/////////////////////////////////////////////////////////////////////////////// // Name: string.h // Purpose: wxString and wxArrayString classes // Author: Vadim Zeitlin // Modified by: // Created: 29/01/98 // RCS-ID: $Id$ // Copyright: (c) 1998 Vadim Zeitlin // Licence: wxWindows license /////////////////////////////////////////////////////////////////////////////// /* Efficient string class [more or less] compatible with MFC CString, wxWindows version 1 wxString and std::string and some handy functions missing from string.h. */ #ifndef _WX_WXSTRINGH__ #define _WX_WXSTRINGH__ #ifdef __GNUG__ #pragma interface "string.h" #endif // ---------------------------------------------------------------------------- // conditinal compilation // ---------------------------------------------------------------------------- // compile the std::string compatibility functions if defined #define wxSTD_STRING_COMPATIBILITY // define to derive wxString from wxObject (deprecated!) #ifdef WXSTRING_IS_WXOBJECT #undef WXSTRING_IS_WXOBJECT #endif // ---------------------------------------------------------------------------- // headers // ---------------------------------------------------------------------------- #if defined(__WXMAC__) || defined(__VISAGECPP__) #include #endif #ifdef __EMX__ #include #endif #if defined(__VISAGECPP__) && __IBMCPP__ >= 400 // problem in VACPP V4 with including stdlib.h multiple times // strconv includes it anyway # include # include # include # include #else # include # include # include # include # include #endif #ifdef HAVE_STRINGS_H #include // for strcasecmp() #endif // AIX #include "wx/defs.h" // everybody should include this #include "wx/debug.h" // for wxASSERT() #include "wx/wxchar.h" // for wxChar #include "wx/buffer.h" // for wxCharBuffer #include "wx/strconv.h" // for wxConvertXXX() macros and wxMBConv classes #ifndef WX_PRECOMP #ifdef WXSTRING_IS_WXOBJECT #include "wx/object.h" // base class #endif #endif // !PCH // --------------------------------------------------------------------------- // macros // --------------------------------------------------------------------------- // 'naughty' cast #define WXSTRINGCAST (wxChar *)(const wxChar *) #define wxCSTRINGCAST (wxChar *)(const wxChar *) #define wxMBSTRINGCAST (char *)(const char *) #define wxWCSTRINGCAST (wchar_t *)(const wchar_t *) // implementation only #define ASSERT_VALID_INDEX(i) wxASSERT( (unsigned)(i) <= Len() ) // ---------------------------------------------------------------------------- // constants // ---------------------------------------------------------------------------- // maximum possible length for a string means "take all string" everywhere // (as sizeof(StringData) is unknown here, we substract 100) const unsigned int wxSTRING_MAXLEN = UINT_MAX - 100; // ---------------------------------------------------------------------------- // global data // ---------------------------------------------------------------------------- // global pointer to empty string WXDLLEXPORT_DATA(extern const wxChar*) wxEmptyString; // --------------------------------------------------------------------------- // global functions complementing standard C string library replacements for // strlen() and portable strcasecmp() //--------------------------------------------------------------------------- // Use wxXXX() functions from wxchar.h instead! These functions are for // backwards compatibility only. // checks whether the passed in pointer is NULL and if the string is empty inline bool WXDLLEXPORT IsEmpty(const char *p) { return (!p || !*p); } // safe version of strlen() (returns 0 if passed NULL pointer) inline size_t WXDLLEXPORT Strlen(const char *psz) { return psz ? strlen(psz) : 0; } // portable strcasecmp/_stricmp inline int WXDLLEXPORT Stricmp(const char *psz1, const char *psz2) { #if defined(__VISUALC__) || ( defined(__MWERKS__) && defined(__INTEL__) ) return _stricmp(psz1, psz2); #elif defined(__SC__) return _stricmp(psz1, psz2); #elif defined(__SALFORDC__) return stricmp(psz1, psz2); #elif defined(__BORLANDC__) return stricmp(psz1, psz2); #elif defined(__WATCOMC__) return stricmp(psz1, psz2); #elif defined(__EMX__) return stricmp(psz1, psz2); #elif defined(__WXPM__) return stricmp(psz1, psz2); #elif defined(__UNIX__) || defined(__GNUWIN32__) return strcasecmp(psz1, psz2); #elif defined(__MWERKS__) && !defined(__INTEL__) register char c1, c2; do { c1 = tolower(*psz1++); c2 = tolower(*psz2++); } while ( c1 && (c1 == c2) ); return c1 - c2; #else // almost all compilers/libraries provide this function (unfortunately under // different names), that's why we don't implement our own which will surely // be more efficient than this code (uncomment to use): /* register char c1, c2; do { c1 = tolower(*psz1++); c2 = tolower(*psz2++); } while ( c1 && (c1 == c2) ); return c1 - c2; */ #error "Please define string case-insensitive