• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 //===--- IdentifierTable.h - Hash table for identifier lookup ---*- C++ -*-===//
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 /// \file
11 /// \brief Defines the clang::IdentifierInfo, clang::IdentifierTable, and
12 /// clang::Selector interfaces.
13 ///
14 //===----------------------------------------------------------------------===//
15 
16 #ifndef LLVM_CLANG_BASIC_IDENTIFIERTABLE_H
17 #define LLVM_CLANG_BASIC_IDENTIFIERTABLE_H
18 
19 #include "clang/Basic/LLVM.h"
20 #include "clang/Basic/TokenKinds.h"
21 #include "llvm/ADT/StringMap.h"
22 #include "llvm/ADT/StringRef.h"
23 #include <cassert>
24 #include <string>
25 
26 namespace llvm {
27   template <typename T> struct DenseMapInfo;
28 }
29 
30 namespace clang {
31   class LangOptions;
32   class IdentifierInfo;
33   class IdentifierTable;
34   class SourceLocation;
35   class MultiKeywordSelector; // private class used by Selector
36   class DeclarationName;      // AST class that stores declaration names
37 
38   /// \brief A simple pair of identifier info and location.
39   typedef std::pair<IdentifierInfo*, SourceLocation> IdentifierLocPair;
40 
41 
42 /// One of these records is kept for each identifier that
43 /// is lexed.  This contains information about whether the token was \#define'd,
44 /// is a language keyword, or if it is a front-end token of some sort (e.g. a
45 /// variable or function name).  The preprocessor keeps this information in a
46 /// set, and all tok::identifier tokens have a pointer to one of these.
47 class IdentifierInfo {
48   unsigned TokenID            : 9; // Front-end token ID or tok::identifier.
49   // Objective-C keyword ('protocol' in '@protocol') or builtin (__builtin_inf).
50   // First NUM_OBJC_KEYWORDS values are for Objective-C, the remaining values
51   // are for builtins.
52   unsigned ObjCOrBuiltinID    :11;
53   bool HasMacro               : 1; // True if there is a #define for this.
54   bool HadMacro               : 1; // True if there was a #define for this.
55   bool IsExtension            : 1; // True if identifier is a lang extension.
56   bool IsCXX11CompatKeyword   : 1; // True if identifier is a keyword in C++11.
57   bool IsPoisoned             : 1; // True if identifier is poisoned.
58   bool IsCPPOperatorKeyword   : 1; // True if ident is a C++ operator keyword.
59   bool NeedsHandleIdentifier  : 1; // See "RecomputeNeedsHandleIdentifier".
60   bool IsFromAST              : 1; // True if identifier was loaded (at least
61                                    // partially) from an AST file.
62   bool ChangedAfterLoad       : 1; // True if identifier has changed from the
63                                    // definition loaded from an AST file.
64   bool RevertedTokenID        : 1; // True if RevertTokenIDToIdentifier was
65                                    // called.
66   bool OutOfDate              : 1; // True if there may be additional
67                                    // information about this identifier
68                                    // stored externally.
69   bool IsModulesImport        : 1; // True if this is the 'import' contextual
70                                    // keyword.
71   // 32-bit word is filled.
72 
73   void *FETokenInfo;               // Managed by the language front-end.
74   llvm::StringMapEntry<IdentifierInfo*> *Entry;
75 
76   IdentifierInfo(const IdentifierInfo&) LLVM_DELETED_FUNCTION;
77   void operator=(const IdentifierInfo&) LLVM_DELETED_FUNCTION;
78 
79   friend class IdentifierTable;
80 
81 public:
82   IdentifierInfo();
83 
84 
85   /// \brief Return true if this is the identifier for the specified string.
86   ///
87   /// This is intended to be used for string literals only: II->isStr("foo").
88   template <std::size_t StrLen>
isStr(const char (& Str)[StrLen])89   bool isStr(const char (&Str)[StrLen]) const {
90     return getLength() == StrLen-1 && !memcmp(getNameStart(), Str, StrLen-1);
91   }
92 
93   /// \brief Return the beginning of the actual null-terminated string for this
94   /// identifier.
95   ///
getNameStart()96   const char *getNameStart() const {
97     if (Entry) return Entry->getKeyData();
98     // FIXME: This is gross. It would be best not to embed specific details
99     // of the PTH file format here.
100     // The 'this' pointer really points to a
101     // std::pair<IdentifierInfo, const char*>, where internal pointer
102     // points to the external string data.
103     typedef std::pair<IdentifierInfo, const char*> actualtype;
104     return ((const actualtype*) this)->second;
105   }
106 
107   /// \brief Efficiently return the length of this identifier info.
