X-Git-Url: https://git.saurik.com/wxWidgets.git/blobdiff_plain/64a044d5a64dd92473b4cc666a6877db78bd37d3..a8be206acc1893075b6e91646cf9a2c4c859291e:/include/wx/string.h

diff --git a/include/wx/string.h b/include/wx/string.h
index b72a312f45..08eafdc4b4 100644
--- a/include/wx/string.h
+++ b/include/wx/string.h
@@ -57,6 +57,36 @@
 #include "wx/stringops.h"
 #include "wx/unichar.h"
 
+// by default we cache the mapping of the positions in UTF-8 string to the byte
+// offset as this results in noticeable performance improvements for loops over
+// strings using indices; comment out this line to disable this
+//
+// notice that this optimization is well worth using even in debug builds as it
+// changes asymptotic complexity of algorithms using indices to iterate over
+// wxString back to expected linear from quadratic
+//
+// also notice that wxTLS_TYPE() (__declspec(thread) in this case) is unsafe to
+// use in DLL build under pre-Vista Windows so we disable this code for now, if
+// anybody really needs to use UTF-8 build under Windows with this optimization
+// it would have to be re-tested and probably corrected
+#if wxUSE_UNICODE_UTF8 && !defined(__WXMSW__)
+    #define wxUSE_STRING_POS_CACHE 1
+#else
+    #define wxUSE_STRING_POS_CACHE 0
+#endif
+
+#if wxUSE_STRING_POS_CACHE
+    #include "wx/tls.h"
+
+    // change this 0 to 1 to enable additional (very expensive) asserts
+    // verifying that string caching logic works as expected
+    #if 0
+        #define wxSTRING_CACHE_ASSERT(cond) wxASSERT(cond)
+    #else
+        #define wxSTRING_CACHE_ASSERT(cond)
+    #endif
+#endif // wxUSE_STRING_POS_CACHE
+
 class WXDLLIMPEXP_FWD_BASE wxString;
 
 // unless this symbol is predefined to disable the compatibility functions, do
@@ -508,10 +538,13 @@ private:
   static size_t LenToImpl(size_t len) { return len; }
   static size_t PosFromImpl(size_t pos) { return pos; }
 
-  // we don't want to define this as an empty inline function as it could
+  // we don't want to define these as empty inline functions as it could
   // result in noticeable (and quite unnecessary in non-UTF-8 build) slowdown
   // in debug build where the inline functions are not effectively inlined
-  #define wxSTRING_INVALIDATE_INDEX_CACHE()
+  #define wxSTRING_INVALIDATE_CACHE()
+  #define wxSTRING_INVALIDATE_CACHED_LENGTH()
+  #define wxSTRING_UPDATE_CACHED_LENGTH(n)
+  #define wxSTRING_SET_CACHED_LENGTH(n)
 
 #else // wxUSE_UNICODE_UTF8
 
@@ -527,62 +560,250 @@ private:
   static SubstrBufFromWC ImplStr(const wchar_t* str, size_t n)
     { return ConvertStr(str, n, wxMBConvUTF8()); }
 
-  // this is an extremely simple cache used by PosToImpl()
-  struct PosToImplCache
+#if wxUSE_STRING_POS_CACHE
+  // this is an extremely simple cache used by PosToImpl(): each cache element
+  // contains the string it applies to and the index corresponding to the last
+  // used position in this wxString in its m_impl string
+  //
+  // NB: notice that this struct (and nested Element one) must be a POD or we
+  //     wouldn't be able to use a thread-local variable of this type, in
+  //     particular it should have no ctor -- we rely on statics being
+  //     initialized to 0 instead
+  struct Cache
   {
-      const wxString *str;
-      size_t pos,
-             impl;
+      enum { SIZE = 8 };
+
+      struct Element
+      {
+          const wxString *str;  // the string to which this element applies
+          size_t pos,           // the cached index in this string
+                 impl,          // the corresponding position in its m_impl
+                 len;           // cached length or npos if unknown
+
+          // reset cached index to 0
+          void ResetPos() { pos = impl = 0; }
+
+          // reset position and length
+          void Reset() { ResetPos(); len = npos; }
+      };
+
+      // cache the indices mapping for the last few string used
+      Element cached[SIZE];
+
+      // the last used index
+      unsigned lastUsed;
   };
 