compare for your OS/compiler" #endif // OS/compiler } // wxSnprintf() is like snprintf() if it's available and sprintf() (always // available, but dangerous!) if not extern int WXDLLEXPORT wxSnprintf(wxChar *buf, size_t len, const wxChar *format, ...); // and wxVsnprintf() is like vsnprintf() or vsprintf() extern int WXDLLEXPORT wxVsnprintf(wxChar *buf, size_t len, const wxChar *format, va_list argptr); // return an empty wxString class WXDLLEXPORT wxString; // not yet defined inline const wxString& wxGetEmptyString() { return *(wxString *)&wxEmptyString; } // --------------------------------------------------------------------------- // string data prepended with some housekeeping info (used by wxString class), // is never used directly (but had to be put here to allow inlining) // --------------------------------------------------------------------------- struct WXDLLEXPORT wxStringData { int nRefs; // reference count size_t nDataLength, // actual string length nAllocLength; // allocated memory size // mimics declaration 'wxChar data[nAllocLength]' wxChar* data() const { return (wxChar*)(this + 1); } // empty string has a special ref count so it's never deleted bool IsEmpty() const { return (nRefs == -1); } bool IsShared() const { return (nRefs > 1); } // lock/unlock void Lock() { if ( !IsEmpty() ) nRefs++; } void Unlock() { if ( !IsEmpty() && --nRefs == 0) free(this); } // if we had taken control over string memory (GetWriteBuf), it's // intentionally put in invalid state void Validate(bool b) { nRefs = (b ? 1 : 0); } bool IsValid() const { return (nRefs != 0); } }; // --------------------------------------------------------------------------- // This is (yet another one) String class for C++ programmers. It doesn't use // any of "advanced" C++ features (i.e. templates, exceptions, namespaces...) // thus you should be able to compile it with practicaly any C++ compiler. // This class uses copy-on-write technique, i.e. identical strings share the // same memory as long as neither of them is changed. // // This class aims to be as compatible as possible with the new standard // std::string class, but adds some additional functions and should be at // least as efficient than the standard implementation. // // Performance note: it's more efficient to write functions which take "const // String&" arguments than "const char *" if you assign the argument to // another string. // // It was compiled and tested under Win32, Linux (libc 5 & 6), Solaris 5.5. // // To do: // - ressource support (string tables in ressources) // - more wide character (UNICODE) support // - regular expressions support // --------------------------------------------------------------------------- #ifdef WXSTRING_IS_WXOBJECT class WXDLLEXPORT wxString : public wxObject { DECLARE_DYNAMIC_CLASS(wxString) #else //WXSTRING_IS_WXOBJECT class WXDLLEXPORT wxString { #endif //WXSTRING_IS_WXOBJECT friend class WXDLLEXPORT wxArrayString; // NB: special care was taken in arranging the member functions in such order // that all inline functions can be effectively inlined, verify that all // performace critical functions are still inlined if you change order! private: // points to data preceded by wxStringData structure with ref count info wxChar *m_pchData; // accessor to string data wxStringData* GetStringData() const { return (wxStringData*)m_pchData - 1; } // string (re)initialization functions // initializes the string to the empty value (must be called only from // ctors, use Reinit() otherwise) void Init() { m_pchData = (wxChar *)wxEmptyString; } // initializaes the string with (a part of) C-string void InitWith(const wxChar *psz, size_t nPos = 0, size_t nLen = wxSTRING_MAXLEN); // as Init, but also frees old data void Reinit() { GetStringData()->Unlock(); Init(); } // memory allocation // allocates memory for string of lenght nLen void AllocBuffer(size_t nLen); // copies data to another string void AllocCopy(wxString&, int, int) const; // effectively copies data to string void AssignCopy(size_t, const wxChar *); // append a (sub)string void ConcatSelf(int nLen, const wxChar *src); // functions called before writing to the string: they copy it if there // are other references to our data (should be the only owner when writing) void CopyBeforeWrite(); void AllocBeforeWrite(size_t); // this method is not implemented - there is _no_ conversion from int to // string, you're doing something wrong if the compiler wants to call it! // // try `s << i' or `s.Printf("%d", i)' instead wxString(int); wxString(long); public: // constructors and destructor // ctor for an empty string wxString() { Init(); } // copy ctor wxString(const wxString& stringSrc) { wxASSERT( stringSrc.GetStringData()->IsValid() ); if ( stringSrc.IsEmpty() ) { // nothing to do for an empty string Init(); } else { m_pchData = stringSrc.m_pchData; // share same data