207  case HASH(LEN, FIRST, THIRD): \
208    return memcmp(Name, #NAME, LEN) ? tok::pp_not_keyword : tok::pp_ ## NAME
209
210  unsigned Len = getLength();
211  if (Len < 2) return tok::pp_not_keyword;
212  const char *Name = getNameStart();
213  switch (HASH(Len, Name[0], Name[2])) {
214  default: return tok::pp_not_keyword;
215  CASE( 2, 'i', '\0', if);
216  CASE( 4, 'e', 'i', elif);
217  CASE( 4, 'e', 's', else);
218  CASE( 4, 'l', 'n', line);
219  CASE( 4, 's', 'c', sccs);
220  CASE( 5, 'e', 'd', endif);
221  CASE( 5, 'e', 'r', error);
222  CASE( 5, 'i', 'e', ident);
223  CASE( 5, 'i', 'd', ifdef);
224  CASE( 5, 'u', 'd', undef);
225
226  CASE( 6, 'a', 's', assert);
227  CASE( 6, 'd', 'f', define);
228  CASE( 6, 'i', 'n', ifndef);
229  CASE( 6, 'i', 'p', import);
230  CASE( 6, 'p', 'a', pragma);
231
232  CASE( 7, 'd', 'f', defined);
233  CASE( 7, 'i', 'c', include);
234  CASE( 7, 'w', 'r', warning);
235
236  CASE( 8, 'u', 'a', unassert);
237  CASE(12, 'i', 'c', include_next);
238
239  CASE(14, '_', 'p', __public_macro);
240
241  CASE(15, '_', 'p', __private_macro);
242
243  CASE(16, '_', 'i', __include_macros);
244#undef CASE
245#undef HASH
246  }
247}
248
249//===----------------------------------------------------------------------===//
250// Stats Implementation
251//===----------------------------------------------------------------------===//
252
253/// PrintStats - Print statistics about how well the identifier table is doing
254/// at hashing identifiers.
255void IdentifierTable::PrintStats() const {
256  unsigned NumBuckets = HashTable.getNumBuckets();
257  unsigned NumIdentifiers = HashTable.getNumItems();
258  unsigned NumEmptyBuckets = NumBuckets-NumIdentifiers;
259  unsigned AverageIdentifierSize = 0;
260  unsigned MaxIdentifierLength = 0;
261
262  // TODO: Figure out maximum times an identifier had to probe for -stats.
263  for (llvm::StringMap<IdentifierInfo*, llvm::BumpPtrAllocator>::const_iterator
264       I = HashTable.begin(), E = HashTable.end(); I != E; ++I) {
265    unsigned IdLen = I->getKeyLength();
266    AverageIdentifierSize += IdLen;
267    if (MaxIdentifierLength < IdLen)
268      MaxIdentifierLength = IdLen;
269  }
270
271  fprintf(stderr, "\n*** Identifier Table Stats:\n");
272  fprintf(stderr, "# Identifiers:   %d\n", NumIdentifiers);
273  fprintf(stderr, "# Empty Buckets: %d\n", NumEmptyBuckets);
274  fprintf(stderr, "Hash density (#identifiers per bucket): %f\n",
275          NumIdentifiers/(double)NumBuckets);
276  fprintf(stderr, "Ave identifier length: %f\n",
277          (AverageIdentifierSize/(double)NumIdentifiers));
278  fprintf(stderr, "Max identifier length: %d\n", MaxIdentifierLength);
279
280  // Compute statistics about the memory allocated for identifiers.
281  HashTable.getAllocator().PrintStats();
282}
283
284//===----------------------------------------------------------------------===//
285// SelectorTable Implementation
286//===----------------------------------------------------------------------===//
287
288unsigned llvm::DenseMapInfo<clang::Selector>::getHashValue(clang::Selector S) {
289  return DenseMapInfo<void*>::getHashValue(S.getAsOpaquePtr());
290}
291
292namespace clang {
293/// MultiKeywordSelector - One of these variable length records is kept for each
294/// selector containing more than one keyword. We use a folding set
295/// to unique aggregate names (keyword selectors in ObjC parlance). Access to
296/// this class is provided strictly through Selector.
297class MultiKeywordSelector
298  : public DeclarationNameExtra, public llvm::FoldingSetNode {
299  MultiKeywordSelector(unsigned nKeys) {
300    ExtraKindOrNumArgs = NUM_EXTRA_KINDS + nKeys;
301  }
302public:
303  // Constructor for keyword selectors.
304  MultiKeywordSelector(unsigned nKeys, IdentifierInfo **IIV) {
305    assert((nKeys > 1) && "not a multi-keyword selector");
306    ExtraKindOrNumArgs = NUM_EXTRA_KINDS + nKeys;
307
308    // Fill in the trailing keyword array.