-  static PosToImplCache ms_cache;
+  // notice that we must use an accessor function and not a static variable
+  // because when the TLS variables support is implemented in the library (and
+  // not by the compiler), the global s_cache variable could be not yet
+  // initialized when a ctor of another global object is executed and if that
+  // ctor uses any wxString methods, bad things happen
+  //
+  // also note that for the same reason this function _is_ MT-safe: we know
+  // it's going to be called during the program startup (currently during
+  // globals initialization but even if they ever stop using wxString, it would
+  // still be called by wxInitialize()), i.e. before any threads are created
+  static Cache& GetCache()
+  {
+      static wxTLS_TYPE(Cache) s_cache;
+
+      return wxTLS_VALUE(s_cache);
+  }
+
+  static Cache::Element *GetCacheBegin() { return GetCache().cached; }
+  static Cache::Element *GetCacheEnd() { return GetCacheBegin() + Cache::SIZE; }
+  static unsigned& LastUsedCacheElement() { return GetCache().lastUsed; }
+
+  friend struct wxStrCacheDumper;
+
+  // uncomment this to have access to some profiling statistics on program
+  // termination
+  //#define wxPROFILE_STRING_CACHE
+
+#ifdef wxPROFILE_STRING_CACHE
+  static struct PosToImplCacheStats
+  {
+      unsigned postot,  // total non-trivial calls to PosToImpl
+               poshits, // cache hits from PosToImpl()
+               mishits, // cached position beyond the needed one
+               sumpos,  // sum of all positions, used to compute the
+                        // average position after dividing by postot
+               sumofs,  // sum of all offsets after using the cache, used to
+                        // compute the average after dividing by hits
+               lentot,  // number of total calls to length()
+               lenhits; // number of cache hits in length()
+  } ms_cacheStats;
+
+  friend struct ShowCacheStats;
+
+  #define wxCACHE_PROFILE_FIELD_INC(field) ms_cacheStats.field++
+  #define wxCACHE_PROFILE_FIELD_ADD(field, val) ms_cacheStats.field += (val)
+#else // !wxPROFILE_STRING_CACHE
+  #define wxCACHE_PROFILE_FIELD_INC(field)
+  #define wxCACHE_PROFILE_FIELD_ADD(field, val)
+#endif // wxPROFILE_STRING_CACHE/!wxPROFILE_STRING_CACHE
+
+  // note: it could seem that the functions below shouldn't be inline because
+  // they are big, contain loops and so the compiler shouldn't be able to
+  // inline them anyhow, however moving them into string.cpp does decrease the
+  // code performance by ~5%, at least when using g++ 4.1 so do keep them here
+  // unless tests show that it's not advantageous any more
+
+  // return the pointer to the cache element for this string or NULL if not
+  // cached
+  Cache::Element *FindCacheElement() const
+  {
+      // profiling seems to show a small but consistent gain if we use this
+      // simple loop instead of starting from the last used element (there are
+      // a lot of misses in this function...)
+      for ( Cache::Element *c = GetCacheBegin(); c != GetCacheEnd(); c++ )
+      {
+          if ( c->str == this )
+              return c;
+      }
 
-  size_t PosToImpl(size_t pos) const
+      return NULL;
+  }
+
+  // unlike FindCacheElement(), this one always returns a valid pointer to the
+  // cache element for this string, it may have valid last cached position and
+  // its corresponding index in the byte string or not
+  Cache::Element *GetCacheElement() const
   {
-      if ( pos == 0 || pos == npos )
-          return pos;
+      Cache::Element * const cacheBegin = GetCacheBegin();
+      Cache::Element * const cacheEnd = GetCacheEnd();
+      Cache::Element * const cacheStart = cacheBegin + LastUsedCacheElement();
+
+      // check the last used first, this does no (measurable) harm for a miss
+      // but does help for simple loops addressing the same string all the time
+      if ( cacheStart->str == this )
+          return cacheStart;
 
-      PosToImplCache& cache = ms_cache;
-      if ( this == cache.str )
+      // notice that we're going to check cacheStart again inside this call but
+      // profiling shows that it's still faster to use a simple loop like
+      // inside FindCacheElement() than manually looping with wrapping starting
+      // from the cache entry after the start one
+      Cache::Element *c = FindCacheElement();
+      if ( !c )
       {
-          if ( pos == cache.pos )
-              return cache.impl;
+          // claim the next cache entry for this string
+          c = cacheStart;
+          if ( ++c == cacheEnd )
+              c = cacheBegin;
 
-          // TODO: is it worth complicating this function even further by going
-          //       backwards from the last position? it might be if we're just
-          //       before it...
-          if ( cache.pos > pos )
-          {
-              cache.pos =
-              cache.impl = 0;
-          }
+          c->str = this;
+          c->Reset();
+
+          // and remember the last used element
+          LastUsedCacheElement() = c - cacheBegin;
       }
-      else // data for this string not cached
+
+      return c;
+  }
+
+  size_t DoPosToImpl(size_t pos) const
+  {
+      wxCACHE_PROFILE_FIELD_INC(postot);
+
+      // NB: although the case of pos == 1 (and offset from cached position
+      //     equal to 1) are common, nothing is gained by writing special code
+      //     for handling them, the compiler (at least g++ 4.1 used) seems to
+      //     optimize the code well enough on its own
+
+      wxCACHE_PROFILE_FIELD_ADD(sumpos, pos);
+
+      Cache::Element * const cache = GetCacheElement();
+
+      // cached position can't be 0 so if it is, it means that this entry was
+      // used for length caching only so far, i.e. it doesn't count as a hit
+      // from our point of view
+      if ( cache->pos )
+          wxCACHE_PROFILE_FIELD_INC(poshits);
+
+      if ( pos == cache->pos )
+          return cache->impl;
+
+      // this seems to happen only rarely so just reset the cache in this case
+      // instead of complicating code even further by seeking backwards in this
+      // case
+      if ( cache->pos > pos )
       {
-          cache.str = this;
-          cache.pos =
-          cache.impl = 0;
+          wxCACHE_PROFILE_FIELD_INC(mishits);
+
+          cache->ResetPos();
       }
 
-      wxStringImpl::const_iterator i(m_impl.begin() + cache.impl);
-      for ( size_t n = cache.pos; n < pos; n++ )
+      wxCACHE_PROFILE_FIELD_ADD(sumofs, pos - cache->pos);
+
+
+      wxStringImpl::const_iterator i(m_impl.begin() + cache->impl);
+      for ( size_t n = cache->pos; n < pos; n++ )
           wxStringOperations::IncIter(i);
 