GetStringData()->Lock(); // => one more copy } } // string containing nRepeat copies of ch wxString(wxChar ch, size_t nRepeat = 1); // ctor takes first nLength characters from C string // (default value of wxSTRING_MAXLEN means take all the string) wxString(const wxChar *psz, size_t nLength = wxSTRING_MAXLEN) { InitWith(psz, 0, nLength); } #if wxUSE_UNICODE // from multibyte string // (NB: nLength is right now number of Unicode characters, not // characters in psz! So try not to use it yet!) wxString(const char *psz, wxMBConv& conv = wxConvLibc, size_t nLength = wxSTRING_MAXLEN); // from wxWCharBuffer (i.e. return from wxGetString) wxString(const wxWCharBuffer& psz) { InitWith(psz, 0, wxSTRING_MAXLEN); } #else // ANSI // from C string (for compilers using unsigned char) wxString(const unsigned char* psz, size_t nLength = wxSTRING_MAXLEN) { InitWith((const char*)psz, 0, nLength); } // from multibyte string wxString(const char *psz, wxMBConv& WXUNUSED(conv) , size_t nLength = wxSTRING_MAXLEN) { InitWith(psz, 0, nLength); } #if wxUSE_WCHAR_T // from wide (Unicode) string wxString(const wchar_t *pwz); #endif // !wxUSE_WCHAR_T // from wxCharBuffer wxString(const wxCharBuffer& psz) { InitWith(psz, 0, wxSTRING_MAXLEN); } #endif // Unicode/ANSI // dtor is not virtual, this class must not be inherited from! ~wxString() { GetStringData()->Unlock(); } // generic attributes & operations // as standard strlen() size_t Len() const { return GetStringData()->nDataLength; } // string contains any characters? bool IsEmpty() const { return Len() == 0; } // empty string is "FALSE", so !str will return TRUE bool operator!() const { return IsEmpty(); } // empty string contents void Empty() { if ( !IsEmpty() ) Reinit(); // should be empty wxASSERT( GetStringData()->nDataLength == 0 ); } // empty the string and free memory void Clear() { if ( !GetStringData()->IsEmpty() ) Reinit(); wxASSERT( GetStringData()->nDataLength == 0 ); // should be empty wxASSERT( GetStringData()->nAllocLength == 0 ); // and not own any memory } // contents test // Is an ascii value bool IsAscii() const; // Is a number bool IsNumber() const; // Is a word bool IsWord() const; // data access (all indexes are 0 based) // read access wxChar GetChar(size_t n) const { ASSERT_VALID_INDEX( n ); return m_pchData[n]; } // read/write access wxChar& GetWritableChar(size_t n) { ASSERT_VALID_INDEX( n ); CopyBeforeWrite(); return m_pchData[n]; } // write access void SetChar(size_t n, wxChar ch) { ASSERT_VALID_INDEX( n ); CopyBeforeWrite(); m_pchData[n] = ch; } // get last character wxChar Last() const { wxASSERT( !IsEmpty() ); return m_pchData[Len() - 1]; } // get writable last character wxChar& Last() { wxASSERT( !IsEmpty() ); CopyBeforeWrite(); return m_pchData[Len()-1]; } // operator version of GetChar wxChar operator[](size_t n) const { ASSERT_VALID_INDEX( n ); return m_pchData[n]; } // operator version of GetChar wxChar operator[](int n) const { ASSERT_VALID_INDEX( n ); return m_pchData[n]; } #ifdef __alpha__ // operator version of GetChar wxChar operator[](unsigned int n) const { ASSERT_VALID_INDEX( n ); return m_pchData[n]; } #endif // operator version of GetWriteableChar wxChar& operator[](size_t n) { ASSERT_VALID_INDEX( n ); CopyBeforeWrite(); return m_pchData[n]; } #ifdef __alpha__ // operator version of GetWriteableChar wxChar& operator[](unsigned int n) { ASSERT_VALID_INDEX( n ); CopyBeforeWrite(); return m_pchData[n]; } #endif // implicit conversion to C string operator const wxChar*() const { return m_pchData; } // explicit conversion to C string (use this with printf()!) const wxChar* c_str() const { return m_pchData; } // (and this with [wx]Printf()!) const wxChar* wx_str() const { return m_pchData; } // identical to c_str() const wxChar* GetData() const { return m_pchData; } // conversions with (possible) format convertions: have to return a // buffer with temporary data #if wxUSE_UNICODE const wxCharBuffer mb_str(wxMBConv& conv = wxConvLibc) const { return conv.cWC2MB(m_pchData); } const wxWX2MBbuf mbc_str() const { return mb_str(*wxConvCurrent); } const wxChar* wc_str(wxMBConv& WXUNUSED(conv) = wxConvLibc) const { return m_pchData; } #if wxMBFILES const wxCharBuffer fn_str() const { return mb_str(wxConvFile); } #else // !wxMBFILES const wxChar* fn_str() const { return m_pchData; } #endif // wxMBFILES/!wxMBFILES #else // ANSI #if wxUSE_MULTIBYTE const wxChar* mb_str(wxMBConv& WXUNUSED(conv) = wxConvLibc) const { return m_pchData; } const wxWX2MBbuf mbc_str() const { return mb_str(*wxConvCurrent); } #else // !mmultibyte const wxChar* mb_str() const { return m_pchData; } const wxWX2MBbuf mbc_str() const { return mb_str(); } #endif // multibyte/!multibyte #if wxUSE_WCHAR_T const wxWCharBuffer wc_str(wxMBConv& conv) const { return conv.cMB2WC(m_pchData); } #endif // wxUSE_WCHAR_T const wxChar* fn_str() const { return m_pchData; } #endif // Unicode/ANSI // overloaded