///////////////////////////////////////////////////////////////////////////// // Name: wx/hashmap.h // Purpose: wxHashMap class // Author: Mattia Barbon // Modified by: // Created: 29/01/2002 // Copyright: (c) Mattia Barbon // Licence: wxWindows licence ///////////////////////////////////////////////////////////////////////////// #ifndef _WX_HASHMAP_H_ #define _WX_HASHMAP_H_ #include "wx/string.h" #include "wx/wxcrt.h" // In wxUSE_STD_CONTAINERS build we prefer to use the standard hash map class // but it can be either in non-standard hash_map header (old g++ and some other // STL implementations) or in C++0x standard unordered_map which can in turn be // available either in std::tr1 or std namespace itself // // To summarize: if std::unordered_map is available use it, otherwise use tr1 // and finally fall back to non-standard hash_map #if (defined(HAVE_EXT_HASH_MAP) || defined(HAVE_HASH_MAP)) \ && (defined(HAVE_GNU_CXX_HASH_MAP) || defined(HAVE_STD_HASH_MAP)) #define HAVE_STL_HASH_MAP #endif #if wxUSE_STD_CONTAINERS && \ (defined(HAVE_STD_UNORDERED_MAP) || defined(HAVE_TR1_UNORDERED_MAP)) #if defined(HAVE_STD_UNORDERED_MAP) #include <unordered_map> #define WX_HASH_MAP_NAMESPACE std #elif defined(HAVE_TR1_UNORDERED_MAP) #include <tr1/unordered_map> #define WX_HASH_MAP_NAMESPACE std::tr1 #endif #define _WX_DECLARE_HASH_MAP( KEY_T, VALUE_T, HASH_T, KEY_EQ_T, CLASSNAME, CLASSEXP ) \ typedef WX_HASH_MAP_NAMESPACE::unordered_map< KEY_T, VALUE_T, HASH_T, KEY_EQ_T > CLASSNAME #elif wxUSE_STD_CONTAINERS && defined(HAVE_STL_HASH_MAP) #if defined(HAVE_EXT_HASH_MAP) #include <ext/hash_map> #elif defined(HAVE_HASH_MAP) #include <hash_map> #endif #if defined(HAVE_GNU_CXX_HASH_MAP) #define WX_HASH_MAP_NAMESPACE __gnu_cxx #elif defined(HAVE_STD_HASH_MAP) #define WX_HASH_MAP_NAMESPACE std #endif #define _WX_DECLARE_HASH_MAP( KEY_T, VALUE_T, HASH_T, KEY_EQ_T, CLASSNAME, CLASSEXP ) \ typedef WX_HASH_MAP_NAMESPACE::hash_map< KEY_T, VALUE_T, HASH_T, KEY_EQ_T > CLASSNAME #else // !wxUSE_STD_CONTAINERS || no std::{hash,unordered}_map class available #define wxNEEDS_WX_HASH_MAP #include <stddef.h> // for ptrdiff_t // private struct WXDLLIMPEXP_BASE _wxHashTable_NodeBase { _wxHashTable_NodeBase() : m_next(NULL) {} _wxHashTable_NodeBase* m_next; // Cannot do this: // wxDECLARE_NO_COPY_CLASS(_wxHashTable_NodeBase); // without rewriting the macros, which require a public copy constructor. }; // private class WXDLLIMPEXP_BASE _wxHashTableBase2 { public: typedef void (*NodeDtor)(_wxHashTable_NodeBase*); typedef size_t (*BucketFromNode)(_wxHashTableBase2*,_wxHashTable_NodeBase*); typedef _wxHashTable_NodeBase* (*ProcessNode)(_wxHashTable_NodeBase*); protected: static _wxHashTable_NodeBase* DummyProcessNode(_wxHashTable_NodeBase* node); static void DeleteNodes( size_t buckets, _wxHashTable_NodeBase** table, NodeDtor dtor ); static _wxHashTable_NodeBase* GetFirstNode( size_t buckets, _wxHashTable_NodeBase** table ) { for( size_t i = 0; i < buckets; ++i ) if( table[i] ) return table[i]; return NULL; } // as static const unsigned prime_count = 31 but works with all compilers enum { prime_count = 31 }; static const unsigned long ms_primes[prime_count]; // returns the first prime in