-      cache.pos = pos;
-      return cache.impl = i - m_impl.begin();
+      cache->pos = pos;
+      cache->impl = i - m_impl.begin();
+
+      wxSTRING_CACHE_ASSERT(
+          (int)cache->impl == (begin() + pos).impl() - m_impl.begin() );
+
+      return cache->impl;
+  }
+
+  void InvalidateCache()
+  {
+      Cache::Element * const cache = FindCacheElement();
+      if ( cache )
+          cache->Reset();
+  }
+
+  void InvalidateCachedLength()
+  {
+      Cache::Element * const cache = FindCacheElement();
+      if ( cache )
+          cache->len = npos;
+  }
+
+  void SetCachedLength(size_t len)
+  {
+      // we optimistically cache the length here even if the string wasn't
+      // present in the cache before, this seems to do no harm and the
+      // potential for avoiding length recomputation for long strings looks
+      // interesting
+      GetCacheElement()->len = len;
   }
 
-  void InvalidatePosToImplCache()
+  void UpdateCachedLength(ptrdiff_t delta)
   {
-      PosToImplCache& cache = ms_cache;
-      if ( cache.str == this )
+      Cache::Element * const cache = FindCacheElement();
+      if ( cache && cache->len != npos )
       {
-          cache.pos =
-              cache.impl = 0;
+          wxSTRING_CACHE_ASSERT( (ptrdiff_t)cache->len + delta >= 0 );
+
+          cache->len += delta;
       }
   }
 
-  #define wxSTRING_INVALIDATE_INDEX_CACHE() InvalidatePosToImplCache()
+  #define wxSTRING_INVALIDATE_CACHE() InvalidateCache()
+  #define wxSTRING_INVALIDATE_CACHED_LENGTH() InvalidateCachedLength()
+  #define wxSTRING_UPDATE_CACHED_LENGTH(n) UpdateCachedLength(n)
+  #define wxSTRING_SET_CACHED_LENGTH(n) SetCachedLength(n)
+#else // !wxUSE_STRING_POS_CACHE
+  size_t DoPosToImpl(size_t pos) const
+  {
+      return (begin() + pos).impl() - m_impl.begin();
+  }
+
+  #define wxSTRING_INVALIDATE_CACHE()
+  #define wxSTRING_INVALIDATE_CACHED_LENGTH()
+  #define wxSTRING_UPDATE_CACHED_LENGTH(n)
+  #define wxSTRING_SET_CACHED_LENGTH(n)
+#endif // wxUSE_STRING_POS_CACHE/!wxUSE_STRING_POS_CACHE
+
+  size_t PosToImpl(size_t pos) const
+  {
+      return pos == 0 || pos == npos ? pos : DoPosToImpl(pos);
+  }
 
   void PosLenToImpl(size_t pos, size_t len, size_t *implPos, size_t *implLen) const;
 
@@ -727,7 +948,7 @@ public:
       }
 
       reference operator*()
-        { return wxUniCharRef::CreateForString(m_node, m_cur); }
+        { return wxUniCharRef::CreateForString(*str(), m_cur); }
 
       iterator operator+(ptrdiff_t n) const
         { return iterator(str(), wxStringOperations::AddToIter(m_cur, n)); }
@@ -791,9 +1012,9 @@ public:
   size_t IterToImplPos(wxString::iterator i) const
     { return wxStringImpl::const_iterator(i.impl()) - m_impl.begin(); }
 
-  iterator GetNthIter(size_t n)
+  iterator GetIterForNthChar(size_t n)
     { return iterator(this, m_impl.begin() + PosToImpl(n)); }
-  const_iterator GetNthIter(size_t n) const
+  const_iterator GetIterForNthChar(size_t n) const
     { return const_iterator(this, m_impl.begin() + PosToImpl(n)); }
 #else // !wxUSE_UNICODE_UTF8
 
@@ -847,8 +1068,8 @@ public:
           : m_cur(ptr) {}
   };
 
-  iterator GetNthIter(size_t n) { return begin() + n; }
-  const_iterator GetNthIter(size_t n) const { return begin() + n; }
+  iterator GetIterForNthChar(size_t n) { return begin() + n; }
+  const_iterator GetIterForNthChar(size_t n) const { return begin() + n; }
 #endif // wxUSE_UNICODE_UTF8/!wxUSE_UNICODE_UTF8
 
   #undef WX_STR_ITERATOR_TAG
@@ -1035,6 +1256,17 @@ public:
   wxString(const wxString& str, size_t nLength)
     { assign(str, nLength); }
 
+
+#if wxUSE_STRING_POS_CACHE
+  ~wxString()
+  {
+      // we need to invalidate our cache entry as another string could be
+      // recreated at the same address (unlikely, but still possible, with the
+      // heap-allocated strings but perfectly common with stack-allocated ones)
+      InvalidateCache();
+  }
+#endif // wxUSE_STRING_POS_CACHE
+
   // even if we're not built with wxUSE_STL == 1 it is very convenient to allow
   // implicit conversions from std::string to wxString and vice verse as this
   // allows to use the same strings in non-GUI and GUI code, however we don't
@@ -1110,7 +1342,32 @@ public:
 