assignment // from another wxString wxString& operator=(const wxString& stringSrc); // from a character wxString& operator=(wxChar ch); // from a C string wxString& operator=(const wxChar *psz); #if wxUSE_UNICODE // from wxWCharBuffer wxString& operator=(const wxWCharBuffer& psz) { return operator=((const wchar_t *)psz); } #else // ANSI // from another kind of C string wxString& operator=(const unsigned char* psz); #if wxUSE_WCHAR_T // from a wide string wxString& operator=(const wchar_t *pwz); #endif // from wxCharBuffer wxString& operator=(const wxCharBuffer& psz) { return operator=((const char *)psz); } #endif // Unicode/ANSI // string concatenation // in place concatenation /* Concatenate and return the result. Note that the left to right associativity of << allows to write things like "str << str1 << str2 << ..." (unlike with +=) */ // string += string wxString& operator<<(const wxString& s) { wxASSERT( s.GetStringData()->IsValid() ); ConcatSelf(s.Len(), s); return *this; } // string += C string wxString& operator<<(const wxChar *psz) { ConcatSelf(wxStrlen(psz), psz); return *this; } // string += char wxString& operator<<(wxChar ch) { ConcatSelf(1, &ch); return *this; } // string += string void operator+=(const wxString& s) { (void)operator<<(s); } // string += C string void operator+=(const wxChar *psz) { (void)operator<<(psz); } // string += char void operator+=(wxChar ch) { (void)operator<<(ch); } // string += buffer (i.e. from wxGetString) #if wxUSE_UNICODE wxString& operator<<(const wxWCharBuffer& s) { (void)operator<<((const wchar_t *)s); return *this; } void operator+=(const wxWCharBuffer& s) { (void)operator<<((const wchar_t *)s); } #else wxString& operator<<(const wxCharBuffer& s) { (void)operator<<((const char *)s); return *this; } void operator+=(const wxCharBuffer& s) { (void)operator<<((const char *)s); } #endif // string += C string wxString& Append(const wxChar* psz) { ConcatSelf(wxStrlen(psz), psz); return *this; } // append count copies of given character wxString& Append(wxChar ch, size_t count = 1u) { wxString str(ch, count); return *this << str; } // prepend a string, return the string itself wxString& Prepend(const wxString& str) { *this = str + *this; return *this; } // non-destructive concatenation // friend wxString WXDLLEXPORT operator+(const wxString& string1, const wxString& string2); // friend wxString WXDLLEXPORT operator+(const wxString& string, wxChar ch); // friend wxString WXDLLEXPORT operator+(wxChar ch, const wxString& string); // friend wxString WXDLLEXPORT operator+(const wxString& string, const wxChar *psz); // friend wxString WXDLLEXPORT operator+(const wxChar *psz, const wxString& string); // stream-like functions // insert an int into string wxString& operator<<(int i); // insert a float into string wxString& operator<<(float f); // insert a double into string wxString& operator<<(double d); // string comparison // case-sensitive comparison (returns a value < 0, = 0 or > 0) int Cmp(const wxChar *psz) const { return wxStrcmp(c_str(), psz); } // same as Cmp() but not case-sensitive int CmpNoCase(const wxChar *psz) const { return wxStricmp(c_str(), psz); } // test for the string equality, either considering case or not // (if compareWithCase then the case matters) bool IsSameAs(const wxChar *psz, bool compareWithCase = TRUE) const { return (compareWithCase ? Cmp(psz) : CmpNoCase(psz)) == 0; } // comparison with a signle character: returns TRUE if equal bool IsSameAs(wxChar c, bool compareWithCase = TRUE) const { return (Len() == 1) && (compareWithCase ? GetChar(0u) == c : wxToupper(GetChar(0u)) == wxToupper(c)); } // simple sub-string extraction // return substring starting at nFirst of length nCount (or till the end // if nCount = default value) wxString Mid(size_t nFirst, size_t nCount = wxSTRING_MAXLEN) const; // operator version of Mid() wxString operator()(size_t start, size_t len) const { return Mid(start, len); } // get first nCount characters wxString Left(size_t nCount) const; // get last nCount characters wxString Right(size_t nCount) const; // get all characters before the first occurence of ch // (returns the whole string if ch not found) wxString BeforeFirst(wxChar ch) const; // get all characters before the last occurence of ch // (returns empty string if ch not found) wxString BeforeLast(wxChar ch) const; // get all characters after the first occurence of ch // (returns empty string if ch not found) wxString AfterFirst(wxChar ch) const; // get all characters after the last occurence of ch // (returns the whole string if ch not found) wxString AfterLast(wxChar ch) const; // for compatibility only, use more explicitly named functions above wxString Before(wxChar ch) const { return BeforeLast(ch); } wxString After(wxChar ch) const { return AfterFirst(ch); } // case conversion // convert to upper case in place, return the string itself