X-Git-Url: https://git.saurik.com/wxWidgets.git/blobdiff_plain/4fdfe2f3ae110c1f1ba2d925749cac5b761eff4b..4b5d2be39657bb364c9111c4d13b1eecb8f3b69f:/src/common/string.cpp diff --git a/src/common/string.cpp b/src/common/string.cpp index d36f85bd5a..595caf7125 100644 --- a/src/common/string.cpp +++ b/src/common/string.cpp @@ -35,10 +35,6 @@ #include #include -#ifdef __SALFORDC__ - #include -#endif - #include "wx/hashmap.h" // string handling functions used by wxString: @@ -62,6 +58,105 @@ //According to STL _must_ be a -1 size_t const size_t wxString::npos = (size_t) -1; +#if wxUSE_STRING_POS_CACHE + +#ifdef wxHAS_COMPILER_TLS + +wxTLS_TYPE(wxString::Cache) wxString::ms_cache; + +#else // !wxHAS_COMPILER_TLS + +struct wxStrCacheInitializer +{ + wxStrCacheInitializer() + { + // calling this function triggers s_cache initialization in it, and + // from now on it becomes safe to call from multiple threads + wxString::GetCache(); + } +}; + +/* +wxString::Cache& wxString::GetCache() +{ + static wxTLS_TYPE(Cache) s_cache; + + return wxTLS_VALUE(s_cache); +} +*/ + +static wxStrCacheInitializer gs_stringCacheInit; + +#endif // wxHAS_COMPILER_TLS/!wxHAS_COMPILER_TLS + +// gdb seems to be unable to display thread-local variables correctly, at least +// not my 6.4.98 version under amd64, so provide this debugging helper to do it +#ifdef __WXDEBUG__ + +struct wxStrCacheDumper +{ + static void ShowAll() + { + puts("*** wxString cache dump:"); + for ( unsigned n = 0; n < wxString::Cache::SIZE; n++ ) + { + const wxString::Cache::Element& + c = wxString::GetCacheBegin()[n]; + + printf("\t%u%s\t%p: pos=(%lu, %lu), len=%ld\n", + n, + n == wxString::LastUsedCacheElement() ? " [*]" : "", + c.str, + (unsigned long)c.pos, + (unsigned long)c.impl, + (long)c.len); + } + } +}; + +void wxDumpStrCache() { wxStrCacheDumper::ShowAll(); } + +#endif // __WXDEBUG__ + +#ifdef wxPROFILE_STRING_CACHE + +wxString::CacheStats wxString::ms_cacheStats; + +struct wxStrCacheStatsDumper +{ + ~wxStrCacheStatsDumper() + { + const wxString::CacheStats& stats = wxString::ms_cacheStats; + + if ( stats.postot ) + { + puts("*** wxString cache statistics:"); + printf("\tTotal non-trivial calls to PosToImpl(): %u\n", + stats.postot); + printf("\tHits %u (of which %u not used) or %.2f%%\n", + stats.poshits, + stats.mishits, + 100.*float(stats.poshits - stats.mishits)/stats.postot); + printf("\tAverage position requested: %.2f\n", + float(stats.sumpos) / stats.postot); + printf("\tAverage offset after cached hint: %.2f\n", + float(stats.sumofs) / stats.postot); + } + + if ( stats.lentot ) + { + printf("\tNumber of calls to length(): %u, hits=%.2f%%\n", + stats.lentot, 100.*float(stats.lenhits)/stats.lentot); + } + } +}; + +static wxStrCacheStatsDumper s_showCacheStats; + +#endif // wxPROFILE_STRING_CACHE + +#endif // wxUSE_STRING_POS_CACHE + // ---------------------------------------------------------------------------- // global functions // ---------------------------------------------------------------------------- @@ -72,11 +167,10 @@ const size_t wxString::npos = (size_t) -1; wxSTD ostream& operator<<(wxSTD ostream& os, const wxCStrData& str) { -// FIXME-UTF8: always, not only if wxUSE_UNICODE -#if wxUSE_UNICODE && !defined(__BORLANDC__) - return os << (const wchar_t*)str.AsWCharBuf(); +#if wxUSE_UNICODE && !wxUSE_UNICODE_UTF8 + return os << (const char *)str.AsCharBuf(); #else - return os << (const char*)str.AsCharBuf(); + return os << str.AsInternal(); #endif } @@ -97,6 +191,25 @@ wxSTD ostream& operator<<(wxSTD ostream& os, const wxWCharBuffer& str) } #endif +#if wxUSE_UNICODE && defined(HAVE_WOSTREAM) + +wxSTD wostream& operator<<(wxSTD wostream& wos, const wxString& str) +{ + return wos << str.wc_str(); +} + +wxSTD wostream& operator<<(wxSTD wostream& wos, const wxCStrData& str) +{ + return wos << str.AsWChar(); +} + +wxSTD