108   ///
getLength()109   unsigned getLength() const {
110     if (Entry) return Entry->getKeyLength();
111     // FIXME: This is gross. It would be best not to embed specific details
112     // of the PTH file format here.
113     // The 'this' pointer really points to a
114     // std::pair<IdentifierInfo, const char*>, where internal pointer
115     // points to the external string data.
116     typedef std::pair<IdentifierInfo, const char*> actualtype;
117     const char* p = ((const actualtype*) this)->second - 2;
118     return (((unsigned) p[0]) | (((unsigned) p[1]) << 8)) - 1;
119   }
120 
121   /// \brief Return the actual identifier string.
getName()122   StringRef getName() const {
123     return StringRef(getNameStart(), getLength());
124   }
125 
126   /// \brief Return true if this identifier is \#defined to some other value.
hasMacroDefinition()127   bool hasMacroDefinition() const {
128     return HasMacro;
129   }
setHasMacroDefinition(bool Val)130   void setHasMacroDefinition(bool Val) {
131     if (HasMacro == Val) return;
132 
133     HasMacro = Val;
134     if (Val) {
135       NeedsHandleIdentifier = 1;
136       HadMacro = true;
137     } else {
138       RecomputeNeedsHandleIdentifier();
139     }
140   }
141   /// \brief Returns true if this identifier was \#defined to some value at any
142   /// moment. In this case there should be an entry for the identifier in the
143   /// macro history table in Preprocessor.
hadMacroDefinition()144   bool hadMacroDefinition() const {
145     return HadMacro;
146   }
147 
148   /// If this is a source-language token (e.g. 'for'), this API
149   /// can be used to cause the lexer to map identifiers to source-language
150   /// tokens.
getTokenID()151   tok::TokenKind getTokenID() const { return (tok::TokenKind)TokenID; }
152 
153   /// \brief True if RevertTokenIDToIdentifier() was called.
hasRevertedTokenIDToIdentifier()154   bool hasRevertedTokenIDToIdentifier() const { return RevertedTokenID; }
155 
156   /// \brief Revert TokenID to tok::identifier; used for GNU libstdc++ 4.2
157   /// compatibility.
158   ///
159   /// TokenID is normally read-only but there are 2 instances where we revert it
160   /// to tok::identifier for libstdc++ 4.2. Keep track of when this happens
161   /// using this method so we can inform serialization about it.
RevertTokenIDToIdentifier()162   void RevertTokenIDToIdentifier() {
163     assert(TokenID != tok::identifier && "Already at tok::identifier");
164     TokenID = tok::identifier;
165     RevertedTokenID = true;
166   }
167 
168   /// \brief Return the preprocessor keyword ID for this identifier.
169   ///
170   /// For example, "define" will return tok::pp_define.
171   tok::PPKeywordKind getPPKeywordID() const;
172 
173   /// \brief Return the Objective-C keyword ID for the this identifier.
174   ///
175   /// For example, 'class' will return tok::objc_class if ObjC is enabled.
getObjCKeywordID()176   tok::ObjCKeywordKind getObjCKeywordID() const {
177     if (ObjCOrBuiltinID < tok::NUM_OBJC_KEYWORDS)
178       return tok::ObjCKeywordKind(ObjCOrBuiltinID);
179     else
180       return tok::objc_not_keyword;
181   }
setObjCKeywordID(tok::ObjCKeywordKind ID)182   void setObjCKeywordID(tok::ObjCKeywordKind ID) { ObjCOrBuiltinID = ID; }
183 
184   /// \brief Return a value indicating whether this is a builtin function.
185   ///
186   /// 0 is not-built-in.  1 is builtin-for-some-nonprimary-target.
187   /// 2+ are specific builtin functions.
getBuiltinID()188   unsigned getBuiltinID() const {
189     if (ObjCOrBuiltinID >= tok::NUM_OBJC_KEYWORDS)
190       return ObjCOrBuiltinID - tok::NUM_OBJC_KEYWORDS;
191     else
192       return 0;
193   }
setBuiltinID(unsigned ID)194   void setBuiltinID(unsigned ID) {
195     ObjCOrBuiltinID = ID + tok::NUM_OBJC_KEYWORDS;
196     assert(ObjCOrBuiltinID - unsigned(tok::NUM_OBJC_KEYWORDS) == ID
197            && "ID too large for field!");
198   }
199 
getObjCOrBuiltinID()200   unsigned getObjCOrBuiltinID() const { return ObjCOrBuiltinID; }
setObjCOrBuiltinID(unsigned ID)201   void setObjCOrBuiltinID(unsigned ID) { ObjCOrBuiltinID = ID; }
202 
203   /// get/setExtension - Initialize information about whether or not this
204   /// language token is an extension.  This controls extension warnings, and is
205   /// only valid if a custom token ID is set.