ms_primes greater than n static unsigned long GetNextPrime( unsigned long n ); // returns the first prime in ms_primes smaller than n // ( or ms_primes[0] if n is very small ) static unsigned long GetPreviousPrime( unsigned long n ); static void CopyHashTable( _wxHashTable_NodeBase** srcTable, size_t srcBuckets, _wxHashTableBase2* dst, _wxHashTable_NodeBase** dstTable, BucketFromNode func, ProcessNode proc ); static void** AllocTable( size_t sz ) { return (void **)calloc(sz, sizeof(void*)); } static void FreeTable(void *table) { free(table); } }; #define _WX_DECLARE_HASHTABLE( VALUE_T, KEY_T, HASH_T, KEY_EX_T, KEY_EQ_T,\ PTROPERATOR, CLASSNAME, CLASSEXP, \ SHOULD_GROW, SHOULD_SHRINK ) \ CLASSEXP CLASSNAME : protected _wxHashTableBase2 \ { \ public: \ typedef KEY_T key_type; \ typedef VALUE_T value_type; \ typedef HASH_T hasher; \ typedef KEY_EQ_T key_equal; \ \ typedef size_t size_type; \ typedef ptrdiff_t difference_type; \ typedef value_type* pointer; \ typedef const value_type* const_pointer; \ typedef value_type& reference; \ typedef const value_type& const_reference; \ /* should these be protected? */ \ typedef const KEY_T const_key_type; \ typedef const VALUE_T const_mapped_type; \ public: \ typedef KEY_EX_T key_extractor; \ typedef CLASSNAME Self; \ protected: \ _wxHashTable_NodeBase** m_table; \ size_t m_tableBuckets; \ size_t m_items; \ hasher m_hasher; \ key_equal m_equals; \ key_extractor m_getKey; \ public: \ struct Node:public _wxHashTable_NodeBase \ { \ public: \ Node( const value_type& value ) \ : m_value( value ) {} \ Node* next() { return static_cast<Node*>(m_next); } \ \ value_type m_value; \ }; \ \ protected: \ static void DeleteNode( _wxHashTable_NodeBase* node ) \ { \ delete static_cast<Node*>(node); \ } \ public: \ /* */ \ /* forward iterator */ \ /* */ \ CLASSEXP Iterator \ { \ public: \ Node* m_node; \ Self* m_ht; \ \ Iterator() : m_node(NULL), m_ht(NULL) {} \ Iterator( Node* node, const Self* ht ) \ : m_node(node), m_ht(const_cast<Self*>(ht)) {} \ bool operator ==( const Iterator& it ) const \ { return m_node == it.m_node; } \ bool operator !=( const Iterator& it ) const \ { return m_node != it.m_node; } \ protected: \ Node* GetNextNode() \ { \ size_type bucket = GetBucketForNode(m_ht,m_node); \ for( size_type i = bucket + 1; i < m_ht->m_tableBuckets; ++i ) \ { \ if( m_ht->m_table[i] ) \ return static_cast<Node*>(m_ht->m_table[i]); \ } \ return NULL; \ } \ \ void PlusPlus() \ { \ Node* next = m_node->next(); \ m_node = next ? next : GetNextNode(); \ } \ }; \ friend class Iterator; \ \ public: \ CLASSEXP iterator : public Iterator \ { \ public: \ iterator() : Iterator() {} \ iterator( Node* node, Self* ht ) : Iterator( node, ht ) {} \ iterator& operator++() { PlusPlus(); return *this; } \ iterator operator++(int) { iterator it=*this;PlusPlus();return it; } \ reference operator *() const { return m_node->m_value; } \ PTROPERATOR(pointer) \ }; \ \ CLASSEXP const_iterator : public Iterator \ { \ public: \ const_iterator() : Iterator() {} \ const_iterator(iterator i) : Iterator(i) {} \ const_iterator( Node* node, const Self* ht ) \ : Iterator(node, const_cast<Self*>(ht)) {} \ const_iterator& operator++() { PlusPlus();return *this; } \ const_iterator operator++(int) { const_iterator it=*this;PlusPlus();return it; } \ const_reference operator *() const { return m_node->m_value; } \ PTROPERATOR(const_pointer) \ }; \ \ CLASSNAME( size_type sz = 10, const hasher& hfun = hasher(), \ const key_equal& k_eq = key_equal(), \ const key_extractor& k_ex = key_extractor() ) \ : m_tableBuckets( GetNextPrime( (unsigned long) sz ) ), \ m_items( 0 ), \ m_hasher( hfun ), \ m_equals( k_eq ), \ m_getKey( k_ex ) \ { \ m_table = (_wxHashTable_NodeBase**)AllocTable(m_tableBuckets); \ } \ \ CLASSNAME( const Self& ht ) \ : m_table(NULL), \ m_tableBuckets( 0 ), \ m_items( ht.m_items ), \ m_hasher( ht.m_hasher ), \ m_equals( ht.m_equals ), \ m_getKey( ht.m_getKey ) \ { \ HashCopy( ht ); \ } \ \ const Self& operator=( const Self& ht ) \ { \ if (&ht != this) \ { \ clear(); \ m_hasher = ht.m_hasher; \ m_equals = ht.m_equals; \ m_getKey = ht.m_getKey; \ m_items = ht.m_items; \ HashCopy( ht ); \ } \ return *this; \ } \ \ ~CLASSNAME() \ { \ clear(); \ \ FreeTable(m_table); \ } \ \ hasher hash_funct() { return m_hasher; } \ key_equal key_eq() { return m_equals; } \ \ /* removes all elements from the hash table, but does not */ \ /* shrink it ( perhaps it should ) */ \ void clear() \ { \ DeleteNodes(m_tableBuckets, m_table, DeleteNode); \ m_items = 0; \ } \ \ size_type size() const { return m_items; } \ size_type max_size() const { return size_type(-1); } \ bool empty() const { return size() == 0; } \ \ const_iterator end() const { return const_iterator(NULL, this); } \ iterator end() { return iterator(NULL, this); } \ const_iterator begin() const \ { return const_iterator(static_cast<Node*>(GetFirstNode(m_tableBuckets, m_table)), this); } \ iterator begin() \ { return iterator(static_cast<Node*>(GetFirstNode(m_tableBuckets, m_table)), this); } \ \ size_type erase( const const_key_type& key ) \ { \ _wxHashTable_NodeBase** node = GetNodePtr(key); \ \ if( !node ) \ return 0; \ \ --m_items; \ _wxHashTable_NodeBase* temp = (*node)->m_next; \ delete static_cast<Node*>(*node); \ (*node) = temp; \ if( SHOULD_SHRINK( m_tableBuckets, m_items ) ) \ ResizeTable( GetPreviousPrime( (unsigned long) m_tableBuckets ) - 1 ); \ return 1; \ } \ \ protected: \ static size_type GetBucketForNode( Self* ht, Node* node ) \ { \ return ht->m_hasher( ht->m_getKey( node->m_value ) ) \ % ht->m_tableBuckets; \ } \ static Node* CopyNode( Node* node ) { return new Node( *node ); } \ \ Node* GetOrCreateNode( const value_type& value, bool& created ) \ { \ const const_key_type& key = m_getKey( value ); \ size_t bucket = m_hasher( key ) % m_tableBuckets; \ Node* node = static_cast<Node*>(m_table[bucket]); \ \ while( node ) \ { \ if( m_equals( m_getKey( node->m_value ), key ) ) \ { \ created = false; \ return node; \ } \ node = node->next(); \ } \ created = true; \ return CreateNode( value, bucket); \ }\ Node * CreateNode( const value_type& value, size_t bucket ) \ {\ Node* node = new Node( value ); \ node->m_next = m_table[bucket]; \ m_table[bucket] = node; \ \ /* must be after the node is inserted */ \ ++m_items; \ if( SHOULD_GROW( m_tableBuckets, m_items ) ) \ ResizeTable( m_tableBuckets ); \ \ return node; \ } \ void CreateNode( const