wostream& operator<<(wxSTD wostream& wos, const wxWCharBuffer& str) +{ + return wos << str.data(); +} + +#endif // wxUSE_UNICODE && defined(HAVE_WOSTREAM) + #endif // wxUSE_STD_IOSTREAM // =========================================================================== @@ -109,22 +222,30 @@ void wxString::PosLenToImpl(size_t pos, size_t len, size_t *implPos, size_t *implLen) const { if ( pos == npos ) + { *implPos = npos; - else + } + else // have valid start position { - const_iterator i = begin() + pos; - *implPos = wxStringImpl::const_iterator(i.impl()) - m_impl.begin(); + const const_iterator b = GetIterForNthChar(pos); + *implPos = wxStringImpl::const_iterator(b.impl()) - m_impl.begin(); if ( len == npos ) + { *implLen = npos; - else + } + else // have valid length too { - // too large length is interpreted as "to the end of the string" - // FIXME-UTF8: verify this is the case in std::string, assert - // otherwise - if ( pos + len > length() ) - len = length() - pos; - - *implLen = (i + len).impl() - i.impl(); + // we need to handle the case of length specifying a substring + // going beyond the end of the string, just as std::string does + const const_iterator e(end()); + const_iterator i(b); + while ( len && i <= e ) + { + ++i; + --len; + } + + *implLen = i.impl() - b.impl(); } } } @@ -233,9 +354,35 @@ const char* wxCStrData::AsChar() const wxString *str = wxConstCast(m_str, wxString); // convert the string: + // + // FIXME-UTF8: we'd like to do the conversion in the existing buffer (if we + // have it) but it's unfortunately not obvious to implement + // because we don't know how big buffer do we need for the + // given string length (in case of multibyte encodings, e.g. + // ISO-2022-JP or UTF-8 when internal representation is wchar_t) + // + // One idea would be to store more than just m_convertedToChar + // in wxString: then we could record the length of the string + // which was converted the last time and try to reuse the same + // buffer if the current length is not greater than it (this + // could still fail because string could have been modified in + // place but it would work most of the time, so we'd do it and + // only allocate the new buffer if in-place conversion returned + // an error). We could also store a bit saying if the string + // was modified since the last conversion (and update it in all + // operation modifying the string, of course) to avoid unneeded + // consequential conversions. But both of these ideas require + // adding more fields to wxString and require profiling results + // to be sure that we really gain enough from them to justify + // doing it. wxCharBuffer buf(str->mb_str()); - // FIXME-UTF8: do the conversion in-place in the existing buffer + // if it failed, return empty string and not NULL to avoid crashes in code + // written with either wxWidgets 2 wxString or std::string behaviour in + // mind: neither of them ever returns NULL and so we shouldn't neither + if ( !buf ) + return ""; + if ( str->m_convertedToChar && strlen(buf) == strlen(str->m_convertedToChar) ) { @@ -261,6 +408,10 @@ const wchar_t* wxCStrData::AsWChar() const // convert the string: wxWCharBuffer buf(str->wc_str()); + // notice that here, unlike above in AsChar(), conversion can't fail as our + // internal UTF-8 is always well-formed -- or the string was corrupted and + // all bets are off anyhow + // FIXME-UTF8: do the conversion in-place in the existing buffer if ( str->m_convertedToWChar && wxWcslen(buf) == wxWcslen(str->m_convertedToWChar) ) @@ -325,6 +476,9 @@ wxString::SubstrBufFromMB wxString::ConvertStr(const char *psz, size_t nLength, // UTF-8 sequence and psz may be invalid: if ( wxStringOperations::IsValidUtf8String(psz, nLength) ) { + // we must pass the real string length to SubstrBufFromMB ctor + if ( nLength == npos ) + nLength = psz ? strlen(psz) : 0; return