wxString& MakeUpper(); // convert to upper case, return the copy of the string // Here's something to remember: BC++ doesn't like returns in inlines. wxString Upper() const ; // convert to lower case in place, return the string itself wxString& MakeLower(); // convert to lower case, return the copy of the string wxString Lower() const ; // trimming/padding whitespace (either side) and truncating // remove spaces from left or from right (default) side wxString& Trim(bool bFromRight = TRUE); // add nCount copies chPad in the beginning or at the end (default) wxString& Pad(size_t nCount, wxChar chPad = wxT(' '), bool bFromRight = TRUE); // truncate string to given length wxString& Truncate(size_t uiLen); // searching and replacing // searching (return starting index, or -1 if not found) int Find(wxChar ch, bool bFromEnd = FALSE) const; // like strchr/strrchr // searching (return starting index, or -1 if not found) int Find(const wxChar *pszSub) const; // like strstr // replace first (or all of bReplaceAll) occurences of substring with // another string, returns the number of replacements made size_t Replace(const wxChar *szOld, const wxChar *szNew, bool bReplaceAll = TRUE); // check if the string contents matches a mask containing '*' and '?' bool Matches(const wxChar *szMask) const; // conversion to numbers: all functions return TRUE only if the whole string // is a number and put the value of this number into the pointer provided // convert to a signed integer bool ToLong(long *val) const; // convert to an unsigned integer bool ToULong(unsigned long *val) const; // convert to a double bool ToDouble(double *val) const; // formated input/output // as sprintf(), returns the number of characters written or < 0 on error int Printf(const wxChar *pszFormat, ...); // as vprintf(), returns the number of characters written or < 0 on error int PrintfV(const wxChar* pszFormat, va_list argptr); // returns the string containing the result of Printf() to it static wxString Format(const wxChar *pszFormat, ...); // the same as above, but takes a va_list static wxString FormatV(const wxChar *pszFormat, va_list argptr); // raw access to string memory // ensure that string has space for at least nLen characters // only works if the data of this string is not shared void Alloc(size_t nLen); // minimize the string's memory // only works if the data of this string is not shared void Shrink(); // get writable buffer of at least nLen bytes. Unget() *must* be called // a.s.a.p. to put string back in a reasonable state! wxChar *GetWriteBuf(size_t nLen); // call this immediately after GetWriteBuf() has been used void UngetWriteBuf(); // wxWindows version 1 compatibility functions // use Mid() wxString SubString(size_t from, size_t to) const { return Mid(from, (to - from + 1)); } // values for second parameter of CompareTo function enum caseCompare {exact, ignoreCase}; // values for first parameter of Strip function enum stripType {leading = 0x1, trailing = 0x2, both = 0x3}; // use Printf() int sprintf(const wxChar *pszFormat, ...); // use Cmp() inline int CompareTo(const wxChar* psz, caseCompare cmp = exact) const { return cmp == exact ? Cmp(psz) : CmpNoCase(psz); } // use Len size_t Length() const { return Len(); } // Count the number of characters int Freq(wxChar ch) const; // use MakeLower void LowerCase() { MakeLower(); } // use MakeUpper void UpperCase() { MakeUpper(); } // use Trim except that it doesn't change this string wxString Strip(stripType w = trailing) const; // use Find (more general variants not yet supported) size_t Index(const wxChar* psz) const { return Find(psz); } size_t Index(wxChar ch) const { return Find(ch); } // use Truncate wxString& Remove(size_t pos) { return Truncate(pos); } wxString& RemoveLast() { return Truncate(Len() - 1); } wxString& Remove(size_t nStart, size_t nLen) { return erase( nStart, nLen ); } // use Find() int First( const wxChar ch ) const { return Find(ch); } int First( const wxChar* psz ) const { return Find(psz); } int First( const wxString &str ) const { return Find(str); } int Last( const wxChar ch ) const { return Find(ch, TRUE); } bool Contains(const wxString& str) const { return Find(str) != -1; } // use IsEmpty() bool IsNull() const { return IsEmpty(); } #ifdef wxSTD_STRING_COMPATIBILITY // std::string compatibility functions // standard types typedef wxChar value_type; typedef const value_type *const_iterator; // an 'invalid' value for string index static const size_t npos; // constructors // take nLen chars starting at nPos wxString(const wxString& str, size_t nPos, size_t nLen) { wxASSERT( str.GetStringData()->IsValid() ); InitWith(str.c_str(), nPos, nLen == npos ? 