   // std::string methods:
 #if wxUSE_UNICODE_UTF8
-  size_t length() const { return end() - begin(); } // FIXME-UTF8: optimize!
+  size_t length() const
+  {
+#if wxUSE_STRING_POS_CACHE
+      wxCACHE_PROFILE_FIELD_INC(lentot);
+
+      Cache::Element * const cache = GetCacheElement();
+
+      if ( cache->len == npos )
+      {
+          // it's probably not worth trying to be clever and using cache->pos
+          // here as it's probably 0 anyhow -- you usually call length() before
+          // starting to index the string
+          cache->len = end() - begin();
+      }
+      else
+      {
+          wxCACHE_PROFILE_FIELD_INC(lenhits);
+
+          wxSTRING_CACHE_ASSERT( (int)cache->len == end() - begin() );
+      }
+
+      return cache->len;
+#else // !wxUSE_STRING_POS_CACHE
+      return end() - begin();
+#endif // wxUSE_STRING_POS_CACHE/!wxUSE_STRING_POS_CACHE
+  }
 #else
   size_t length() const { return m_impl.length(); }
 #endif
@@ -1120,8 +1377,9 @@ public:
 
   bool empty() const { return m_impl.empty(); }
 
-  size_type capacity() const { return m_impl.capacity(); } // FIXME-UTF8
-  void reserve(size_t sz) { m_impl.reserve(sz); } // FIXME-UTF8
+  // NB: these methods don't have a well-defined meaning in UTF-8 case
+  size_type capacity() const { return m_impl.capacity(); }
+  void reserve(size_t sz) { m_impl.reserve(sz); }
 
   void resize(size_t nSize, wxUniChar ch = wxT('\0'))
   {
@@ -1132,7 +1390,7 @@ public:
 #if wxUSE_UNICODE_UTF8
     if ( nSize < len )
     {
-        wxSTRING_INVALIDATE_INDEX_CACHE();
+        wxSTRING_INVALIDATE_CACHE();
 
         // we can't use wxStringImpl::resize() for truncating the string as it
         // counts in bytes, not characters
@@ -1143,10 +1401,16 @@ public:
     // we also can't use (presumably more efficient) resize() if we have to
     // append characters taking more than one byte
     if ( !ch.IsAscii() )
+    {
         append(nSize - len, ch);
-    else
+    }
+    else // can use (presumably faster) resize() version
 #endif // wxUSE_UNICODE_UTF8
+    {
+        wxSTRING_INVALIDATE_CACHED_LENGTH();
+
         m_impl.resize(nSize, (wxStringCharType)ch);
+    }
   }
 
   wxString substr(size_t nStart = 0, size_t nLen = npos) const
@@ -1191,27 +1455,12 @@ public:
       { return at(n); }
     // read/write access
     wxUniCharRef at(size_t n)
-      { return *GetNthIter(n); }
+      { return *GetIterForNthChar(n); }
     wxUniCharRef GetWritableChar(size_t n)
       { return at(n); }
     // write access
     void SetChar(size_t n, wxUniChar ch)
-    {
-        wxUniCharRef ref(at(n));
-
-#if wxUSE_UNICODE_UTF8
-        // if the new character takes the same number of bytes as the old one,
-        // we can avoid invalidating the index cache and it's an important case
-        // as it is exercised by a simple for loop using indices
-        if ( wxStringOperations::GetUtf8CharLength(ref)
-                != wxStringOperations::GetUtf8CharLength(ch) )
-        {
-            wxSTRING_INVALIDATE_INDEX_CACHE();
-        }
-#endif // wxUSE_UNICODE_UTF8
-
-        ref = ch;
-    }
+      { at(n) = ch; }
 
     // get last character
     wxUniChar Last() const
@@ -1503,7 +1752,7 @@ public:
   {
     if ( this != &stringSrc )
     {
-        wxSTRING_INVALIDATE_INDEX_CACHE();
+        wxSTRING_INVALIDATE_CACHE();
 
         m_impl = stringSrc.m_impl;
     }
@@ -1516,7 +1765,7 @@ public:
     // from a character
   wxString& operator=(wxUniChar ch)
   {
-    wxSTRING_INVALIDATE_INDEX_CACHE();
+    wxSTRING_INVALIDATE_CACHE();
 
 #if wxUSE_UNICODE_UTF8
     if ( !ch.IsAscii() )
@@ -1540,7 +1789,7 @@ public:
 #if wxUSE_STL_BASED_WXSTRING
   wxString& operator=(const char *psz)
   {
-      wxSTRING_INVALIDATE_INDEX_CACHE();
+      wxSTRING_INVALIDATE_CACHE();
 
       if ( psz )
           m_impl = ImplStr(psz);
@@ -1552,7 +1801,7 @@ public:
 
   wxString& operator=(const wchar_t *pwz)
   {
-      wxSTRING_INVALIDATE_INDEX_CACHE();
+      wxSTRING_INVALIDATE_CACHE();
 
       if ( pwz )
           m_impl = ImplStr(pwz);
@@ -1564,7 +1813,7 @@ public:
 #else // !wxUSE_STL_BASED_WXSTRING
   wxString& operator=(const char *psz)
   {
-      wxSTRING_INVALIDATE_INDEX_CACHE();
+      wxSTRING_INVALIDATE_CACHE();
 
       m_impl = ImplStr(psz);
 
@@ -1573,7 +1822,7 @@ public:
 
   wxString& operator=(const wchar_t *pwz)
   {
-      wxSTRING_INVALIDATE_INDEX_CACHE();
+      wxSTRING_INVALIDATE_CACHE();
 
       m_impl = ImplStr(pwz);
 