isExtensionToken()206   bool isExtensionToken() const { return IsExtension; }
setIsExtensionToken(bool Val)207   void setIsExtensionToken(bool Val) {
208     IsExtension = Val;
209     if (Val)
210       NeedsHandleIdentifier = 1;
211     else
212       RecomputeNeedsHandleIdentifier();
213   }
214 
215   /// is/setIsCXX11CompatKeyword - Initialize information about whether or not
216   /// this language token is a keyword in C++11. This controls compatibility
217   /// warnings, and is only true when not parsing C++11. Once a compatibility
218   /// problem has been diagnosed with this keyword, the flag will be cleared.
isCXX11CompatKeyword()219   bool isCXX11CompatKeyword() const { return IsCXX11CompatKeyword; }
setIsCXX11CompatKeyword(bool Val)220   void setIsCXX11CompatKeyword(bool Val) {
221     IsCXX11CompatKeyword = Val;
222     if (Val)
223       NeedsHandleIdentifier = 1;
224     else
225       RecomputeNeedsHandleIdentifier();
226   }
227 
228   /// setIsPoisoned - Mark this identifier as poisoned.  After poisoning, the
229   /// Preprocessor will emit an error every time this token is used.
230   void setIsPoisoned(bool Value = true) {
231     IsPoisoned = Value;
232     if (Value)
233       NeedsHandleIdentifier = 1;
234     else
235       RecomputeNeedsHandleIdentifier();
236   }
237 
238   /// \brief Return true if this token has been poisoned.
isPoisoned()239   bool isPoisoned() const { return IsPoisoned; }
240 
241   /// isCPlusPlusOperatorKeyword/setIsCPlusPlusOperatorKeyword controls whether
242   /// this identifier is a C++ alternate representation of an operator.
243   void setIsCPlusPlusOperatorKeyword(bool Val = true) {
244     IsCPPOperatorKeyword = Val;
245     if (Val)
246       NeedsHandleIdentifier = 1;
247     else
248       RecomputeNeedsHandleIdentifier();
249   }
isCPlusPlusOperatorKeyword()250   bool isCPlusPlusOperatorKeyword() const { return IsCPPOperatorKeyword; }
251 
252   /// getFETokenInfo/setFETokenInfo - The language front-end is allowed to
253   /// associate arbitrary metadata with this token.
254   template<typename T>
getFETokenInfo()255   T *getFETokenInfo() const { return static_cast<T*>(FETokenInfo); }
setFETokenInfo(void * T)256   void setFETokenInfo(void *T) { FETokenInfo = T; }
257 
258   /// \brief Return true if the Preprocessor::HandleIdentifier must be called
259   /// on a token of this identifier.
260   ///
261   /// If this returns false, we know that HandleIdentifier will not affect
262   /// the token.
isHandleIdentifierCase()263   bool isHandleIdentifierCase() const { return NeedsHandleIdentifier; }
264 
265   /// \brief Return true if the identifier in its current state was loaded
266   /// from an AST file.
isFromAST()267   bool isFromAST() const { return IsFromAST; }
268 
setIsFromAST()269   void setIsFromAST() { IsFromAST = true; }
270 
271   /// \brief Determine whether this identifier has changed since it was loaded
272   /// from an AST file.
hasChangedSinceDeserialization()273   bool hasChangedSinceDeserialization() const {
274     return ChangedAfterLoad;
275   }
276 
277   /// \brief Note that this identifier has changed since it was loaded from
278   /// an AST file.
setChangedSinceDeserialization()279   void setChangedSinceDeserialization() {
280     ChangedAfterLoad = true;
281   }
282 
283   /// \brief Determine whether the information for this identifier is out of
284   /// date with respect to the external source.
isOutOfDate()285   bool isOutOfDate() const { return OutOfDate; }
286 
287   /// \brief Set whether the information for this identifier is out of
288   /// date with respect to the external source.
setOutOfDate(bool OOD)289   void setOutOfDate(bool OOD) {
290     OutOfDate = OOD;
291     if (OOD)
292       NeedsHandleIdentifier = true;
293     else
294       RecomputeNeedsHandleIdentifier();
295   }
296 
297   /// \brief Determine whether this is the contextual keyword \c import.
isModulesImport()298   bool isModulesImport() const { return IsModulesImport; }
299 
300   /// \brief Set whether this identifier is the contextual keyword \c import.
setModulesImport(bool I)301   void setModulesImport(bool I) {
302     IsModulesImport = I;
303     if (I)
304       NeedsHandleIdentifier = true;
305     else
306       RecomputeNeedsHandleIdentifier();
307   }
308 
309 private:
310   /// The Preprocessor::HandleIdentifier does several special (but rare)
311   /// things to identifiers of various sorts.  For example, it changes the
312   /// \c for keyword token from tok::identifier to tok::for.