SubstrBufFromMB(wxCharBuffer::CreateNonOwned(psz), nLength); } // else: do the roundtrip through wchar_t* @@ -407,7 +561,7 @@ const wxCharBuffer wxString::mb_str(const wxMBConv& conv) const if ( !wcLen ) return wxCharBuffer(""); - return conv.cWC2MB(wcBuf, wcLen, NULL); + return conv.cWC2MB(wcBuf, wcLen+1, NULL); } #else // ANSI @@ -945,13 +1099,12 @@ int wxString::CmpNoCase(const wxString& s) const { // FIXME-UTF8: use wxUniChar::ToLower/ToUpper once added - size_t idx = 0; const_iterator i1 = begin(); const_iterator end1 = end(); const_iterator i2 = s.begin(); const_iterator end2 = s.end(); - for ( ; i1 != end1 && i2 != end2; ++idx, ++i1, ++i2 ) + for ( ; i1 != end1 && i2 != end2; ++i1, ++i2 ) { wxUniChar lower1 = (wxChar)wxTolower(*i1); wxUniChar lower2 = (wxChar)wxTolower(*i2); @@ -986,7 +1139,7 @@ wxString wxString::FromAscii(const char *ascii, size_t len) wxString res; { - wxImplStringBuffer buf(res, len); + wxStringInternalBuffer buf(res, len); wxStringCharType *dest = buf; for ( ; len > 0; --len ) @@ -1007,7 +1160,7 @@ wxString wxString::FromAscii(const char *ascii) return FromAscii(ascii, wxStrlen(ascii)); } -wxString wxString::FromAscii(const char ascii) +wxString wxString::FromAscii(char ascii) { // What do we do with '\0' ? @@ -1191,35 +1344,59 @@ size_t wxString::Replace(const wxString& strOld, wxCHECK_MSG( !strOld.empty(), 0, _T("wxString::Replace(): invalid parameter") ); + wxSTRING_INVALIDATE_CACHE(); + size_t uiCount = 0; // count of replacements made - size_t uiOldLen = strOld.length(); - size_t uiNewLen = strNew.length(); + // optimize the special common case: replacement of one character by + // another one (in UTF-8 case we can only do this for ASCII characters) + // + // benchmarks show that this special version is around 3 times faster + // (depending on the proportion of matching characters and UTF-8/wchar_t + // build) + if ( strOld.m_impl.length() == 1 && strNew.m_impl.length() == 1 ) + { + const wxStringCharType chOld = strOld.m_impl[0], + chNew = strNew.m_impl[0]; + + // this loop is the simplified version of the one below + for ( size_t pos = 0; ; ) + { + pos = m_impl.find(chOld, pos); + if ( pos == npos ) + break; - size_t dwPos = 0; + m_impl[pos++] = chNew; - while ( (*this)[dwPos] != wxT('\0') ) + uiCount++; + + if ( !bReplaceAll ) + break; + } + } + else // general case { - //DO NOT USE STRSTR HERE - //this string can contain embedded null characters, - //so strstr will function incorrectly - dwPos = find(strOld, dwPos); - if ( dwPos == npos ) - break; // exit the loop - else + const size_t uiOldLen = strOld.m_impl.length(); + const size_t uiNewLen = strNew.m_impl.length(); + + for ( size_t pos = 0; ; ) { - //replace this occurance of the old string with the new one - replace(dwPos, uiOldLen, strNew, uiNewLen); + pos = m_impl.find(strOld.m_impl, pos); + if ( pos == npos ) + break; - //move up pos past the string that was replaced - dwPos += uiNewLen; + // replace this occurrence of the old string with the new one + m_impl.replace(pos, uiOldLen, strNew.m_impl); - //increase replace count - ++uiCount; + // move up pos past the string that was replaced + pos += uiNewLen; - // stop now? + // increase replace count + uiCount++; + + // stop after the first one? if ( !bReplaceAll ) - break; // exit the loop + break; } } @@ -1295,14 +1472,28 @@ wxString& wxString::MakeLower() return *this; } +wxString& wxString::MakeCapitalized() +{ + const iterator en = end(); + iterator it = begin(); + if ( it != en ) + { + *it = (wxChar)wxToupper(*it); + for ( ++it; it != en; ++it ) + *it = (wxChar)wxTolower(*it); + } + + return *this; +} + // --------------------------------------------------------------------------- // trimming and padding // --------------------------------------------------------------------------- // some compilers (VC++ 6.0 not to name them) return