@@ -2067,30 +2316,54 @@ public:
     // append elements str[pos], ..., str[pos+n]
   wxString& append(const wxString& str, size_t pos, size_t n)
   {
-    size_t from, len;
-    str.PosLenToImpl(pos, n, &from, &len);
-    m_impl.append(str.m_impl, from, len);
-    return *this;
+      wxSTRING_UPDATE_CACHED_LENGTH(n);
+
+      size_t from, len;
+      str.PosLenToImpl(pos, n, &from, &len);
+      m_impl.append(str.m_impl, from, len);
+      return *this;
   }
     // append a string
   wxString& append(const wxString& str)
-    { m_impl.append(str.m_impl); return *this; }
+  {
+      wxSTRING_UPDATE_CACHED_LENGTH(str.length());
+
+      m_impl.append(str.m_impl);
+      return *this;
+  }
+
     // append first n (or all if n == npos) characters of sz
   wxString& append(const char *sz)
-    { m_impl.append(ImplStr(sz)); return *this; }
+  {
+      wxSTRING_INVALIDATE_CACHED_LENGTH();
+
+      m_impl.append(ImplStr(sz));
+      return *this;
+  }
+
   wxString& append(const wchar_t *sz)
-    { m_impl.append(ImplStr(sz)); return *this; }
+  {
+      wxSTRING_INVALIDATE_CACHED_LENGTH();
+
+      m_impl.append(ImplStr(sz));
+      return *this;
+  }
+
   wxString& append(const char *sz, size_t n)
   {
-    SubstrBufFromMB str(ImplStr(sz, n));
-    m_impl.append(str.data, str.len);
-    return *this;
+      wxSTRING_INVALIDATE_CACHED_LENGTH();
+
+      SubstrBufFromMB str(ImplStr(sz, n));
+      m_impl.append(str.data, str.len);
+      return *this;
   }
   wxString& append(const wchar_t *sz, size_t n)
   {
-    SubstrBufFromWC str(ImplStr(sz, n));
-    m_impl.append(str.data, str.len);
-    return *this;
+      wxSTRING_UPDATE_CACHED_LENGTH(n);
+
+      SubstrBufFromWC str(ImplStr(sz, n));
+      m_impl.append(str.data, str.len);
+      return *this;
   }
 
   wxString& append(const wxCStrData& str)
@@ -2110,13 +2383,23 @@ public:
   wxString& append(size_t n, wxUniChar ch)
   {
 #if wxUSE_UNICODE_UTF8
-    if ( !ch.IsAscii() )
-        m_impl.append(wxStringOperations::EncodeNChars(n, ch));
-    else
+      if ( !ch.IsAscii() )
+      {
+          wxSTRING_INVALIDATE_CACHED_LENGTH();
+
+          m_impl.append(wxStringOperations::EncodeNChars(n, ch));
+      }
+      else // ASCII
 #endif
-        m_impl.append(n, (wxStringCharType)ch);
-    return *this;
+      {
+          wxSTRING_UPDATE_CACHED_LENGTH(n);
+
+          m_impl.append(n, (wxStringCharType)ch);
+      }
+
+      return *this;
   }
+
   wxString& append(size_t n, wxUniCharRef ch)
     { return append(n, wxUniChar(ch)); }
   wxString& append(size_t n, char ch)
@@ -2128,7 +2411,12 @@ public:
 
     // append from first to last
   wxString& append(const_iterator first, const_iterator last)
-    { m_impl.append(first.impl(), last.impl()); return *this; }
+  {
+      wxSTRING_INVALIDATE_CACHED_LENGTH();
+
+      m_impl.append(first.impl(), last.impl());
+      return *this;
+  }
 #if WXWIN_COMPATIBILITY_STRING_PTR_AS_ITER
   wxString& append(const char *first, const char *last)
     { return append(first, last - first); }
@@ -2141,7 +2429,7 @@ public:
     // same as `this_string = str'
   wxString& assign(const wxString& str)
   {
-      wxSTRING_INVALIDATE_INDEX_CACHE();
+      wxSTRING_SET_CACHED_LENGTH(str.length());
 
       m_impl = str.m_impl;
 
@@ -2150,7 +2438,7 @@ public:
 
   wxString& assign(const wxString& str, size_t len)
   {
-      wxSTRING_INVALIDATE_INDEX_CACHE();
+      wxSTRING_SET_CACHED_LENGTH(len);
 
       m_impl.assign(str.m_impl, 0, str.LenToImpl(len));
 
@@ -2160,18 +2448,21 @@ public:
     // same as ` = str[pos..pos + n]
   wxString& assign(const wxString& str, size_t pos, size_t n)
   {
-      wxSTRING_INVALIDATE_INDEX_CACHE();
-
       size_t from, len;
       str.PosLenToImpl(pos, n, &from, &len);
       m_impl.assign(str.m_impl, from, len);
+
+      // it's important to call this after PosLenToImpl() above in case str is
+      // the same string as this one
+      wxSTRING_SET_CACHED_LENGTH(n);
+
       return *this;
   }
 
     // same as `= first n (or all if n == npos) characters of sz'
   wxString& assign(const char *sz)
   {
-      wxSTRING_INVALIDATE_INDEX_CACHE();
+      wxSTRING_INVALIDATE_CACHE();
 
       m_impl.assign(ImplStr(sz));
 
@@ -2180,7 +2471,7 @@ public:
 
   wxString& assign(const wchar_t *sz)
   {
-      wxSTRING_INVALIDATE_INDEX_CACHE();
+      wxSTRING_INVALIDATE_CACHE();
 
       m_impl.assign(ImplStr(sz));
 
@@ -2189,7 +2480,7 @@ public:
 
   wxString& assign(const char *sz, size_t n)
   {
-      wxSTRING_INVALIDATE_INDEX_CACHE();
+      wxSTRING_SET_CACHED_LENGTH(n);
 
       SubstrBufFromMB str(ImplStr(sz, n));
       m_impl.assign(str.data, str.len);
@@ -2199,7 +2490,7 @@ public:
 
   wxString& assign(const wchar_t *sz, size_t n)
   {
-      wxSTRING_INVALIDATE_INDEX_CACHE();
+      wxSTRING_SET_CACHED_LENGTH(n);
 