313   ///
314   /// This method is very tied to the definition of HandleIdentifier.  Any
315   /// change to it should be reflected here.
RecomputeNeedsHandleIdentifier()316   void RecomputeNeedsHandleIdentifier() {
317     NeedsHandleIdentifier =
318       (isPoisoned() | hasMacroDefinition() | isCPlusPlusOperatorKeyword() |
319        isExtensionToken() | isCXX11CompatKeyword() || isOutOfDate() ||
320        isModulesImport());
321   }
322 };
323 
324 /// \brief An RAII object for [un]poisoning an identifier within a scope.
325 ///
326 /// \p II is allowed to be null, in which case objects of this type have
327 /// no effect.
328 class PoisonIdentifierRAIIObject {
329   IdentifierInfo *const II;
330   const bool OldValue;
331 public:
PoisonIdentifierRAIIObject(IdentifierInfo * II,bool NewValue)332   PoisonIdentifierRAIIObject(IdentifierInfo *II, bool NewValue)
333     : II(II), OldValue(II ? II->isPoisoned() : false) {
334     if(II)
335       II->setIsPoisoned(NewValue);
336   }
337 
~PoisonIdentifierRAIIObject()338   ~PoisonIdentifierRAIIObject() {
339     if(II)
340       II->setIsPoisoned(OldValue);
341   }
342 };
343 
344 /// \brief An iterator that walks over all of the known identifiers
345 /// in the lookup table.
346 ///
347 /// Since this iterator uses an abstract interface via virtual
348 /// functions, it uses an object-oriented interface rather than the
349 /// more standard C++ STL iterator interface. In this OO-style
350 /// iteration, the single function \c Next() provides dereference,
351 /// advance, and end-of-sequence checking in a single
352 /// operation. Subclasses of this iterator type will provide the
353 /// actual functionality.
354 class IdentifierIterator {
355 private:
356   IdentifierIterator(const IdentifierIterator &) LLVM_DELETED_FUNCTION;
357   void operator=(const IdentifierIterator &) LLVM_DELETED_FUNCTION;
358 
359 protected:
IdentifierIterator()360   IdentifierIterator() { }
361 
362 public:
363   virtual ~IdentifierIterator();
364 
365   /// \brief Retrieve the next string in the identifier table and
366   /// advances the iterator for the following string.
367   ///
368   /// \returns The next string in the identifier table. If there is
369   /// no such string, returns an empty \c StringRef.
370   virtual StringRef Next() = 0;
371 };
372 
373 /// \brief Provides lookups to, and iteration over, IdentiferInfo objects.
374 class IdentifierInfoLookup {
375 public:
376   virtual ~IdentifierInfoLookup();
377 
378   /// \brief Return the IdentifierInfo for the specified named identifier.
379   ///
380   /// Unlike the version in IdentifierTable, this returns a pointer instead
381   /// of a reference.  If the pointer is null then the IdentifierInfo cannot
382   /// be found.
383   virtual IdentifierInfo* get(StringRef Name) = 0;
384 
385   /// \brief Retrieve an iterator into the set of all identifiers
386   /// known to this identifier lookup source.
387   ///
388   /// This routine provides access to all of the identifiers known to
389   /// the identifier lookup, allowing access to the contents of the
390   /// identifiers without introducing the overhead of constructing
391   /// IdentifierInfo objects for each.
392   ///
393   /// \returns A new iterator into the set of known identifiers. The
394   /// caller is responsible for deleting this iterator.
395   virtual IdentifierIterator *getIdentifiers();
396 };
397 
398 /// \brief An abstract class used to resolve numerical identifier
399 /// references (meaningful only to some external source) into
400 /// IdentifierInfo pointers.
401 class ExternalIdentifierLookup {
402 public:
403   virtual ~ExternalIdentifierLookup();
404 
405   /// \brief Return the identifier associated with the given ID number.
406   ///
407   /// The ID 0 is associated with the NULL identifier.
408   virtual IdentifierInfo *GetIdentifier(unsigned ID) = 0;
409 };
410 
411 /// \brief Implements an efficient mapping from strings to IdentifierInfo nodes.