true for a call to -// isspace('ê') in the C locale which seems to be broken to me, but we have to -// live with this by checking that the character is a 7 bit one - even if this -// may fail to detect some spaces (I don't know if Unicode doesn't have +// isspace('\xEA') in the C locale which seems to be broken to me, but we have +// to live with this by checking that the character is a 7 bit one - even if +// this may fail to detect some spaces (I don't know if Unicode doesn't have // space-like symbols somewhere except in the first 128 chars), it is arguably // still better than trimming away accented letters inline int wxSafeIsspace(wxChar ch) { return (ch < 127) && wxIsspace(ch); } @@ -1323,7 +1514,7 @@ wxString& wxString::Trim(bool bFromRight) // find last non-space character reverse_iterator psz = rbegin(); while ( (psz != rend()) && wxSafeIsspace(*psz) ) - psz++; + ++psz; // truncate at trailing space start erase(psz.base(), end()); @@ -1333,7 +1524,7 @@ wxString& wxString::Trim(bool bFromRight) // find first non-space character iterator psz = begin(); while ( (psz != end()) && wxSafeIsspace(*psz) ) - psz++; + ++psz; // fix up data and length erase(begin(), psz); @@ -1391,70 +1582,68 @@ int wxString::Find(wxUniChar ch, bool bFromEnd) const // it out. Note that number extraction works correctly on UTF-8 strings, so // we can use wxStringCharType and wx_str() for maximum efficiency. -template -bool wxStringToIntType(const wxStringCharType *start, - T *val, - int base, - T (*func)(const wxStringCharType*, wxStringCharType**, int)) -{ - wxCHECK_MSG( val, false, _T("NULL output pointer") ); - wxASSERT_MSG( !base || (base > 1 && base <= 36), _T("invalid base") ); - #ifndef __WXWINCE__ - errno = 0; + #define DO_IF_NOT_WINCE(x) x +#else + #define DO_IF_NOT_WINCE(x) #endif - wxStringCharType *end; - *val = (*func)(start, &end, base); +#define WX_STRING_TO_INT_TYPE(out, base, func, T) \ + wxCHECK_MSG( out, false, _T("NULL output pointer") ); \ + wxASSERT_MSG( !base || (base > 1 && base <= 36), _T("invalid base") ); \ + \ + DO_IF_NOT_WINCE( errno = 0; ) \ + \ + const wxStringCharType *start = wx_str(); \ + wxStringCharType *end; \ + T val = func(start, &end, base); \ + \ + /* return true only if scan was stopped by the terminating NUL and */ \ + /* if the string was not empty to start with and no under/overflow */ \ + /* occurred: */ \ + if ( *end || end == start DO_IF_NOT_WINCE(|| errno == ERANGE) ) \ + return false; \ + *out = val; \ + return true - // return true only if scan was stopped by the terminating NUL and if the - // string was not empty to start with and no under/overflow occurred - return !*end && (end != start) -#ifndef __WXWINCE__ - && (errno != ERANGE) -#endif - ; -} - -bool wxString::ToLong(long *val, int base) const +bool wxString::ToLong(long *pVal, int base) const { - return wxStringToIntType(wx_str(), val, base, wxStrtol); + WX_STRING_TO_INT_TYPE(pVal, base, wxStrtol, long); } -bool wxString::ToULong(unsigned long *val, int base) const +bool wxString::ToULong(unsigned long *pVal, int base) const { - return wxStringToIntType(wx_str(), val, base, wxStrtoul); + WX_STRING_TO_INT_TYPE(pVal, base, wxStrtoul, unsigned long); } -bool wxString::ToLongLong(wxLongLong_t *val, int base) const +bool wxString::ToLongLong(wxLongLong_t *pVal, int base) const { - return wxStringToIntType(wx_str(), val, base, wxStrtoll); + WX_STRING_TO_INT_TYPE(pVal, base, wxStrtoll, wxLongLong_t); } -bool wxString::ToULongLong(wxULongLong_t *val, int base) const +bool wxString::ToULongLong(wxULongLong_t *pVal, int base) const { - return wxStringToIntType(wx_str(), val, base, wxStrtoull); + WX_STRING_TO_INT_TYPE(pVal, base, wxStrtoull, wxULongLong_t); } -bool wxString::ToDouble(double *val) const +bool wxString::ToDouble(double *pVal) const { - wxCHECK_MSG( val, false, _T("NULL pointer in wxString::ToDouble") ); + wxCHECK_MSG( pVal, false, _T("NULL output pointer") ); -#ifndef __WXWINCE__ - errno = 0; -#endif + DO_IF_NOT_WINCE( errno = 0; ) const wxChar *start = c_str(); wxChar *end; - *val = wxStrtod(start, &end); + double val = wxStrtod(start, &end); // return true only if scan was stopped by the terminating NUL and if the // string was not empty to start with and no under/overflow occurred - return !