       SubstrBufFromWC str(ImplStr(sz, n));
       m_impl.assign(str.data, str.len);
@@ -2223,15 +2514,16 @@ public:
     // same as `= n copies of ch'
   wxString& assign(size_t n, wxUniChar ch)
   {
-    wxSTRING_INVALIDATE_INDEX_CACHE();
+      wxSTRING_SET_CACHED_LENGTH(n);
 
 #if wxUSE_UNICODE_UTF8
-    if ( !ch.IsAscii() )
-        m_impl.assign(wxStringOperations::EncodeNChars(n, ch));
-    else
+      if ( !ch.IsAscii() )
+          m_impl.assign(wxStringOperations::EncodeNChars(n, ch));
+      else
 #endif
-        m_impl.assign(n, (wxStringCharType)ch);
-    return *this;
+          m_impl.assign(n, (wxStringCharType)ch);
+
+      return *this;
   }
 
   wxString& assign(size_t n, wxUniCharRef ch)
@@ -2246,7 +2538,7 @@ public:
     // assign from first to last
   wxString& assign(const_iterator first, const_iterator last)
   {
-      wxSTRING_INVALIDATE_INDEX_CACHE();
+      wxSTRING_INVALIDATE_CACHE();
 
       m_impl.assign(first.impl(), last.impl());
 
@@ -2284,11 +2576,11 @@ public:
 
     // insert another string
   wxString& insert(size_t nPos, const wxString& str)
-    { insert(GetNthIter(nPos), str.begin(), str.end()); return *this; }
+    { insert(GetIterForNthChar(nPos), str.begin(), str.end()); return *this; }
     // insert n chars of str starting at nStart (in str)
   wxString& insert(size_t nPos, const wxString& str, size_t nStart, size_t n)
   {
-      wxSTRING_INVALIDATE_INDEX_CACHE();
+      wxSTRING_UPDATE_CACHED_LENGTH(n);
 
       size_t from, len;
       str.PosLenToImpl(nStart, n, &from, &len);
@@ -2300,7 +2592,7 @@ public:
     // insert first n (or all if n == npos) characters of sz
   wxString& insert(size_t nPos, const char *sz)
   {
-      wxSTRING_INVALIDATE_INDEX_CACHE();
+      wxSTRING_INVALIDATE_CACHE();
 
       m_impl.insert(PosToImpl(nPos), ImplStr(sz));
 
@@ -2309,14 +2601,14 @@ public:
 
   wxString& insert(size_t nPos, const wchar_t *sz)
   {
-      wxSTRING_INVALIDATE_INDEX_CACHE();
+      wxSTRING_INVALIDATE_CACHE();
 
       m_impl.insert(PosToImpl(nPos), ImplStr(sz)); return *this;
   }
 
   wxString& insert(size_t nPos, const char *sz, size_t n)
   {
-      wxSTRING_INVALIDATE_INDEX_CACHE();
+      wxSTRING_UPDATE_CACHED_LENGTH(n);
 
       SubstrBufFromMB str(ImplStr(sz, n));
       m_impl.insert(PosToImpl(nPos), str.data, str.len);
@@ -2326,7 +2618,7 @@ public:
 
   wxString& insert(size_t nPos, const wchar_t *sz, size_t n)
   {
-      wxSTRING_INVALIDATE_INDEX_CACHE();
+      wxSTRING_UPDATE_CACHED_LENGTH(n);
 
       SubstrBufFromWC str(ImplStr(sz, n));
       m_impl.insert(PosToImpl(nPos), str.data, str.len);
@@ -2337,36 +2629,36 @@ public:
     // insert n copies of ch
   wxString& insert(size_t nPos, size_t n, wxUniChar ch)
   {
-    wxSTRING_INVALIDATE_INDEX_CACHE();
+      wxSTRING_UPDATE_CACHED_LENGTH(n);
 
 #if wxUSE_UNICODE_UTF8
-    if ( !ch.IsAscii() )
-        m_impl.insert(PosToImpl(nPos), wxStringOperations::EncodeNChars(n, ch));
-    else
+      if ( !ch.IsAscii() )
+          m_impl.insert(PosToImpl(nPos), wxStringOperations::EncodeNChars(n, ch));
+      else
 #endif
-        m_impl.insert(PosToImpl(nPos), n, (wxStringCharType)ch);
-    return *this;
+          m_impl.insert(PosToImpl(nPos), n, (wxStringCharType)ch);
+      return *this;
   }
 
   iterator insert(iterator it, wxUniChar ch)
   {
-    wxSTRING_INVALIDATE_INDEX_CACHE();
+      wxSTRING_UPDATE_CACHED_LENGTH(1);
 
 #if wxUSE_UNICODE_UTF8
-    if ( !ch.IsAscii() )
-    {
-        size_t pos = IterToImplPos(it);
-        m_impl.insert(pos, wxStringOperations::EncodeChar(ch));
-        return iterator(this, m_impl.begin() + pos);
-    }
-    else
+      if ( !ch.IsAscii() )
+      {
+          size_t pos = IterToImplPos(it);
+          m_impl.insert(pos, wxStringOperations::EncodeChar(ch));
+          return iterator(this, m_impl.begin() + pos);
+      }
+      else
 #endif
-        return iterator(this, m_impl.insert(it.impl(), (wxStringCharType)ch));
+          return iterator(this, m_impl.insert(it.impl(), (wxStringCharType)ch));
   }
 
   void insert(iterator it, const_iterator first, const_iterator last)
   {
-      wxSTRING_INVALIDATE_INDEX_CACHE();
+      wxSTRING_INVALIDATE_CACHE();
 
       m_impl.insert(it.impl(), first.impl(), last.impl());
   }
@@ -2382,20 +2674,20 @@ public:
 
   void insert(iterator it, size_type n, wxUniChar ch)
   {
-    wxSTRING_INVALIDATE_INDEX_CACHE();
+      wxSTRING_UPDATE_CACHED_LENGTH(n);
 