0 : nLen); } // take all characters from pStart to pEnd wxString(const void *pStart, const void *pEnd); // lib.string.capacity // return the length of the string size_t size() const { return Len(); } // return the length of the string size_t length() const { return Len(); } // return the maximum size of the string size_t max_size() const { return wxSTRING_MAXLEN; } // resize the string, filling the space with c if c != 0 void resize(size_t nSize, wxChar ch = wxT('\0')); // delete the contents of the string void clear() { Empty(); } // returns true if the string is empty bool empty() const { return IsEmpty(); } // inform string about planned change in size void reserve(size_t size) { Alloc(size); } // lib.string.access // return the character at position n wxChar at(size_t n) const { return GetChar(n); } // returns the writable character at position n wxChar& at(size_t n) { return GetWritableChar(n); } // first valid index position const_iterator begin() const { return wx_str(); } // position one after the last valid one const_iterator end() const { return wx_str() + length(); } // lib.string.modifiers // append a string wxString& append(const wxString& str) { *this += str; return *this; } // append elements str[pos], ..., str[pos+n] wxString& append(const wxString& str, size_t pos, size_t n) { ConcatSelf(n, str.c_str() + pos); return *this; } // append first n (or all if n == npos) characters of sz wxString& append(const wxChar *sz, size_t n = npos) { ConcatSelf(n == npos ? wxStrlen(sz) : n, sz); return *this; } // append n copies of ch wxString& append(size_t n, wxChar ch) { return Pad(n, ch); } // same as `this_string = str' wxString& assign(const wxString& str) { return (*this) = str; } // same as ` = str[pos..pos + n] wxString& assign(const wxString& str, size_t pos, size_t n) { return *this = wxString((const wxChar *)str + pos, n); } // same as `= first n (or all if n == npos) characters of sz' wxString& assign(const wxChar *sz, size_t n = npos) { return *this = wxString(sz, n); } // same as `= n copies of ch' wxString& assign(size_t n, wxChar ch) { return *this = wxString(ch, n); } // insert another string wxString& insert(size_t nPos, const wxString& str); // insert n chars of str starting at nStart (in str) wxString& insert(size_t nPos, const wxString& str, size_t nStart, size_t n) { return insert(nPos, wxString((const wxChar *)str + nStart, n)); } // insert first n (or all if n == npos) characters of sz wxString& insert(size_t nPos, const wxChar *sz, size_t n = npos) { return insert(nPos, wxString(sz, n)); } // insert n copies of ch wxString& insert(size_t nPos, size_t n, wxChar ch) { return insert(nPos, wxString(ch, n)); } // delete characters from nStart to nStart + nLen wxString& erase(size_t nStart = 0, size_t nLen = npos); // replaces the substring of length nLen starting at nStart wxString& replace(size_t nStart, size_t nLen, const wxChar* sz); // replaces the substring with nCount copies of ch wxString& replace(size_t nStart, size_t nLen, size_t nCount, wxChar ch); // replaces a substring with another substring wxString& replace(size_t nStart, size_t nLen, const wxString& str, size_t nStart2, size_t nLen2); // replaces the substring with first nCount chars of sz wxString& replace(size_t nStart, size_t nLen, const wxChar* sz, size_t nCount); // swap two strings void swap(wxString& str); // All find() functions take the nStart argument which specifies the // position to start the search on, the default value is 0. All functions // return npos if there were no match. // find a substring size_t find(const wxString& str, size_t nStart = 0) const; // VC++ 1.5 can't cope with this syntax. #if !defined(__VISUALC__) || defined(__WIN32__) // find first n characters of sz size_t find(const wxChar* sz, size_t nStart = 0, size_t n = npos) const; #endif // Gives a duplicate symbol (presumably a case-insensitivity problem) #if !defined(__BORLANDC__) // find the first occurence of character ch after nStart size_t find(wxChar ch, size_t nStart = 0) const; #endif // rfind() family is exactly like find() but works right to left // as find, but from the end size_t rfind(const wxString& str, size_t nStart = npos) const; // VC++ 1.5 can't cope with this syntax. #if !defined(__VISUALC__) || defined(__WIN32__) // as find, but from the end size_t rfind(const wxChar* sz, size_t nStart = npos, size_t n = npos) const; // as find, but from the end size_t rfind(wxChar ch, size_t nStart = npos) const; #endif // find first/last occurence of any character in the set // as strpbrk() but starts at nStart, returns npos if not found size_t find_first_of(const wxString& str, size_t nStart = 0) const { return find_first_of(str.c_str(), nStart); } // same as above size_t find_first_of(const wxChar* sz, size_t nStart = 0) const; // same as find(char, size_t) size_t find_first_of(wxChar c, size_t nStart = 0) const { return find(c, nStart); } // find the last (starting from nStart) char from str in this string size_t find_last_of (const wxString& str, size_t nStart = npos) const { return find_last_of(str.c_str(), nStart); } // same as above size_t