309    IdentifierInfo **KeyInfo = reinterpret_cast<IdentifierInfo **>(this+1);
310    for (unsigned i = 0; i != nKeys; ++i)
311      KeyInfo[i] = IIV[i];
312  }
313
314  // getName - Derive the full selector name and return it.
315  std::string getName() const;
316
317  unsigned getNumArgs() const { return ExtraKindOrNumArgs - NUM_EXTRA_KINDS; }
318
319  typedef IdentifierInfo *const *keyword_iterator;
320  keyword_iterator keyword_begin() const {
321    return reinterpret_cast<keyword_iterator>(this+1);
322  }
323  keyword_iterator keyword_end() const {
324    return keyword_begin()+getNumArgs();
325  }
326  IdentifierInfo *getIdentifierInfoForSlot(unsigned i) const {
327    assert(i < getNumArgs() && "getIdentifierInfoForSlot(): illegal index");
328    return keyword_begin()[i];
329  }
330  static void Profile(llvm::FoldingSetNodeID &ID,
331                      keyword_iterator ArgTys, unsigned NumArgs) {
332    ID.AddInteger(NumArgs);
333    for (unsigned i = 0; i != NumArgs; ++i)
334      ID.AddPointer(ArgTys[i]);
335  }
336  void Profile(llvm::FoldingSetNodeID &ID) {
337    Profile(ID, keyword_begin(), getNumArgs());
338  }
339};
340} // end namespace clang.
341
342unsigned Selector::getNumArgs() const {
343  unsigned IIF = getIdentifierInfoFlag();
344  if (IIF <= ZeroArg)
345    return 0;
346  if (IIF == OneArg)
347    return 1;
348  // We point to a MultiKeywordSelector.
349  MultiKeywordSelector *SI = getMultiKeywordSelector();
350  return SI->getNumArgs();
351}
352
353IdentifierInfo *Selector::getIdentifierInfoForSlot(unsigned argIndex) const {
354  if (getIdentifierInfoFlag() < MultiArg) {
355    assert(argIndex == 0 && "illegal keyword index");
356    return getAsIdentifierInfo();
357  }
358  // We point to a MultiKeywordSelector.
359  MultiKeywordSelector *SI = getMultiKeywordSelector();
360  return SI->getIdentifierInfoForSlot(argIndex);
361}
362
363StringRef Selector::getNameForSlot(unsigned int argIndex) const {
364  IdentifierInfo *II = getIdentifierInfoForSlot(argIndex);
365  return II? II->getName() : StringRef();
366}
367
368std::string MultiKeywordSelector::getName() const {
369  SmallString<256> Str;
370  llvm::raw_svector_ostream OS(Str);
371  for (keyword_iterator I = keyword_begin(), E = keyword_end(); I != E; ++I) {
372    if (*I)
373      OS << (*I)->getName();
374    OS << ':';
375  }
376
377  return OS.str();
378}
379
380std::string Selector::getAsString() const {
381  if (InfoPtr == 0)
382    return "<null selector>";
383
384  if (getIdentifierInfoFlag() < MultiArg) {
385    IdentifierInfo *II = getAsIdentifierInfo();
386
387    // If the number of arguments is 0 then II is guaranteed to not be null.
388    if (getNumArgs() == 0)
389      return II->getName();
390
391    if (!II)
392      return ":";
393
394    return II->getName().str() + ":";
395  }
396
397  // We have a multiple keyword selector.
398  return getMultiKeywordSelector()->getName();
399}
400
401/// Interpreting the given string using the normal CamelCase
402/// conventions, determine whether the given string starts with the
403/// given "word", which is assumed to end in a lowercase letter.
404static bool startsWithWord(StringRef name, StringRef word) {
405  if (name.size() < word.size()) return false;
406  return ((name.size() == word.size() || !isLowercase(name[word.size()])) &&
407          name.startswith(word));
408}
409
410ObjCMethodFamily Selector::getMethodFamilyImpl(Selector sel) {
411  IdentifierInfo *first = sel.getIdentifierInfoForSlot(0);
412  if (!first) return OMF_None;
413
414  StringRef name = first->getName();
415  if (sel.isUnarySelector()) {
416    if (name == "autorelease") return OMF_autorelease;
417    if (name == "dealloc") return OMF_dealloc;
418    if (name == "finalize") return OMF_finalize;
419    if (name == "release") return OMF_release;
420    if (name == "retain") return OMF_retain;
421    if (name == "retainCount") return OMF_retainCount;
422    if (name == "self") return OMF_self;
423  }
424
425  if (name == "performSelector") return OMF_performSelector;
426
427  // The other method families may begin with a prefix of underscores.