 #if wxUSE_UNICODE_UTF8
-    if ( !ch.IsAscii() )
-        m_impl.insert(IterToImplPos(it), wxStringOperations::EncodeNChars(n, ch));
-    else
+      if ( !ch.IsAscii() )
+          m_impl.insert(IterToImplPos(it), wxStringOperations::EncodeNChars(n, ch));
+      else
 #endif
-        m_impl.insert(it.impl(), n, (wxStringCharType)ch);
+          m_impl.insert(it.impl(), n, (wxStringCharType)ch);
   }
 
     // delete characters from nStart to nStart + nLen
   wxString& erase(size_type pos = 0, size_type n = npos)
   {
-      wxSTRING_INVALIDATE_INDEX_CACHE();
+      wxSTRING_INVALIDATE_CACHE();
 
       size_t from, len;
       PosLenToImpl(pos, n, &from, &len);
@@ -2407,14 +2699,14 @@ public:
     // delete characters from first up to last
   iterator erase(iterator first, iterator last)
   {
-      wxSTRING_INVALIDATE_INDEX_CACHE();
+      wxSTRING_INVALIDATE_CACHE();
 
       return iterator(this, m_impl.erase(first.impl(), last.impl()));
   }
 
   iterator erase(iterator first)
   {
-      wxSTRING_INVALIDATE_INDEX_CACHE();
+      wxSTRING_UPDATE_CACHED_LENGTH(-1);
 
       return iterator(this, m_impl.erase(first.impl()));
   }
@@ -2424,7 +2716,7 @@ public:
 #else
   void clear()
   {
-      wxSTRING_INVALIDATE_INDEX_CACHE();
+      wxSTRING_SET_CACHED_LENGTH(0);
 
       m_impl.clear();
   }
@@ -2433,7 +2725,7 @@ public:
     // replaces the substring of length nLen starting at nStart
   wxString& replace(size_t nStart, size_t nLen, const char* sz)
   {
-      wxSTRING_INVALIDATE_INDEX_CACHE();
+      wxSTRING_INVALIDATE_CACHE();
 
       size_t from, len;
       PosLenToImpl(nStart, nLen, &from, &len);
@@ -2444,7 +2736,7 @@ public:
 
   wxString& replace(size_t nStart, size_t nLen, const wchar_t* sz)
   {
-      wxSTRING_INVALIDATE_INDEX_CACHE();
+      wxSTRING_INVALIDATE_CACHE();
 
       size_t from, len;
       PosLenToImpl(nStart, nLen, &from, &len);
@@ -2456,7 +2748,7 @@ public:
     // replaces the substring of length nLen starting at nStart
   wxString& replace(size_t nStart, size_t nLen, const wxString& str)
   {
-      wxSTRING_INVALIDATE_INDEX_CACHE();
+      wxSTRING_INVALIDATE_CACHE();
 
       size_t from, len;
       PosLenToImpl(nStart, nLen, &from, &len);
@@ -2468,7 +2760,7 @@ public:
     // replaces the substring with nCount copies of ch
   wxString& replace(size_t nStart, size_t nLen, size_t nCount, wxUniChar ch)
   {
-      wxSTRING_INVALIDATE_INDEX_CACHE();
+      wxSTRING_INVALIDATE_CACHE();
 
       size_t from, len;
       PosLenToImpl(nStart, nLen, &from, &len);
@@ -2486,7 +2778,7 @@ public:
   wxString& replace(size_t nStart, size_t nLen,
                     const wxString& str, size_t nStart2, size_t nLen2)
   {
-      wxSTRING_INVALIDATE_INDEX_CACHE();
+      wxSTRING_INVALIDATE_CACHE();
 
       size_t from, len;
       PosLenToImpl(nStart, nLen, &from, &len);
@@ -2503,7 +2795,7 @@ public:
   wxString& replace(size_t nStart, size_t nLen,
                     const char* sz, size_t nCount)
   {
-      wxSTRING_INVALIDATE_INDEX_CACHE();
+      wxSTRING_INVALIDATE_CACHE();
 
       size_t from, len;
       PosLenToImpl(nStart, nLen, &from, &len);
@@ -2518,7 +2810,7 @@ public:
   wxString& replace(size_t nStart, size_t nLen,
                     const wchar_t* sz, size_t nCount)
   {
-      wxSTRING_INVALIDATE_INDEX_CACHE();
+      wxSTRING_INVALIDATE_CACHE();
 
       size_t from, len;
       PosLenToImpl(nStart, nLen, &from, &len);
@@ -2533,7 +2825,7 @@ public:
   wxString& replace(size_t nStart, size_t nLen,
                     const wxString& s, size_t nCount)
   {
-      wxSTRING_INVALIDATE_INDEX_CACHE();
+      wxSTRING_INVALIDATE_CACHE();
 
       size_t from, len;
       PosLenToImpl(nStart, nLen, &from, &len);
@@ -2544,7 +2836,7 @@ public:
 
   wxString& replace(iterator first, iterator last, const char* s)
   {
-      wxSTRING_INVALIDATE_INDEX_CACHE();
+      wxSTRING_INVALIDATE_CACHE();
 
       m_impl.replace(first.impl(), last.impl(), ImplStr(s));
 