value_type& value ) \ {\ CreateNode(value, m_hasher( m_getKey(value) ) % m_tableBuckets ); \ }\ \ /* returns NULL if not found */ \ _wxHashTable_NodeBase** GetNodePtr(const const_key_type& key) const \ { \ size_t bucket = m_hasher( key ) % m_tableBuckets; \ _wxHashTable_NodeBase** node = &m_table[bucket]; \ \ while( *node ) \ { \ if (m_equals(m_getKey(static_cast<Node*>(*node)->m_value), key)) \ return node; \ node = &(*node)->m_next; \ } \ \ return NULL; \ } \ \ /* returns NULL if not found */ \ /* expressing it in terms of GetNodePtr is 5-8% slower :-( */ \ Node* GetNode( const const_key_type& key ) const \ { \ size_t bucket = m_hasher( key ) % m_tableBuckets; \ Node* node = static_cast<Node*>(m_table[bucket]); \ \ while( node ) \ { \ if( m_equals( m_getKey( node->m_value ), key ) ) \ return node; \ node = node->next(); \ } \ \ return NULL; \ } \ \ void ResizeTable( size_t newSize ) \ { \ newSize = GetNextPrime( (unsigned long)newSize ); \ _wxHashTable_NodeBase** srcTable = m_table; \ size_t srcBuckets = m_tableBuckets; \ m_table = (_wxHashTable_NodeBase**)AllocTable( newSize ); \ m_tableBuckets = newSize; \ \ CopyHashTable( srcTable, srcBuckets, \ this, m_table, \ (BucketFromNode)GetBucketForNode,\ (ProcessNode)&DummyProcessNode ); \ FreeTable(srcTable); \ } \ \ /* this must be called _after_ m_table has been cleaned */ \ void HashCopy( const Self& ht ) \ { \ ResizeTable( ht.size() ); \ CopyHashTable( ht.m_table, ht.m_tableBuckets, \ (_wxHashTableBase2*)this, \ m_table, \ (BucketFromNode)GetBucketForNode, \ (ProcessNode)CopyNode ); \ } \ }; // defines an STL-like pair class CLASSNAME storing two fields: first of type // KEY_T and second of type VALUE_T #define _WX_DECLARE_PAIR( KEY_T, VALUE_T, CLASSNAME, CLASSEXP ) \ CLASSEXP CLASSNAME \ { \ public: \ typedef KEY_T first_type; \ typedef VALUE_T second_type; \ typedef KEY_T t1; \ typedef VALUE_T t2; \ typedef const KEY_T const_t1; \ typedef const VALUE_T const_t2; \ \ CLASSNAME(const const_t1& f, const const_t2& s) \ : first(const_cast<t1&>(f)), second(const_cast<t2&>(s)) {} \ \ t1 first; \ t2 second; \ }; // defines the class CLASSNAME returning the key part (of type KEY_T) from a // pair of type PAIR_T #define _WX_DECLARE_HASH_MAP_KEY_EX( KEY_T, PAIR_T, CLASSNAME, CLASSEXP ) \ CLASSEXP CLASSNAME \ { \ typedef KEY_T key_type; \ typedef PAIR_T pair_type; \ typedef const key_type const_key_type; \ typedef const pair_type const_pair_type; \ typedef const_key_type& const_key_reference; \ typedef const_pair_type& const_pair_reference; \ public: \ CLASSNAME() { } \ const_key_reference operator()( const_pair_reference pair ) const { return pair.first; }\ }; // grow/shrink predicates inline bool never_grow( size_t, size_t ) { return false; } inline bool never_shrink( size_t, size_t ) { return false; } inline bool grow_lf70( size_t buckets, size_t items ) { return float(items)/float(buckets) >= 0.85f; } #endif // various hash map implementations // ---------------------------------------------------------------------------- // hashing and comparison functors // ---------------------------------------------------------------------------- #ifndef wxNEEDS_WX_HASH_MAP // integer types struct WXDLLIMPEXP_BASE