412 ///
413 /// This has no other purpose, but this is an extremely performance-critical
414 /// piece of the code, as each occurrence of every identifier goes through
415 /// here when lexed.
416 class IdentifierTable {
417   // Shark shows that using MallocAllocator is *much* slower than using this
418   // BumpPtrAllocator!
419   typedef llvm::StringMap<IdentifierInfo*, llvm::BumpPtrAllocator> HashTableTy;
420   HashTableTy HashTable;
421 
422   IdentifierInfoLookup* ExternalLookup;
423 
424 public:
425   /// \brief Create the identifier table, populating it with info about the
426   /// language keywords for the language specified by \p LangOpts.
427   IdentifierTable(const LangOptions &LangOpts,
428                   IdentifierInfoLookup* externalLookup = nullptr);
429 
430   /// \brief Set the external identifier lookup mechanism.
setExternalIdentifierLookup(IdentifierInfoLookup * IILookup)431   void setExternalIdentifierLookup(IdentifierInfoLookup *IILookup) {
432     ExternalLookup = IILookup;
433   }
434 
435   /// \brief Retrieve the external identifier lookup object, if any.
getExternalIdentifierLookup()436   IdentifierInfoLookup *getExternalIdentifierLookup() const {
437     return ExternalLookup;
438   }
439 
getAllocator()440   llvm::BumpPtrAllocator& getAllocator() {
441     return HashTable.getAllocator();
442   }
443 
444   /// \brief Return the identifier token info for the specified named
445   /// identifier.
get(StringRef Name)446   IdentifierInfo &get(StringRef Name) {
447     llvm::StringMapEntry<IdentifierInfo*> &Entry =
448       HashTable.GetOrCreateValue(Name);
449 
450     IdentifierInfo *II = Entry.getValue();
451     if (II) return *II;
452 
453     // No entry; if we have an external lookup, look there first.
454     if (ExternalLookup) {
455       II = ExternalLookup->get(Name);
456       if (II) {
457         // Cache in the StringMap for subsequent lookups.
458         Entry.setValue(II);
459         return *II;
460       }
461     }
462 
463     // Lookups failed, make a new IdentifierInfo.
464     void *Mem = getAllocator().Allocate<IdentifierInfo>();
465     II = new (Mem) IdentifierInfo();
466     Entry.setValue(II);
467 
468     // Make sure getName() knows how to find the IdentifierInfo
469     // contents.
470     II->Entry = &Entry;
471 
472     return *II;
473   }
474 
get(StringRef Name,tok::TokenKind TokenCode)475   IdentifierInfo &get(StringRef Name, tok::TokenKind TokenCode) {
476     IdentifierInfo &II = get(Name);
477     II.TokenID = TokenCode;
478     assert(II.TokenID == (unsigned) TokenCode && "TokenCode too large");
479     return II;
480   }
481 
482   /// \brief Gets an IdentifierInfo for the given name without consulting
483   ///        external sources.
484   ///
485   /// This is a version of get() meant for external sources that want to
486   /// introduce or modify an identifier. If they called get(), they would
487   /// likely end up in a recursion.
getOwn(StringRef Name)488   IdentifierInfo &getOwn(StringRef Name) {
489     llvm::StringMapEntry<IdentifierInfo*> &Entry =
490       HashTable.GetOrCreateValue(Name);
491 
492     IdentifierInfo *II = Entry.getValue();
493     if (!II) {
494 
495       // Lookups failed, make a new IdentifierInfo.
496       void *Mem = getAllocator().Allocate<IdentifierInfo>();
497       II = new (Mem) IdentifierInfo();
498       Entry.setValue(II);
499 
500       // Make sure getName() knows how to find the IdentifierInfo
501       // contents.
502       II->Entry = &Entry;
503 
504       // If this is the 'import' contextual keyword, mark it as such.
505       if (Name.equals("import"))
506         II->setModulesImport(true);
507     }
508 
509     return *II;
510   }
511 
512   typedef HashTableTy::const_iterator iterator;
513   typedef HashTableTy::const_iterator const_iterator;
514 
begin()515   iterator begin() const { return HashTable.begin(); }
end()516   iterator end() const   { return HashTable.end(); }
size()517   unsigned size() const { return HashTable.size(); }
518 
519   /// \brief Print some statistics to stderr that indicate how well the
520   /// hashing is doing.
521   void PrintStats() const;
522 
523   void AddKeywords(const LangOptions &LangOpts);
524 };
525 
526 /// \brief A family of Objective-C methods.