*end && (end != start) -#ifndef __WXWINCE__ - && (errno != ERANGE) -#endif - ; + if ( *end || end == start DO_IF_NOT_WINCE(|| errno == ERANGE) ) + return false; + + *pVal = val; + + return true; } // --------------------------------------------------------------------------- @@ -1546,6 +1735,60 @@ int wxString::DoPrintfUtf8(const char *format, ...) } #endif // wxUSE_UNICODE_UTF8 +/* + Uses wxVsnprintf and places the result into the this string. + + In ANSI build, wxVsnprintf is effectively vsnprintf but in Unicode build + it is vswprintf. Due to a discrepancy between vsnprintf and vswprintf in + the ISO C99 (and thus SUSv3) standard the return value for the case of + an undersized buffer is inconsistent. For conforming vsnprintf + implementations the function must return the number of characters that + would have been printed had the buffer been large enough. For conforming + vswprintf implementations the function must return a negative number + and set errno. + + What vswprintf sets errno to is undefined but Darwin seems to set it to + EOVERFLOW. The only expected errno are EILSEQ and EINVAL. Both of + those are defined in the standard and backed up by several conformance + statements. Note that ENOMEM mentioned in the manual page does not + apply to swprintf, only wprintf and fwprintf. + + Official manual page: + http://www.opengroup.org/onlinepubs/009695399/functions/swprintf.html + + Some conformance statements (AIX, Solaris): + http://www.opengroup.org/csq/view.mhtml?RID=ibm%2FSD1%2F3 + http://www.theopengroup.org/csq/view.mhtml?norationale=1&noreferences=1&RID=Fujitsu%2FSE2%2F10 + + Since EILSEQ and EINVAL are rather common but EOVERFLOW is not and since + EILSEQ and EINVAL are specifically defined to mean the error is other than + an undersized buffer and no other errno are defined we treat those two + as meaning hard errors and everything else gets the old behavior which + is to keep looping and increasing buffer size until the function succeeds. + + In practice it's impossible to determine before compilation which behavior + may be used. The vswprintf function may have vsnprintf-like behavior or + vice-versa. Behavior detected on one release can theoretically change + with an updated release. Not to mention that configure testing for it + would require the test to be run on the host system, not the build system + which makes cross compilation difficult. Therefore, we make no assumptions + about behavior and try our best to handle every known case, including the + case where wxVsnprintf returns a negative number and fails to set errno. + + There is yet one more non-standard implementation and that is our own. + Fortunately, that can be detected at compile-time. + + On top of all that, ISO C99 explicitly defines snprintf to write a null + character to the last position of the specified buffer. That would be at + at the given buffer size minus 1. It is supposed to do this even if it + turns out that the buffer is sized too small. + + Darwin (tested on 10.5) follows the C99 behavior exactly. + + Glibc 2.6 almost follows the C99 behavior except vswprintf never sets + errno even when it fails. However, it only seems to ever fail due + to an undersized buffer. +*/ #if wxUSE_UNICODE_UTF8 template #else @@ -1583,12 +1826,19 @@ static int DoStringPrintfV(wxString& str, // only a copy va_list argptrcopy; wxVaCopy(argptrcopy, argptr); + +#ifndef __WXWINCE__ + // Set errno to 0 to make