find_last_of (const wxChar* sz, size_t nStart = npos) const; // same as above size_t find_last_of(wxChar c, size_t nStart = npos) const { return rfind(c, nStart); } // find first/last occurence of any character not in the set // as strspn() (starting from nStart), returns npos on failure size_t find_first_not_of(const wxString& str, size_t nStart = 0) const { return find_first_not_of(str.c_str(), nStart); } // same as above size_t find_first_not_of(const wxChar* sz, size_t nStart = 0) const; // same as above size_t find_first_not_of(wxChar ch, size_t nStart = 0) const; // as strcspn() size_t find_last_not_of(const wxString& str, size_t nStart=npos) const; // same as above size_t find_last_not_of(const wxChar* sz, size_t nStart = npos) const; // same as above size_t find_last_not_of(wxChar ch, size_t nStart = npos) const; // All compare functions return -1, 0 or 1 if the [sub]string is less, // equal or greater than the compare() argument. // just like strcmp() int compare(const wxString& str) const { return Cmp(str); } // comparison with a substring int compare(size_t nStart, size_t nLen, const wxString& str) const; // comparison of 2 substrings int compare(size_t nStart, size_t nLen, const wxString& str, size_t nStart2, size_t nLen2) const; // just like strcmp() int compare(const wxChar* sz) const { return Cmp(sz); } // substring comparison with first nCount characters of sz int compare(size_t nStart, size_t nLen, const wxChar* sz, size_t nCount = npos) const; // substring extraction wxString substr(size_t nStart = 0, size_t nLen = npos) const { return Mid(nStart, nLen); } #endif // wxSTD_STRING_COMPATIBILITY }; // ---------------------------------------------------------------------------- // The string array uses it's knowledge of internal structure of the wxString // class to optimize string storage. Normally, we would store pointers to // string, but as wxString is, in fact, itself a pointer (sizeof(wxString) is // sizeof(char *)) we store these pointers instead. The cast to "wxString *" is // really all we need to turn such pointer into a string! // // Of course, it can be called a dirty hack, but we use twice less memory and // this approach is also more speed efficient, so it's probably worth it. // // Usage notes: when a string is added/inserted, a new copy of it is created, // so the original string may be safely deleted. When a string is retrieved // from the array (operator[] or Item() method), a reference is returned. // ---------------------------------------------------------------------------- class WXDLLEXPORT wxArrayString { public: // type of function used by wxArrayString::Sort() typedef int (*CompareFunction)(const wxString& first, const wxString& second); // constructors and destructor // default ctor: if autoSort is TRUE, the array is always sorted (in // alphabetical order) wxArrayString(bool autoSort = FALSE); // copy ctor wxArrayString(const wxArrayString& array); // assignment operator wxArrayString& operator=(const wxArrayString& src); // not virtual, this class should not be derived from ~wxArrayString(); // memory management // empties the list, but doesn't release memory void Empty(); // empties the list and releases memory void Clear(); // preallocates memory for given number of items void Alloc(size_t nCount); // minimzes the memory usage (by freeing all extra memory) void Shrink(); // simple accessors // number of elements in the array size_t GetCount() const { return m_nCount; } // is it empty? bool IsEmpty() const { return m_nCount == 0; } // number of elements in the array (GetCount is preferred API) size_t Count() const { return m_nCount; } // items access (range checking is done in debug version) // get item at position uiIndex wxString& Item(size_t nIndex) const { wxASSERT( nIndex < m_nCount ); return *(wxString *)&(m_pItems[nIndex]); } // same as Item() wxString& operator[](size_t nIndex) const { return Item(nIndex); } // get last item wxString& Last() const { wxASSERT( !IsEmpty() ); return Item(Count() - 1); } // item management // Search the element in the array, starting from the beginning if // bFromEnd is FALSE or from end otherwise. If bCase, comparison is case // sensitive (default). Returns index of the first item matched or // wxNOT_FOUND int Index (const wxChar *sz, bool bCase = TRUE, bool bFromEnd = FALSE) const; // add new element at the end (if the array is not sorted), return its // index size_t Add(const wxString& str); // add new element at given position void Insert(const wxString& str, size_t uiIndex); // remove first item matching this value void Remove(const wxChar *sz); // remove item by index void Remove(size_t nIndex); // sorting // sort array elements in alphabetical order (or reversed alphabetical // order if reverseOrder parameter is TRUE) void Sort(bool reverseOrder = FALSE); // sort array elements using specified comparaison function void