527 ///
528 /// These families have no inherent meaning in the language, but are
529 /// nonetheless central enough in the existing implementations to
530 /// merit direct AST support.  While, in theory, arbitrary methods can
531 /// be considered to form families, we focus here on the methods
532 /// involving allocation and retain-count management, as these are the
533 /// most "core" and the most likely to be useful to diverse clients
534 /// without extra information.
535 ///
536 /// Both selectors and actual method declarations may be classified
537 /// into families.  Method families may impose additional restrictions
538 /// beyond their selector name; for example, a method called '_init'
539 /// that returns void is not considered to be in the 'init' family
540 /// (but would be if it returned 'id').  It is also possible to
541 /// explicitly change or remove a method's family.  Therefore the
542 /// method's family should be considered the single source of truth.
543 enum ObjCMethodFamily {
544   /// \brief No particular method family.
545   OMF_None,
546 
547   // Selectors in these families may have arbitrary arity, may be
548   // written with arbitrary leading underscores, and may have
549   // additional CamelCase "words" in their first selector chunk
550   // following the family name.
551   OMF_alloc,
552   OMF_copy,
553   OMF_init,
554   OMF_mutableCopy,
555   OMF_new,
556 
557   // These families are singletons consisting only of the nullary
558   // selector with the given name.
559   OMF_autorelease,
560   OMF_dealloc,
561   OMF_finalize,
562   OMF_release,
563   OMF_retain,
564   OMF_retainCount,
565   OMF_self,
566 
567   // performSelector families
568   OMF_performSelector
569 };
570 
571 /// Enough bits to store any enumerator in ObjCMethodFamily or
572 /// InvalidObjCMethodFamily.
573 enum { ObjCMethodFamilyBitWidth = 4 };
574 
575 /// \brief An invalid value of ObjCMethodFamily.
576 enum { InvalidObjCMethodFamily = (1 << ObjCMethodFamilyBitWidth) - 1 };
577 
578 /// \brief A family of Objective-C methods.
579 ///
580 /// These are family of methods whose result type is initially 'id', but
581 /// but are candidate for the result type to be changed to 'instancetype'.
582 enum ObjCInstanceTypeFamily {
583   OIT_None,
584   OIT_Array,
585   OIT_Dictionary,
586   OIT_Singleton,
587   OIT_Init,
588   OIT_ReturnsSelf
589 };
590 
591 /// \brief Smart pointer class that efficiently represents Objective-C method
592 /// names.
593 ///
594 /// This class will either point to an IdentifierInfo or a
595 /// MultiKeywordSelector (which is private). This enables us to optimize
596 /// selectors that take no arguments and selectors that take 1 argument, which
597 /// accounts for 78% of all selectors in Cocoa.h.
598 class Selector {
599   friend class Diagnostic;
600 
601   enum IdentifierInfoFlag {
602     // Empty selector = 0.
603     ZeroArg  = 0x1,
604     OneArg   = 0x2,
605     MultiArg = 0x3,
606     ArgFlags = ZeroArg|OneArg
607   };
608   uintptr_t InfoPtr; // a pointer to the MultiKeywordSelector or IdentifierInfo.
609 
Selector(IdentifierInfo * II,unsigned nArgs)610   Selector(IdentifierInfo *II, unsigned nArgs) {
611     InfoPtr = reinterpret_cast<uintptr_t>(II);
612     assert((InfoPtr & ArgFlags) == 0 &&"Insufficiently aligned IdentifierInfo");
613     assert(nArgs < 2 && "nArgs not equal to 0/1");
614     InfoPtr |= nArgs+1;
615   }
Selector(MultiKeywordSelector * SI)616   Selector(MultiKeywordSelector *SI) {
617     InfoPtr = reinterpret_cast<uintptr_t>(SI);
618     assert((InfoPtr & ArgFlags) == 0 &&"Insufficiently aligned IdentifierInfo");
619     InfoPtr |= MultiArg;
620   }
621 
getAsIdentifierInfo()622   IdentifierInfo *getAsIdentifierInfo() const {
623     if (getIdentifierInfoFlag() < MultiArg)
624       return reinterpret_cast<IdentifierInfo *>(InfoPtr & ~ArgFlags);
625     return nullptr;
626   }
getMultiKeywordSelector()627   MultiKeywordSelector *getMultiKeywordSelector() const {
628     return reinterpret_cast<MultiKeywordSelector *>(InfoPtr & ~ArgFlags);
629   }
630 
getIdentifierInfoFlag()631   unsigned getIdentifierInfoFlag() const {
632     return InfoPtr & ArgFlags;
633   }
634 
635   static ObjCMethodFamily getMethodFamilyImpl(Selector sel);
636 
637 public:
638   friend class SelectorTable; // only the SelectorTable can create these
639   friend class DeclarationName; // and the AST's DeclarationName.