wxIntegerHash { private: WX_HASH_MAP_NAMESPACE::hash<long> longHash; WX_HASH_MAP_NAMESPACE::hash<unsigned long> ulongHash; WX_HASH_MAP_NAMESPACE::hash<int> intHash; WX_HASH_MAP_NAMESPACE::hash<unsigned int> uintHash; WX_HASH_MAP_NAMESPACE::hash<short> shortHash; WX_HASH_MAP_NAMESPACE::hash<unsigned short> ushortHash; #ifdef wxHAS_LONG_LONG_T_DIFFERENT_FROM_LONG // hash<wxLongLong_t> ought to work but doesn't on some compilers #if (!defined SIZEOF_LONG_LONG && SIZEOF_LONG == 4) \ || (defined SIZEOF_LONG_LONG && SIZEOF_LONG_LONG == SIZEOF_LONG * 2) size_t longlongHash( wxLongLong_t x ) const { return longHash( wx_truncate_cast(long, x) ) ^ longHash( wx_truncate_cast(long, x >> (sizeof(long) * 8)) ); } #elif defined SIZEOF_LONG_LONG && SIZEOF_LONG_LONG == SIZEOF_LONG WX_HASH_MAP_NAMESPACE::hash<long> longlongHash; #else WX_HASH_MAP_NAMESPACE::hash<wxLongLong_t> longlongHash; #endif #endif // wxHAS_LONG_LONG_T_DIFFERENT_FROM_LONG public: wxIntegerHash() wxNOEXCEPT { } size_t operator()( long x ) const wxNOEXCEPT { return longHash( x ); } size_t operator()( unsigned long x ) const wxNOEXCEPT { return ulongHash( x ); } size_t operator()( int x ) const wxNOEXCEPT { return intHash( x ); } size_t operator()( unsigned int x ) const wxNOEXCEPT { return uintHash( x ); } size_t operator()( short x ) const wxNOEXCEPT { return shortHash( x ); } size_t operator()( unsigned short x ) const wxNOEXCEPT { return ushortHash( x ); } #ifdef wxHAS_LONG_LONG_T_DIFFERENT_FROM_LONG size_t operator()( wxLongLong_t x ) const wxNOEXCEPT { return longlongHash(x); } size_t operator()( wxULongLong_t x ) const wxNOEXCEPT { return longlongHash(x); } #endif // wxHAS_LONG_LONG_T_DIFFERENT_FROM_LONG }; #else // wxNEEDS_WX_HASH_MAP // integer types struct WXDLLIMPEXP_BASE wxIntegerHash { wxIntegerHash() wxNOEXCEPT { } unsigned long operator()( long x ) const wxNOEXCEPT { return (unsigned long)x; } unsigned long operator()( unsigned long x ) const wxNOEXCEPT { return x; } unsigned long operator()( int x ) const wxNOEXCEPT { return (unsigned long)x; } unsigned long operator()( unsigned int x ) const wxNOEXCEPT { return x; } unsigned long operator()( short x ) const wxNOEXCEPT { return (unsigned long)x; } unsigned long operator()( unsigned short x ) const wxNOEXCEPT { return x; } #ifdef wxHAS_LONG_LONG_T_DIFFERENT_FROM_LONG wxULongLong_t operator()( wxLongLong_t x ) const wxNOEXCEPT { return static_cast<wxULongLong_t>(x); } wxULongLong_t operator()( wxULongLong_t x ) const wxNOEXCEPT { return x; } #endif // wxHAS_LONG_LONG_T_DIFFERENT_FROM_LONG }; #endif // !wxNEEDS_WX_HASH_MAP/wxNEEDS_WX_HASH_MAP struct WXDLLIMPEXP_BASE wxIntegerEqual { wxIntegerEqual() wxNOEXCEPT { } bool operator()( long a, long b ) const wxNOEXCEPT { return a == b; } bool operator()( unsigned long a, unsigned long b ) const wxNOEXCEPT { return a == b; } bool operator()( int a, int b ) const wxNOEXCEPT { return a == b; } bool operator()( unsigned int a, unsigned int b ) const wxNOEXCEPT { return a == b; } bool operator()( short a, short b ) const wxNOEXCEPT { return a == b; } bool operator()( unsigned short a, unsigned short b ) const wxNOEXCEPT { return a == b; } #ifdef wxHAS_LONG_LONG_T_DIFFERENT_FROM_LONG bool operator()( wxLongLong_t