@@ -2553,7 +2845,7 @@ public:
 
   wxString& replace(iterator first, iterator last, const wchar_t* s)
   {
-      wxSTRING_INVALIDATE_INDEX_CACHE();
+      wxSTRING_INVALIDATE_CACHE();
 
       m_impl.replace(first.impl(), last.impl(), ImplStr(s));
 
@@ -2562,7 +2854,7 @@ public:
 
   wxString& replace(iterator first, iterator last, const char* s, size_type n)
   {
-      wxSTRING_INVALIDATE_INDEX_CACHE();
+      wxSTRING_INVALIDATE_CACHE();
 
       SubstrBufFromMB str(ImplStr(s, n));
       m_impl.replace(first.impl(), last.impl(), str.data, str.len);
@@ -2572,7 +2864,7 @@ public:
 
   wxString& replace(iterator first, iterator last, const wchar_t* s, size_type n)
   {
-      wxSTRING_INVALIDATE_INDEX_CACHE();
+      wxSTRING_INVALIDATE_CACHE();
 
       SubstrBufFromWC str(ImplStr(s, n));
       m_impl.replace(first.impl(), last.impl(), str.data, str.len);
@@ -2582,7 +2874,7 @@ public:
 
   wxString& replace(iterator first, iterator last, const wxString& s)
   {
-      wxSTRING_INVALIDATE_INDEX_CACHE();
+      wxSTRING_INVALIDATE_CACHE();
 
       m_impl.replace(first.impl(), last.impl(), s.m_impl);
 
@@ -2591,7 +2883,7 @@ public:
 
   wxString& replace(iterator first, iterator last, size_type n, wxUniChar ch)
   {
-      wxSTRING_INVALIDATE_INDEX_CACHE();
+      wxSTRING_INVALIDATE_CACHE();
 
 #if wxUSE_UNICODE_UTF8
       if ( !ch.IsAscii() )
@@ -2607,7 +2899,7 @@ public:
   wxString& replace(iterator first, iterator last,
                     const_iterator first1, const_iterator last1)
   {
-      wxSTRING_INVALIDATE_INDEX_CACHE();
+      wxSTRING_INVALIDATE_CACHE();
 
       m_impl.replace(first.impl(), last.impl(), first1.impl(), last1.impl());
 
@@ -2624,7 +2916,13 @@ public:
   // swap two strings
   void swap(wxString& str)
   {
-      wxSTRING_INVALIDATE_INDEX_CACHE();
+#if wxUSE_STRING_POS_CACHE
+      // we modify not only this string but also the other one directly so we
+      // need to invalidate cache for both of them (we could also try to
+      // exchange their cache entries but it seems unlikely to be worth it)
+      InvalidateCache();
+      str.InvalidateCache();
+#endif // wxUSE_STRING_POS_CACHE
 
       m_impl.swap(str.m_impl);
   }
@@ -2930,14 +3228,34 @@ public:
 
       // string += string
   wxString& operator+=(const wxString& s)
-    { m_impl += s.m_impl; return *this; }
+  {
+      wxSTRING_INVALIDATE_CACHED_LENGTH();
+
+      m_impl += s.m_impl;
+      return *this;
+  }
       // string += C string
   wxString& operator+=(const char *psz)
-    { m_impl += ImplStr(psz); return *this; }
+  {
+      wxSTRING_INVALIDATE_CACHED_LENGTH();
+
+      m_impl += ImplStr(psz);
+      return *this;
+  }
   wxString& operator+=(const wchar_t *pwz)
-    { m_impl += ImplStr(pwz); return *this; }
+  {
+      wxSTRING_INVALIDATE_CACHED_LENGTH();
+
+      m_impl += ImplStr(pwz);
+      return *this;
+  }
   wxString& operator+=(const wxCStrData& s)
-    { m_impl += s.AsString().m_impl; return *this; }
+  {
+      wxSTRING_INVALIDATE_CACHED_LENGTH();
+
+      m_impl += s.AsString().m_impl;
+      return *this;
+  }
   wxString& operator+=(const wxCharBuffer& s)
     { return operator+=(s.data()); }
   wxString& operator+=(const wxWCharBuffer& s)
@@ -2945,13 +3263,15 @@ public:
       // string += char
   wxString& operator+=(wxUniChar ch)
   {
+      wxSTRING_UPDATE_CACHED_LENGTH(1);
+
 #if wxUSE_UNICODE_UTF8
-    if ( !ch.IsAscii() )
-        m_impl += wxStringOperations::EncodeChar(ch);
-    else
+      if ( !ch.IsAscii() )
+          m_impl += wxStringOperations::EncodeChar(ch);
+      else
 #endif
-        m_impl += (wxStringCharType)ch;
-    return *this;
+          m_impl += (wxStringCharType)ch;
+      return *this;
   }
   wxString& operator+=(wxUniCharRef ch) { return *this += wxUniChar(ch); }
   wxString& operator+=(int ch) { return *this += wxUniChar(ch); }
@@ -2969,14 +3289,14 @@ private:
 
   void DoUngetWriteBuf()
   {
-      wxSTRING_INVALIDATE_INDEX_CACHE();
+      wxSTRING_INVALIDATE_CACHE();
 
       m_impl.DoUngetWriteBuf();
   }
 
   void DoUngetWriteBuf(size_t nLen)
   {
-      wxSTRING_INVALIDATE_INDEX_CACHE();
+      wxSTRING_SET_CACHED_LENGTH(nLen);
 
       m_impl.DoUngetWriteBuf(nLen);
   }