it determinate if wxVsnprintf fails to set it. + errno = 0; +#endif int len = wxVsnprintf(buf, size, format, argptrcopy); va_end(argptrcopy); // some implementations of vsnprintf() don't NUL terminate // the string if there is not enough space for it so // always do it manually + // FIXME: This really seems to be the wrong and would be an off-by-one + // bug except the code above allocates an extra character. buf[size] = _T('\0'); // vsnprintf() may return either -1 (traditional Unix behaviour) or the @@ -1610,19 +1860,33 @@ static int DoStringPrintfV(wxString& str, // assume it only returns error if there is not enough space, but // as we don't know how much we need, double the current size of // the buffer - size *= 2; +#ifndef __WXWINCE__ + if( (errno == EILSEQ) || (errno == EINVAL) ) + // If errno was set to one of the two well-known hard errors + // then fail immediately to avoid an infinite loop. + return -1; + else +#endif // __WXWINCE__ + // still not enough, as we don't know how much we need, double the + // current size of the buffer + size *= 2; #endif // wxUSE_WXVSNPRINTF/!wxUSE_WXVSNPRINTF } else if ( len >= size ) { #if wxUSE_WXVSNPRINTF - // we know that our own implementation of wxVsnprintf() returns + // we know that our own implementation of wxVsnprintf() returns // size+1 when there's not enough space but that's not the size // of the required buffer! size *= 2; // so we just double the current size of the buffer #else // some vsnprintf() implementations NUL-terminate the buffer and // some don't in len == size case, to be safe always add 1 + // FIXME: I don't quite understand this comment. The vsnprintf + // function is specifically defined to return the number of + // characters printed not including the null terminator. + // So OF COURSE you need to add 1 to get the right buffer size. + // The following line is definitely correct, no question. size = len + 1; #endif } @@ -1644,7 +1908,7 @@ int wxString::PrintfV(const wxString& format, va_list argptr) #if wxUSE_STL_BASED_WXSTRING typedef wxStringTypeBuffer Utf8Buffer; #else - typedef wxImplStringBuffer Utf8Buffer; + typedef wxStringInternalBuffer Utf8Buffer; #endif #endif @@ -1826,9 +2090,47 @@ int wxString::Freq(wxUniChar ch) const return count; } -// convert to upper case, return the copy of the string -wxString wxString::Upper() const -{ wxString s(*this); return s.MakeUpper(); } +// ---------------------------------------------------------------------------- +// wxUTF8StringBuffer +// ---------------------------------------------------------------------------- -// convert to lower case, return the copy of the string -wxString wxString::Lower() const { wxString s(*this); return s.MakeLower(); } +#if wxUSE_UNICODE_WCHAR +wxUTF8StringBuffer::~wxUTF8StringBuffer() +{ + wxMBConvStrictUTF8 conv; + size_t wlen = conv.ToWChar(NULL, 0, m_buf); + wxCHECK_RET( wlen != wxCONV_FAILED, "invalid UTF-8 data in string buffer?" ); + + wxStringInternalBuffer wbuf(m_str, wlen); + conv.ToWChar(wbuf, wlen, m_buf); +} + +wxUTF8StringBufferLength::~wxUTF8StringBufferLength() +{ + wxCHECK_RET(m_lenSet, "length not set"); + + wxMBConvStrictUTF8 conv; + size_t wlen = conv.ToWChar(NULL, 0, m_buf, m_len); + wxCHECK_RET( wlen != wxCONV_FAILED, "invalid UTF-8 data in string buffer?" ); + + wxStringInternalBufferLength wbuf(m_str, wlen); + conv.ToWChar(wbuf, wlen, m_buf, m_len); + wbuf.SetLength(wlen); +} +#endif // wxUSE_UNICODE_WCHAR + +// ---------------------------------------------------------------------------- +// wxCharBufferType +// ---------------------------------------------------------------------------- + +#ifndef __VMS_BROKEN_TEMPLATES +template<> +#endif +wxCharTypeBuffer::Data +wxCharTypeBuffer::NullData(NULL); + +#ifndef __VMS_BROKEN_TEMPLATES +template<> +#endif +wxCharTypeBuffer::Data +wxCharTypeBuffer::NullData(NULL);