a, wxLongLong_t b ) const wxNOEXCEPT { return a == b; } bool operator()( wxULongLong_t a, wxULongLong_t b ) const wxNOEXCEPT { return a == b; } #endif // wxHAS_LONG_LONG_T_DIFFERENT_FROM_LONG }; // pointers struct WXDLLIMPEXP_BASE wxPointerHash { wxPointerHash() wxNOEXCEPT { } #ifdef wxNEEDS_WX_HASH_MAP wxUIntPtr operator()( const void* k ) const wxNOEXCEPT { return wxPtrToUInt(k); } #else size_t operator()( const void* k ) const wxNOEXCEPT { return (size_t)k; } #endif }; struct WXDLLIMPEXP_BASE wxPointerEqual { wxPointerEqual() wxNOEXCEPT { } bool operator()( const void* a, const void* b ) const wxNOEXCEPT { return a == b; } }; // wxString, char*, wchar_t* struct WXDLLIMPEXP_BASE wxStringHash { wxStringHash() wxNOEXCEPT {} unsigned long operator()( const wxString& x ) const wxNOEXCEPT { return stringHash( x.wx_str() ); } unsigned long operator()( const wchar_t* x ) const wxNOEXCEPT { return stringHash( x ); } unsigned long operator()( const char* x ) const wxNOEXCEPT { return stringHash( x ); } #if WXWIN_COMPATIBILITY_2_8 static unsigned long wxCharStringHash( const wxChar* x ) { return stringHash(x); } #if wxUSE_UNICODE static unsigned long charStringHash( const char* x ) { return stringHash(x); } #endif #endif // WXWIN_COMPATIBILITY_2_8 static unsigned long stringHash( const wchar_t* ); static unsigned long stringHash( const char* ); }; struct WXDLLIMPEXP_BASE wxStringEqual { wxStringEqual() wxNOEXCEPT {} bool operator()( const wxString& a, const wxString& b ) const wxNOEXCEPT { return a == b; } bool operator()( const wxChar* a, const wxChar* b ) const wxNOEXCEPT { return wxStrcmp( a, b ) == 0; } #if wxUSE_UNICODE bool operator()( const char* a, const char* b ) const wxNOEXCEPT { return strcmp( a, b ) == 0; } #endif // wxUSE_UNICODE }; #ifdef wxNEEDS_WX_HASH_MAP #define wxPTROP_NORMAL(pointer) \ pointer operator ->() const { return &(m_node->m_value); } #define wxPTROP_NOP(pointer) #define _WX_DECLARE_HASH_MAP( KEY_T, VALUE_T, HASH_T, KEY_EQ_T, CLASSNAME, CLASSEXP ) \ _WX_DECLARE_PAIR( KEY_T, VALUE_T, CLASSNAME##_wxImplementation_Pair, CLASSEXP ) \ _WX_DECLARE_HASH_MAP_KEY_EX( KEY_T, CLASSNAME##_wxImplementation_Pair, CLASSNAME##_wxImplementation_KeyEx, CLASSEXP ) \ _WX_DECLARE_HASHTABLE( CLASSNAME##_wxImplementation_Pair, KEY_T, HASH_T, \ CLASSNAME##_wxImplementation_KeyEx, KEY_EQ_T, wxPTROP_NORMAL, \ CLASSNAME##_wxImplementation_HashTable, CLASSEXP, grow_lf70, never_shrink ) \ CLASSEXP CLASSNAME:public CLASSNAME##_wxImplementation_HashTable \ { \ public: \ typedef VALUE_T mapped_type; \ _WX_DECLARE_PAIR( iterator, bool, Insert_Result, CLASSEXP ) \ \ explicit CLASSNAME( size_type hint = 100, hasher hf = hasher(), \ key_equal eq = key_equal() ) \ : CLASSNAME##_wxImplementation_HashTable( hint, hf, eq, \ CLASSNAME##_wxImplementation_KeyEx() ) {} \ \ mapped_type& operator[]( const const_key_type& key ) \ { \ bool created; \ return GetOrCreateNode( \ CLASSNAME##_wxImplementation_Pair( key, mapped_type() ), \ created)->m_value.second; \ } \ \ const_iterator find( const const_key_type& key ) const \ { \ return const_iterator( GetNode( key ), this ); \ } \ \ iterator find( const const_key_type& key ) \ { \ return iterator( GetNode( key ), this ); \ } \ \ Insert_Result insert( const value_type& v ) \ { \ bool created; \ Node *node = GetOrCreateNode( \ CLASSNAME##_wxImplementation_Pair( v.first, v.second ), \ created); \ return Insert_Result(iterator(node, this), created); \ } \ \ size_type erase( const key_type& k ) \ { return CLASSNAME##_wxImplementation_HashTable::erase( k ); } \ void erase( const iterator& it ) { erase( (*it).first ); } \ \ /* count() == 0 | 1 */ \ size_type count( const const_key_type& key ) \ { \ return GetNode( key ) ? 1u : 0u; \ } \ } #endif // wxNEEDS_WX_HASH_MAP // these macros are to be used in the user code #define WX_DECLARE_HASH_MAP( KEY_T, VALUE_T, HASH_T, KEY_EQ_T, CLASSNAME) \ _WX_DECLARE_HASH_MAP( KEY_T, VALUE_T, HASH_T, KEY_EQ_T, CLASSNAME, class ) #define WX_DECLARE_STRING_HASH_MAP( VALUE_T, CLASSNAME ) \ _WX_DECLARE_HASH_MAP( wxString, VALUE_T, wxStringHash, wxStringEqual, \ CLASSNAME, class ) #define WX_DECLARE_VOIDPTR_HASH_MAP( VALUE_T, CLASSNAME ) \ _WX_DECLARE_HASH_MAP( void*, VALUE_T, wxPointerHash, wxPointerEqual, \ CLASSNAME, class ) // and these do exactly the same thing but should be used inside the // library // note: DECL is not used since the class is inline #define WX_DECLARE_HASH_MAP_WITH_DECL( KEY_T, VALUE_T, HASH_T, KEY_EQ_T, CLASSNAME, DECL) \ _WX_DECLARE_HASH_MAP( KEY_T, VALUE_T, HASH_T, KEY_EQ_T, CLASSNAME, class ) #define WX_DECLARE_EXPORTED_HASH_MAP( KEY_T, VALUE_T, HASH_T, KEY_EQ_T, CLASSNAME) \ WX_DECLARE_HASH_MAP_WITH_DECL( KEY_T, VALUE_T, HASH_T, KEY_EQ_T, \ CLASSNAME, class WXDLLIMPEXP_CORE ) // note: DECL is not used since the class is inline #define WX_DECLARE_STRING_HASH_MAP_WITH_DECL( VALUE_T, CLASSNAME, DECL ) \ _WX_DECLARE_HASH_MAP( wxString, VALUE_T, wxStringHash, wxStringEqual, \ CLASSNAME, class ) #define WX_DECLARE_EXPORTED_STRING_HASH_MAP( VALUE_T, CLASSNAME ) \ WX_DECLARE_STRING_HASH_MAP_WITH_DECL( VALUE_T, CLASSNAME, \ class WXDLLIMPEXP_CORE ) // note: DECL is not used since the class is inline #define WX_DECLARE_VOIDPTR_HASH_MAP_WITH_DECL( VALUE_T, CLASSNAME, DECL ) \ _WX_DECLARE_HASH_MAP( void*, VALUE_T, wxPointerHash, wxPointerEqual, \ CLASSNAME, class ) #define WX_DECLARE_EXPORTED_VOIDPTR_HASH_MAP( VALUE_T, CLASSNAME ) \ WX_DECLARE_VOIDPTR_HASH_MAP_WITH_DECL( VALUE_T, CLASSNAME, \ class WXDLLIMPEXP_CORE ) // delete all hash elements // // NB: the class declaration of the hash elements must be visible from the // place where you use this macro, otherwise the proper destructor may not // be called (a decent compiler should give a warning about it, but don't // count on it)! #define WX_CLEAR_HASH_MAP(type, hashmap) \ { \ type::iterator it, en; \ for( it = (hashmap).begin(), en = (hashmap).end(); it != en; ++it ) \ delete it->second; \ (hashmap).clear(); \ } //--------------------------------------------------------------------------- // Declarations of common hashmap classes WX_DECLARE_HASH_MAP_WITH_DECL( long, long, wxIntegerHash, wxIntegerEqual, wxLongToLongHashMap, class WXDLLIMPEXP_BASE ); WX_DECLARE_STRING_HASH_MAP_WITH_DECL( wxString, wxStringToStringHashMap, class WXDLLIMPEXP_BASE ); WX_DECLARE_STRING_HASH_MAP_WITH_DECL( wxUIntPtr, wxStringToNumHashMap, class WXDLLIMPEXP_BASE ); #endif // _WX_HASHMAP_H_