640 
641   /// The default ctor should only be used when creating data structures that
642   ///  will contain selectors.
Selector()643   Selector() : InfoPtr(0) {}
Selector(uintptr_t V)644   Selector(uintptr_t V) : InfoPtr(V) {}
645 
646   /// operator==/!= - Indicate whether the specified selectors are identical.
647   bool operator==(Selector RHS) const {
648     return InfoPtr == RHS.InfoPtr;
649   }
650   bool operator!=(Selector RHS) const {
651     return InfoPtr != RHS.InfoPtr;
652   }
getAsOpaquePtr()653   void *getAsOpaquePtr() const {
654     return reinterpret_cast<void*>(InfoPtr);
655   }
656 
657   /// \brief Determine whether this is the empty selector.
isNull()658   bool isNull() const { return InfoPtr == 0; }
659 
660   // Predicates to identify the selector type.
isKeywordSelector()661   bool isKeywordSelector() const {
662     return getIdentifierInfoFlag() != ZeroArg;
663   }
isUnarySelector()664   bool isUnarySelector() const {
665     return getIdentifierInfoFlag() == ZeroArg;
666   }
667   unsigned getNumArgs() const;
668 
669 
670   /// \brief Retrieve the identifier at a given position in the selector.
671   ///
672   /// Note that the identifier pointer returned may be NULL. Clients that only
673   /// care about the text of the identifier string, and not the specific,
674   /// uniqued identifier pointer, should use \c getNameForSlot(), which returns
675   /// an empty string when the identifier pointer would be NULL.
676   ///
677   /// \param argIndex The index for which we want to retrieve the identifier.
678   /// This index shall be less than \c getNumArgs() unless this is a keyword
679   /// selector, in which case 0 is the only permissible value.
680   ///
681   /// \returns the uniqued identifier for this slot, or NULL if this slot has
682   /// no corresponding identifier.
683   IdentifierInfo *getIdentifierInfoForSlot(unsigned argIndex) const;
684 
685   /// \brief Retrieve the name at a given position in the selector.
686   ///
687   /// \param argIndex The index for which we want to retrieve the name.
688   /// This index shall be less than \c getNumArgs() unless this is a keyword
689   /// selector, in which case 0 is the only permissible value.
690   ///
691   /// \returns the name for this slot, which may be the empty string if no
692   /// name was supplied.
693   StringRef getNameForSlot(unsigned argIndex) const;
694 
695   /// \brief Derive the full selector name (e.g. "foo:bar:") and return
696   /// it as an std::string.
697   std::string getAsString() const;
698 
699   /// \brief Prints the full selector name (e.g. "foo:bar:").
700   void print(llvm::raw_ostream &OS) const;
701 
702   /// \brief Derive the conventional family of this method.
getMethodFamily()703   ObjCMethodFamily getMethodFamily() const {
704     return getMethodFamilyImpl(*this);
705   }
706 
getEmptyMarker()707   static Selector getEmptyMarker() {
708     return Selector(uintptr_t(-1));
709   }
getTombstoneMarker()710   static Selector getTombstoneMarker() {
711     return Selector(uintptr_t(-2));
712   }
713 
714   static ObjCInstanceTypeFamily getInstTypeMethodFamily(Selector sel);
715 };
716 
717 /// \brief This table allows us to fully hide how we implement
718 /// multi-keyword caching.
719 class SelectorTable {
720   void *Impl;  // Actually a SelectorTableImpl
721   SelectorTable(const SelectorTable &) LLVM_DELETED_FUNCTION;
722   void operator=(const SelectorTable &) LLVM_DELETED_FUNCTION;
723 public:
724   SelectorTable();
725   ~SelectorTable();
726 
727   /// \brief Can create any sort of selector.
728   ///
729   /// \p NumArgs indicates whether this is a no argument selector "foo", a
730   /// single argument selector "foo:" or multi-argument "foo:bar:".
731   Selector getSelector(unsigned NumArgs, IdentifierInfo **IIV);
732 
getUnarySelector(IdentifierInfo * ID)733   Selector getUnarySelector(IdentifierInfo *ID) {
734     return Selector(ID, 1);
735   }
getNullarySelector(IdentifierInfo * ID)736   Selector getNullarySelector(IdentifierInfo *ID) {
737     return Selector(ID, 0);
738   }
739 
740   /// \brief Return the total amount of memory allocated for managing selectors.
741   size_t getTotalMemory() const;
742 
743   /// \brief Return the default setter name for the given identifier.