428  while (!name.empty() && name.front() == '_')
429    name = name.substr(1);
430
431  if (name.empty()) return OMF_None;
432  switch (name.front()) {
433  case 'a':
434    if (startsWithWord(name, "alloc")) return OMF_alloc;
435    break;
436  case 'c':
437    if (startsWithWord(name, "copy")) return OMF_copy;
438    break;
439  case 'i':
440    if (startsWithWord(name, "init")) return OMF_init;
441    break;
442  case 'm':
443    if (startsWithWord(name, "mutableCopy")) return OMF_mutableCopy;
444    break;
445  case 'n':
446    if (startsWithWord(name, "new")) return OMF_new;
447    break;
448  default:
449    break;
450  }
451
452  return OMF_None;
453}
454
455ObjCInstanceTypeFamily Selector::getInstTypeMethodFamily(Selector sel) {
456  IdentifierInfo *first = sel.getIdentifierInfoForSlot(0);
457  if (!first) return OIT_None;
458
459  StringRef name = first->getName();
460
461  if (name.empty()) return OIT_None;
462  switch (name.front()) {
463    case 'a':
464      if (startsWithWord(name, "alloc")) return OIT_MemManage;
465      else
466        if (startsWithWord(name, "array")) return OIT_Array;
467      break;
468    case 'd':
469      if (startsWithWord(name, "dictionary")) return OIT_Dictionary;
470      if (startsWithWord(name, "default")) return OIT_Singleton;
471      break;
472    case 'i':
473      if (startsWithWord(name, "init")) return OIT_MemManage;
474      break;
475    case 'r':
476      if (startsWithWord(name, "retain")) return OIT_MemManage;
477      break;
478    case 's':
479      if (startsWithWord(name, "shared") ||
480          startsWithWord(name, "standard"))
481        return OIT_Singleton;
482    default:
483      break;
484  }
485  return OIT_None;
486}
487
488namespace {
489  struct SelectorTableImpl {
490    llvm::FoldingSet<MultiKeywordSelector> Table;
491    llvm::BumpPtrAllocator Allocator;
492  };
493} // end anonymous namespace.
494
495static SelectorTableImpl &getSelectorTableImpl(void *P) {
496  return *static_cast<SelectorTableImpl*>(P);
497}
498
499SmallString<64>
500SelectorTable::constructSetterName(StringRef Name) {
501  SmallString<64> SetterName("set");
502  SetterName += Name;
503  SetterName[3] = toUppercase(SetterName[3]);
504  return SetterName;
505}
506
507Selector
508SelectorTable::constructSetterSelector(IdentifierTable &Idents,
509                                       SelectorTable &SelTable,
510                                       const IdentifierInfo *Name) {
511  IdentifierInfo *SetterName =
512    &Idents.get(constructSetterName(Name->getName()));
513  return SelTable.getUnarySelector(SetterName);
514}
515
516size_t SelectorTable::getTotalMemory() const {
517  SelectorTableImpl &SelTabImpl = getSelectorTableImpl(Impl);
518  return SelTabImpl.Allocator.getTotalMemory();
519}
520
521Selector SelectorTable::getSelector(unsigned nKeys, IdentifierInfo **IIV) {
522  if (nKeys < 2)
523    return Selector(IIV[0], nKeys);
524
525  SelectorTableImpl &SelTabImpl = getSelectorTableImpl(Impl);
526
527  // Unique selector, to guarantee there is one per name.
528  llvm::FoldingSetNodeID ID;
529  MultiKeywordSelector::Profile(ID, IIV, nKeys);
530
531  void *InsertPos = 0;
532  if (MultiKeywordSelector *SI =
533        SelTabImpl.Table.FindNodeOrInsertPos(ID, InsertPos))
534    return Selector(SI);
535
536  // MultiKeywordSelector objects are not allocated with new because they have a
537  // variable size array (for parameter types) at the end of them.
538  unsigned Size = sizeof(MultiKeywordSelector) + nKeys*sizeof(IdentifierInfo *);
539  MultiKeywordSelector *SI =
540    (MultiKeywordSelector*)SelTabImpl.Allocator.Allocate(Size,
541                                         llvm::alignOf<MultiKeywordSelector>());
542  new (SI) MultiKeywordSelector(nKeys, IIV);
543  SelTabImpl.Table.InsertNode(SI, InsertPos);
544  return Selector(SI);
545}
546
547SelectorTable::SelectorTable() {
548  Impl = new SelectorTableImpl();
549}
550
551SelectorTable::~SelectorTable() {
552  delete &getSelectorTableImpl(Impl);
553}
554
555const char *clang::getOperatorSpelling(OverloadedOperatorKind Operator) {
556  switch (Operator) {
557  case OO_None:
558  case NUM_OVERLOADED_OPERATORS:
559    return 0;
560
561#define OVERLOADED_OPERATOR(Name,Spelling,Token,Unary,Binary,MemberOnly) \
562  case OO_##Name: return Spelling;
563#include "clang/Basic/OperatorKinds.def"
564  }
565
566  llvm_unreachable("Invalid OverloadedOperatorKind!");
567}
568