Sort(CompareFunction compareFunction); protected: void Copy(const wxArrayString& src); // copies the contents of another array private: void Grow(); // makes array bigger if needed void Free(); // free all the strings stored void DoSort(); // common part of all Sort() variants size_t m_nSize, // current size of the array m_nCount; // current number of elements wxChar **m_pItems; // pointer to data bool m_autoSort; // if TRUE, keep the array always sorted }; class WXDLLEXPORT wxSortedArrayString : public wxArrayString { public: wxSortedArrayString() : wxArrayString(TRUE) { } wxSortedArrayString(const wxArrayString& array) : wxArrayString(TRUE) { Copy(array); } }; // --------------------------------------------------------------------------- // wxString comparison functions: operator versions are always case sensitive // --------------------------------------------------------------------------- // inline bool operator==(const wxString& s1, const wxString& s2) { return (s1.Cmp(s2) == 0); } // inline bool operator==(const wxString& s1, const wxChar * s2) { return (s1.Cmp(s2) == 0); } // inline bool operator==(const wxChar * s1, const wxString& s2) { return (s2.Cmp(s1) == 0); } // inline bool operator!=(const wxString& s1, const wxString& s2) { return (s1.Cmp(s2) != 0); } // inline bool operator!=(const wxString& s1, const wxChar * s2) { return (s1.Cmp(s2) != 0); } // inline bool operator!=(const wxChar * s1, const wxString& s2) { return (s2.Cmp(s1) != 0); } // inline bool operator< (const wxString& s1, const wxString& s2) { return (s1.Cmp(s2) < 0); } // inline bool operator< (const wxString& s1, const wxChar * s2) { return (s1.Cmp(s2) < 0); } // inline bool operator< (const wxChar * s1, const wxString& s2) { return (s2.Cmp(s1) > 0); } // inline bool operator> (const wxString& s1, const wxString& s2) { return (s1.Cmp(s2) > 0); } // inline bool operator> (const wxString& s1, const wxChar * s2) { return (s1.Cmp(s2) > 0); } // inline bool operator> (const wxChar * s1, const wxString& s2) { return (s2.Cmp(s1) < 0); } // inline bool operator<=(const wxString& s1, const wxString& s2) { return (s1.Cmp(s2) <= 0); } // inline bool operator<=(const wxString& s1, const wxChar * s2) { return (s1.Cmp(s2) <= 0); } // inline bool operator<=(const wxChar * s1, const wxString& s2) { return (s2.Cmp(s1) >= 0); } // inline bool operator>=(const wxString& s1, const wxString& s2) { return (s1.Cmp(s2) >= 0); } // inline bool operator>=(const wxString& s1, const wxChar * s2) { return (s1.Cmp(s2) >= 0); } // inline bool operator>=(const wxChar * s1, const wxString& s2) { return (s2.Cmp(s1) <= 0); } // comparison with char inline bool operator==(wxChar c, const wxString& s) { return s.IsSameAs(c); } inline bool operator==(const wxString& s, wxChar c) { return s.IsSameAs(c); } inline bool operator!=(wxChar c, const wxString& s) { return !s.IsSameAs(c); } inline bool operator!=(const wxString& s, wxChar c) { return !s.IsSameAs(c); } #if wxUSE_UNICODE inline bool operator==(const wxString& s1, const wxWCharBuffer& s2) { return (s1.Cmp((const wchar_t *)s2) == 0); } inline bool operator==(const wxWCharBuffer& s1, const wxString& s2) { return (s2.Cmp((const wchar_t *)s1) == 0); } #else inline bool operator==(const wxString& s1, const wxCharBuffer& s2) { return (s1.Cmp((const char *)s2) == 0); } inline bool operator==(const wxCharBuffer& s1, const wxString& s2) { return (s2.Cmp((const char *)s1) == 0); } #endif wxString WXDLLEXPORT operator+(const wxString& string1, const wxString& string2); wxString WXDLLEXPORT operator+(const wxString& string, wxChar ch); wxString WXDLLEXPORT operator+(wxChar ch, const wxString& string); wxString WXDLLEXPORT operator+(const wxString& string, const wxChar *psz); wxString WXDLLEXPORT operator+(const wxChar *psz, const wxString& string); #if wxUSE_UNICODE inline wxString WXDLLEXPORT operator+(const wxString& string, const wxWCharBuffer& buf) { return string + (const wchar_t *)buf; } inline wxString WXDLLEXPORT operator+(const wxWCharBuffer& buf, const wxString& string) { return (const wchar_t *)buf + string; } #else inline wxString WXDLLEXPORT operator+(const wxString& string, const wxCharBuffer& buf) { return string + (const char *)buf; } inline wxString WXDLLEXPORT operator+(const wxCharBuffer& buf, const wxString& string) { return (const char *)buf + string; } #endif // --------------------------------------------------------------------------- // Implementation only from here until the end of file // --------------------------------------------------------------------------- // don't pollute the library user's name space #undef ASSERT_VALID_INDEX #if defined(wxSTD_STRING_COMPATIBILITY) && wxUSE_STD_IOSTREAM #include "wx/ioswrap.h" WXDLLEXPORT istream& operator>>(istream&, wxString&); WXDLLEXPORT ostream& operator<<(ostream&, const wxString&); #endif // wxSTD_STRING_COMPATIBILITY #endif // _WX_WXSTRINGH__