744   ///
745   /// This is "set" + \p Name where the initial character of \p Name
746   /// has been capitalized.
747   static SmallString<64> constructSetterName(StringRef Name);
748 
749   /// \brief Return the default setter selector for the given identifier.
750   ///
751   /// This is "set" + \p Name where the initial character of \p Name
752   /// has been capitalized.
753   static Selector constructSetterSelector(IdentifierTable &Idents,
754                                           SelectorTable &SelTable,
755                                           const IdentifierInfo *Name);
756 };
757 
758 /// DeclarationNameExtra - Common base of the MultiKeywordSelector,
759 /// CXXSpecialName, and CXXOperatorIdName classes, all of which are
760 /// private classes that describe different kinds of names.
761 class DeclarationNameExtra {
762 public:
763   /// ExtraKind - The kind of "extra" information stored in the
764   /// DeclarationName. See @c ExtraKindOrNumArgs for an explanation of
765   /// how these enumerator values are used.
766   enum ExtraKind {
767     CXXConstructor = 0,
768     CXXDestructor,
769     CXXConversionFunction,
770 #define OVERLOADED_OPERATOR(Name,Spelling,Token,Unary,Binary,MemberOnly) \
771     CXXOperator##Name,
772 #include "clang/Basic/OperatorKinds.def"
773     CXXLiteralOperator,
774     CXXUsingDirective,
775     NUM_EXTRA_KINDS
776   };
777 
778   /// ExtraKindOrNumArgs - Either the kind of C++ special name or
779   /// operator-id (if the value is one of the CXX* enumerators of
780   /// ExtraKind), in which case the DeclarationNameExtra is also a
781   /// CXXSpecialName, (for CXXConstructor, CXXDestructor, or
782   /// CXXConversionFunction) CXXOperatorIdName, or CXXLiteralOperatorName,
783   /// it may be also name common to C++ using-directives (CXXUsingDirective),
784   /// otherwise it is NUM_EXTRA_KINDS+NumArgs, where NumArgs is the number of
785   /// arguments in the Objective-C selector, in which case the
786   /// DeclarationNameExtra is also a MultiKeywordSelector.
787   unsigned ExtraKindOrNumArgs;
788 };
789 
790 }  // end namespace clang
791 
792 namespace llvm {
793 /// Define DenseMapInfo so that Selectors can be used as keys in DenseMap and
794 /// DenseSets.
795 template <>
796 struct DenseMapInfo<clang::Selector> {
797   static inline clang::Selector getEmptyKey() {
798     return clang::Selector::getEmptyMarker();
799   }
800   static inline clang::Selector getTombstoneKey() {
801     return clang::Selector::getTombstoneMarker();
802   }
803 
804   static unsigned getHashValue(clang::Selector S);
805 
806   static bool isEqual(clang::Selector LHS, clang::Selector RHS) {
807     return LHS == RHS;
808   }
809 };
810 
811 template <>
812 struct isPodLike<clang::Selector> { static const bool value = true; };
813 
814 template <typename T> class PointerLikeTypeTraits;
815 
816 template<>
817 class PointerLikeTypeTraits<clang::Selector> {
818 public:
819   static inline const void *getAsVoidPointer(clang::Selector P) {
820     return P.getAsOpaquePtr();
821   }
822   static inline clang::Selector getFromVoidPointer(const void *P) {
823     return clang::Selector(reinterpret_cast<uintptr_t>(P));
824   }
825   enum { NumLowBitsAvailable = 0 };
826 };
827 
828 // Provide PointerLikeTypeTraits for IdentifierInfo pointers, which
829 // are not guaranteed to be 8-byte aligned.
830 template<>
831 class PointerLikeTypeTraits<clang::IdentifierInfo*> {
832 public:
833   static inline void *getAsVoidPointer(clang::IdentifierInfo* P) {
834     return P;
835   }
836   static inline clang::IdentifierInfo *getFromVoidPointer(void *P) {
837     return static_cast<clang::IdentifierInfo*>(P);
838   }
839   enum { NumLowBitsAvailable = 1 };
840 };
841 
842 template<>
843 class PointerLikeTypeTraits<const clang::IdentifierInfo*> {
844 public:
845   static inline const void *getAsVoidPointer(const clang::IdentifierInfo* P) {
846     return P;
847   }
848   static inline const clang::IdentifierInfo *getFromVoidPointer(const void *P) {
849     return static_cast<const clang::IdentifierInfo*>(P);
850   }
851   enum { NumLowBitsAvailable = 1 };
852 };
853 
854 }  // end namespace llvm
855 #endif
856